Why RDNA 3 is a Failed Generation

[@vextakes]

Btw, this video is in no hate. I feel like RDNA 3 is a learning generation, hopefully the next will be better.

[@Hexenkind1]

The thing is: FSR 3 isn't even out yet too. Where the fuck is it?

[@muhammadamaar7936]

hmm

[@TheGrizz485]

the full die navi33 rx 7600 ended up %4 faster than navi23. it even uses slightly more power and has %20 more transistors. I am shocked at how broken RDNA3 is.

[@s2131]

@vextakes I am subscribing der8auer and haven't seen the 7800xt simulation. Only saw the one from Igors lab which is flawed in few ways. For example assumes that 7800xt would be the same as W7800 just oced while W6800 was not same as 6800xt. If I would need to bet it would be 7800 non xt priced around current 6800xt, which is same meh as 7600

[@benjib2691]

​​​@@TheGrizz485 I read somewhere that AMD has changed how power consumption is measured between RDNA 2 and RDNA 3. For RDNA 2, the power consumption measured is the GPU core power only, whereas for RDNA 3 it's total board power (GPU + memory + power circuit). This may explain the slightly higher power consumption the 7600 has compared to the 6650 XT (which is GPU core power only). (EDIT After posting the comment I haven't been able to retrieve the info so please take this with a huge grain of salt). Either way the improvement is still disappointing. Working in the hardware industry it seems to me like RDNA 3 has some design flaws and bottlenecks that AMD thought they could solve before launch, but they haven't succeeded. They might have tried to fix the architecture before releasing Navi 33 and Navi 32, which might explain why we haven't seen more cards in the higher-mid-range. I expect RDNA 4 to be much better (especially in the high-end with the chiplet approach), kinda like Zen 2 was to Zen 1/+. It'll have to in order to compete with Blackwell (RTX 5000).

[@hinrei1986]

Nvidia: Hey AMD, let's not compete and push ourselves in this generation. AMD: Good idea.

[@arel508]

I believe that's basically what's going on. That's called price fixing.

[@lilkidsuave]

​@@arel508 and that is illegal

[@daviddesrosiers1946]

I'm really rooting for Intel to get in hard. It'd be a mistake to dismiss them out of hand given their efforts so far. I even bought the Acer version of the A770.

[@georgeindestructible]

No good reason to do better if they know they won't sell as much as they would've wished.

[@Direwoof]

@@lilkidsuave Good luck getting teh fat cats to enforce it.

[@TooBokoo]

Personally, I'm pretty happy with my 7900XTX. I came from a 3080Ti and the uplift is just insane. That said, I didn't wait in line for ten hours to buy one, and I paid less than $1000 and got a free $60 Steam game with it. Not bad considering it beats the 4080 in most cases, has 8GB more VRAM and costs, on average, $200 less. It's no 4090, but it also costs $600+ less than one. The 4090 is an anomaly. We don't usually see a 50% performance bump generation to generation. I'm not sure what people were expecting from RDNA 3. I think the performance uplift is fine. I think AMD's biggest problem and misfire was the price. If the 7900XTX had come in at sub $800 and the 7900XT had been sub $600, it would have been a legendary generation. They would've gotten HUGE props for destroying Nvidia in pricing, would've been applauded for being seen as far less greedy, and people would've been fine with the performance gains at those prices. Sadly, AMD sided with Nvidia and thought those ridiculous Crypto prices for cards would remain and people would be happy to pay them. That said, their biggest problem now is that their old cards still offer a pretty good value, and now they're screwed trying to figure out what to do with their new midrange cards. Pricing is their biggest enemy at this point. They can't make their new midrange cards make sense compared to their old ones, unless they sell these new ones for super-cheap, and beat their previous card prices.

[@Gen_X_Gamer_79]

I agree 7900XTX was a big jump from the 3080Ti & 6900XT. I still have the 6900XT

[@TheFatalintention]

Don't say in a Ngreedia fan boys channel

[@dgillies5420]

7900xtx at $900 and 7900xt at $700 would still be a killer value but AMD doesn't have to undercut by 25% they just need to do 15-20%. And note that you ARE getting much less value since AMD is 1.5 generations behind in ray tracing - their FSR is crap - and cuda / openCL is really feeble on AMD cards. At least in this genereation AMD fixed their lousy mpeg encoders (VP1, h.264).

[@mirceastan94]

The problem is that it only looks good compared to the 4080 at 1200$, if that one was 999$, let alone 899$ or hell, 799$ (the real price from 3080 + inflation), it would have looked much worse.

[@The_Word_Is_The_Way]

@@dgillies5420FSR 2.0 is not crap by any stretch of the imagination. Being outdone doesn’t make it garbage. Why the need for hyperbole is beyond me.

[@Innosos]

I think the story gets even worse when you remind yourself that all that performance and efficiency gain is almost entirely TSMC's work and their improved process node.

[@GeneralSouthParkFan]

It's also made entirely worse by Navi 31 having to sit near its voltage floor (I think like ~800mv?) under heavier load to be able to keep perf% /watt somewhat in check, whereas the 4080/4090 notoriously can hold 1.05V in a variety of titles with ease with their respective 320W/450W tdp. Even running a way more efficient part of the voltage curve, Navi 31 still loses to AD102/AD103 despite Lovelace being predominately a nodeshrink with a massive increase in cache.

[@OneAngrehCat]

Completely false since the 7600 runs off N6 (which is just an optimised N5) and still gains roughly the same performance. It's all monolithic too, and on a smaller die than Navi 23. The 7600 proves that it's neither a question of chiplets nor a "just have a better node bro". Card's got the same perf in the same games as Navi 31, it's strong in the same places and weak in the same places. If anything, the one thing that seems to be a general weakness is that dual issue FP32 that plainly doesn't seem to work well in most places, and those famous "finewine drivers" are just not coming.

[@ofon2000]

actually n6 is an optimized n7...I think that's what you meant to say regardless of the typo.@@OneAngrehCat

[@caffarellaailyze7609]

Feels like how intel is still maturing, RDNA seems still need more time before it really mature.. while NVIDIA is... their greediness is more matured than their product

[@SweatyFeetGirl]

RDNA1 was their first run of their new architecture, RDNA 2 was an improvement of it. RDNA 3 is their first run of MCM gpu architecture, RDNA 4 will be great, just like RTX 50 series using a completely new architecture

[@JimzJimmy]

everyone will do the same if they become a monopoly. market share said that. if you want, buy it. that what they can say to its consumer. if you in nvidia side? wont you do the same? in the end, consumer dont have any choices. nvidia and nvidia. greedy raises when you become monopoly and noone can challenge you and competition isnt working anymore that nvidia becomes sole price and trend setter.

[@yuvanraj2271]

Bro, you are giving so much leeway to AMD. AMD is already matured. The few problems they have are negligible. They need to focus on pricing. Intel is the only one who needs maturing.

[@suchin8524]

@@yuvanraj2271 Intel was formed in 1968, that's over half a century ago. How long does a company need to mature?

[@yuvanraj2271]

@@suchin8524 We are talking about Gpu dude. Are you even on the same page here?

[@Sly_404]

It's very apparent, that AMD doesn't even try to compete this generation. They knew how their cards lined up to the already launched Nvidia cards and took the conscious decision to slide them in around 15%-20% lower than Nvidia. Looking at previous generations, this is sufficient for team red fans to go for AMD but the average Joe still picks Nvidia.

[@hiddedevries8853]

They wanted to compete to the high end but they failed. AMD was always going to slot in around 15 to 20% below Nvidia for the same performance because that is when it sells. They are a company that wants to maximize profits just like every company.

[@hiddedevries8853]

If they had more performance they very likely would have just mad the cards a little more expensive

[@neon_arch]

They could not compete. STFU

[@caribougoo349]

That's the plan. Release at 15% less. Use constant sales to make them a better value without making nvidia react

[@Sly_404]

@@hiddedevries8853That’s literally my point. They know they can sell to their market share at that -15% to -20% margin. But they absolutely do not take market share from Nvidia at that. Given that Nvidia had atrocious cost to performance cards this gen, not seeing this as a golden opportunity is what I challenge.

[@kennyd7667]

I feel like the naming of RDNA 3 is what messed AMD up. They saw the inflated prices of Nvidia and thought "hmm we can price our cards higher and still be cheaper than Nvidia to gain more profit margins" and it backfired on them. Imo, the 7900 xtx should have been called a 7900 xt and the 7900 xt should have been called a 7800 xt and both priced lower by at least $100 then their respective msrp at launch.

[@Ericozzz]

more like $200 above what it should be. It's just that nvidia swung so wildly at the 4090 and 4080 pricing that suddenly they thought 7900xtx would be reasonable at their msrp. They are not even close, and we all knew bullshit was coming when their marketing was around performance to power instead of raw performance

[@kay_a_s]

Agree with this on the naming, though I'm more on the line with the quote-unquote 7900xt being at most in the range of $659 to $699, with the 7900xtx at $899. AMD really missed the mark here, and they would have seen more success if they did not follow Nvidia's pricing.

[@F1neW1ne]

Ya I think we can all agree the names needed to come down a tier. The biggest problem was actually the 7900XT price at launch. It should not have had 20GB. It should have had 16GB and then they could have comfortably priced it at $749 day one. Imagine the power of the 7900XT in a 7800XT at the correct price of $749. That would be the inflation adjusted price of the 6800XT ($650 3 years ago is $750 today). It is 35% - 45% faster than a 6800XT so the name 7800XT would fit. The price would fit and it would have been a smash hit day 1. 7900XTX just needed 1 X removed. If was, and is, $200 cheaper than the 4080. It was priced right against that product. The 6900XT was $999 almost 3 years ago. $999 for the XTX (call it 7900XT) is the same but is 47% faster. Price was right. Name was wrong. Did not help they had a batch of reference coolers fail.

[@chinezut1942]

@@F1neW1neMsrp of 6900xt was overpriced because of crypto mining,it was 8-10% faster than 6800xt for 54% more money.7900xtx(should be a 7800xt-according to previous generations 6800xt was competing with 3080 and now 7900xtx compete with 4080) and be priced at 799$(with adjusted inflation -3080 was 699$),same for 7900xt that compete with 4070-aka 4070ti ,should be called 7800 and launch with 599$).The new card that they want to release it (7800xt) it's a 7700xt that compete with 4060ti (aka.4070) and should be priced at 450$.

[@F1neW1ne]

​@@chinezut1942 And the exact same thing can be said of the 3090 then I guess. Over priced. Are you also saying the RTX4080 should have been $800? OK... Sure. I could get behind that too!

[@RobberClobber]

My 6900xt being half the speed of the 7900xtx at the same prices is not a failed generation for me.

[@marcinkarpiuk7797]

They will probably make it faster in the future with better drivers... Also 6900xt is now half price 7900xtx 😆

[@RobberClobber]

@@marcinkarpiuk7797 yup which is how it should be priced considering. I’m not mad at all I love my card but to say that rdna3 was a flop is sorta lame considering the 40series lol! The 4090 was the only good card to come out of this gen for nvidia.

[@yamsbeans]

@@RobberClobber its the same for amd tho. only good card is 7900xtx. everything else u better off getting previous gen

[@daedalduo]

Vex, I think you nailed it with your theory! AMD is using RDNA3 as a testing grounds for chiplet-based GPUs. This is their way to deal with the fact that Moore’s law is slowing down. If they perfect chiplets for GPUs, they can get better yields and save money. AMD really needs this to work to compete with nvidia in data center accelerators.

[@onomatopoeia162003]

Plus it will age like fine wine, down the road.

[@VladiatorX]

Upgrading from a 2018/2019 GPU today, for double the performance. I see no problem. Generally, I don't think upgrading every generation is a smart thing to do. This week I have upgraded from 5700XT to 7900XTX for 4K. I am pretty sure that it will serve me well for like 3-4 years.

[@inmypaants]

Sweet upgrade my dude! The XTX is a beast, enjoy!

[@TooBokoo]

Yeah, that's a hell of an uplift. Lol

[@piedpiper1172]

As a 5700 XT owner in the market for a top tier card: The 7900xtx is a bit over double what we paid at launch for the 5700xt, four years later. So… yeah, only 2-3 times more power isn’t exactly great for that.

[@inmypaants]

@@piedpiper1172 so paying 100% more for up to 200% more performance and like 300% more VRAM isn’t enough?

[@piedpiper1172]

@@inmypaants Four years apart? No. In the same generation? That’s great. Four years apart? That shows zero progress.

[@AviationAustria]

An rx 7900 xtx is still about 2% faster than the 4080 and costs 300€ less

[@TheGrizz485]

the full die navi33 rx 7600 ended up %4 faster than navi23. it even uses slightly more power and has %20 more transistors. I am shocked at how broken RDNA3 is.

[@mayoofm6049]

is it driver problems?

[@SweatyFeetGirl]

4%? it's 7%

[@belliebeltran4657]

​@@mayoofm6049 Not likely. Since chips has to communicate with each other through the interconnect, some performance will inevitably lost due to latency. It's probably why it also consumes more power, since it's doing more stuff underneath rather than simply calculating right off the bat. Nonetheless, it still isn't good enough explanation as to why a newer chip performs worse than the previous generation, if not barely any better. I reckon the so called "generational improvement" they made is figure out how to create chiplet-design GPUs, and pretty much nothing else on the side.

[@DragonOfTheMortalKombat]

@@belliebeltran4657 There is no interconnect in 7600. It is based on 6nm without chiplets unlike 7900 series.

[@belliebeltran4657]

@@DragonOfTheMortalKombat Oh, sorry. I read wrong.

[@defnotatroll]

Man I LOVE your videos. I follow the hardware space from afar and your explanations/discussions are totally necessary. It's great to get a second opinion after LTT and HWUB

[@carlospuig15]

I think RDNA3 was a test to see if they have the same performance as the 6000 gen. And I think they did, so will see some performance benefit in the future

[@vextakes]

🤞🏼

[@carlospuig15]

@@vextakes Yeah I really hope so too. For competition we all need to this to work. My original comment was a test to see if they have the same performance as a monolithic die. Because from what I understand years ago the performance loss when you go to a chiplet design was to hight.

[@pedroduran8927]

Also remember that RDNA3 are using the same chiplets architecture that they use in ryzen, i think that's a smart design decision.

[@CharlesVanNoland]

I think the biggest issue is how both GPU companies have been keeping the bus width low. The RTX 4060ti's memory bus width is a puny 128 bits. That's insane. The RTX 2060 was 256 bits. It's really like they're intentionally gimping GPUs, just to curb their lifespan so that you have to go out and buy a new one. They could just sell you a good solid honest effort GPU at a good price, but they choose not to. It's like the Phoebus cartel with incandescent lightbulbs 100 years ago all over again.

[@FAHRENHEITDELIO]

7900xt being at most $699, with the 7900xtx at $899. AMD really missed the mark here, and they would have seen more success if they did not follow Nvidia's pricing.

[@RmX.]

Good luck buying top GPU 4000 series for $899 right now

[@FAHRENHEITDELIO]

@@RmX. Here on Amazon Spain the RTX 4070 Ti is 700 euros. so I won't need any luck. other than that. That comment you are replying to was 5 months ago.

[@JHayes90]

The 6700xt being $350 everywhere is the real star of the show for 2023. I just built a 5600x/6700xt/32gb ram system for $1k. Last year that same build would have been closer to $2k. Price to performance is at its peak right now and I believe the prices will continue to fall the later we get into this year.

[@RmX.]

Damn that's a nice rig for $1k. for the same performance with Intel and Nvidia you would pay more than $1.5k

[@JHayes90]

@@RmX. Buy used. Just gotta know what to look for and warning signs of what NOT to buy.

[@akosv96]

Basically we got a prototype to test from AMD. RDNA 4 would actually be the finalized next generation.

[@inmypaants]

Dude the XTX i have us flawless, is over 50% faster than my 3080 consistently and has 140% more VRAM. I do not feel like a beta tester and the tech works perfectly. I have noticed zero differences coming from Nvidia to AMD.

[@TooBokoo]

@inmypaants Same. Dumped my 3080ti for a 7900xtx. The raw rasterisation uplift is pretty massive. 4090 is a rarity for generational improvement. We don't typically see that kind of uplift even on a flagship card.

[@_M....]

RDNA3 is the radeon version of bulldozer.

[@soraaoixxthebluesky]

AMD didn’t anticipate such a performance gain from Nvidia I suppose. We all know how powerful Nvidia is architectural-wise. The only thing that set them back on Ampere was Samsung process node. AMD should be well aware of this knowing that they’ll jump back on TSMC after Covid settled. They should priced their cards accordingly at launch.

[@dra6o0n]

AMD had incidents where they had to deal with double agents working from within to ruin them in the past, they still might have some. For instance the AMD Marketing Team is led by a ex-Nvidia employee.

[@DragonOfTheMortalKombat]

It's not just that , RDNA3 is being bottlenecked. The architecture is powerful with double the no. of transistors, much faster memory , AI Acceleration etc. Don't know what but something is holding it back. Is it the chiplet design ? or drivers ? I don't know.

[@dra6o0n]

@@DragonOfTheMortalKombat The engineers said that in shrinking of the components, the one thing they can't shrink is the cache that is on said dies.... That's impossible to shrink, but these cache needs to be near the cores to operate efficiently and with low latency... They have MCM modules and GCM cores. They managed to shrink everything and separate it from the MCM where the cache is, but they themselves can't be shrunk without severe problems and generates a lot of heat. This could be why it still takes a lot of power for a RDNA 3 die, because the cache itself eats that to keep up with the processing speeds. Of course, Nvidia never ever went chiplets or 3d stacking at all, all of their GPU dies are monolithic. So going monolithic but shrinking the process nodes will net bigger gains as the power usage also shrinks with it as it's retaining it's large die area for the cache like usual... But Chiplets kinda 'slices up' the caches out of the other components and makes it more complicated in order to have a smaller form factor.

[@mastroitek]

though we all knew that nvidia would have switched to TSMC which can already brin a 20-30% perf increase depending on the node (samsung node was not that good to begin with), on top of that one must consider all the architectural progress made in 2y. In my opinion it was kinda obvious that this new gen would have been about 50% faster at the same power draw. I personally think AMD was always aware of that, but simply did not manage to hit the efficiency target (expecially in the mid range and low-end)

[@Dominus_Potatus]

I agree with you that RDNA 3 is a learning generation, it is similar to Zen 1. You know Zen 1 was good because it was cheap and really useful for professionals that needs a lot of cores. Then they revise it, Zen+ was significantly better than Zen. Zen 2 introduce bigger cache which helps a lot and it was a leap. Zen 3 introduce different topology, chaning from 3+3 and 4+4 to 6+0 and 8+8 makes it techincally chiplet since the memory controller and I/O are on different chip. See how they change from 3+3 to 6+0 and 4+4 to 8+0? It made Ryzen 5 5600X and Ryzen 7 5800X so great that 3600x and 3800x were so bad if compared. It also makes the entire Zen3 SKU better, especially for 12 cores and 16 cores SKUs. They had 4+4+4 (3 chiplets) and 4+4+4+4 (4 chiplets) to only 2 chiplets: 6+6 and 8+8. This might be the reason why Ryzen 5 5600 is sold really cheap, in market rn Ryzen 5 5600 is priced USD 105 (tray) or USD 120 (box). 5600 and 5800x are sharing their production line while. Now back to GPU, it becomes... tricky. See how AMD changes from 4 chiplets to 2 chiplets? Basically they reached a conclusion that less chiplets are better and 2 chiplets are the optimum number for CPU. RX 7900 XTX has 96 CU, from the image, I assume they have 6 Chiplets (96 MB L3 total, 16 MB L3 each). RX 7600 XTX has 20 CU coming from, I assume, 2 chiplets since it only has 32 MB L3. So... each chiplet can have 16 CU (8 dual CU), but in RX 7600 XT it has 10 CU (5 dual CU) for each chiplet. Either they have bad yield or they prioritize Z1 extreme that has 12 CU. In the end, Zen was successfull because it was cheap. The once high end product (4c/8t) became entry level Ryzen 3, the once HEDT (8c/16t) became mainstream. It was simply distrubtive. Zen+ was welcomed since its APU was helpful for gaming because all the mining craze. Zen 2 was great for it offers leaping performance and making 6c/12t a really cheap product. Zen 3 was great because AMD delivers more than they promised (remember AM4 planned up to Zen2) and it delivers a leap on performance. RDNA 3... what do they offer? It is neither cheap nor has significantly more performance. I mean it is significantly more efficient but let's get real... most people compare mostly by performance. RDNA 3 is a generation where AMD pays for their sin (read: increasing power consumption for performance). A generation where AMD want to find the sweet spot about chiplets configuration. Should you buy? The question should be "Do you need new graphic card?". If you don't need it, skip the generation. It will mature like milk, similar to how Ryzen 1000 is rarely found on 2nd hand market. It was bad product, it was good because it has 2x cores. My prediction, if AMD learned from Ryzen, is RDNA 4 will be great and they might make each chiplet has 24 CU instead o4 currently 16 CU. So top product will have 96 CU from 4 chiplets of 24 CU

[@Dominus_Potatus]

In short, if you don't need it, don't buy it. While software update might help a bit like how Zen was running better on updated Windows at launch, it still doesn't help the core problem: hardware. You can only do much with software. Or... maybe AMD will go crazy and deliver 6 chiplets of 24 CU with 144 CU total while the mainstream 2 chiplets get 32 CU. Man can dream

[@dalesplitstone6276]

The MCM approach should not have been presented as a means of producing GPUs cheaper, as you pointed out, the cost of a graphics card is not all in the GPU die, but rather a way of producing more powerful GPUs. This should have translated into a major shift in the GPU stack. In particular, the top of the Navi 3 stack should have been a 128 CU compute die with a 512-bit GDDR6 interface to memory. This could have been called N30 and the graphics card the RX 7900XT, even if it carried an MSRP of $1499. With 32MB of VRAM it could have attracted a lot of attention from the content creation market. If the RX 7900XT had an MSRP of $1499, the cut down version, with 28GB of VRAM on a 488-bit bus could have had an MSRP of $1249 or even $1199. Navi 31 is fine, it is just mislabeled, it should have been called the RX 7800XT and the cut down version with the 320-bit bus the RX 7800. The full die, called the RX 7800XT instead of the RX 7900XTX, could have had an MSRP of $899, and the RX 7800 an MSRP of $699. The lower end SKUs should have all been monolithic 6nm dies, with N32 being 64 CUs fed by a 256-bit memory bus, N33 being 48 CUs fed by a 192-bit memory bus, N34(insted of the actual N33) 32 CUs fed by a 128-bit memory bus, N35 24 CUs fed by a 96-bit memory bus, and N36 a renaming of N24, or possibly AMD could have simply continued with N24. Even N35 could have potentially been introduced with 12GB of VRAM, possibly like nVidia’s upcoming 16 GB 4060 Ti, having memory on both sides of the card, a more optimized 9GB VRAM variant could have been introduced later on, then later a cheaper 6GB VRAM variant. N34 (currently called N33) could have similarly been introduced in a 16GB VRAM variant, with cheaper12GB, and later still cheaper 8GB variants following. The 5700 had 8GB of VRAM on a 256-bit bus, implying that wit this capacity memory chips, 4MB of memory could be attached to a 128-bit bus. If this were in addition to, instead of in place of, the 8GB of memory of the 6600/6600XT, this would result in 12GB of VRAM on a 128-bit bus. I don’t know it this could have been done with the 6000 series, but if it could have, it would have been a good way to clear out 6000 series stock in the face of competition from used graphics cards. I any case the suggested N34 die could have been designed to support up to 16GB of VRAM in two banks on a 128-bit bus, also providing the potential of a 12GB variant. With N31 being the GPU of the 7800(XT), the proposed monolithic N32 die would have been the GPU of RX 7700(XT), N33 the 7600(XT), N34 the 7600(XT) and N35 the 7500XT. Thus, instead of the 8GB RX7600 being launched at $269, AMD might have launched the most identical 16GB RX 7500 at between $299 and $329, with the launch of the 8GB variant being delayed until most of the price slashing was done, perhaps just in time for the Christmas shopping season. The proposed 7600(XT) with 40 CUs and 20 GB of VRAM might have been launched at about a 25% price premium over the 7500(XT), perhaps $399 for a 20GB 7600 card with performance in the same raster performance ballpark as the 4060 Ti. The 7600XT with performance about 20% higher than the proposed 7600, and 24GB of VRAM, with an MSRP of perhaps $479, could have been an even better challenger to the 16GB 4060Ti, and like the above proposed 7900XT, attract a lot of attention from the developers of productivity apps. The 7700(XT) with up to 32GB of VRAM, could have brought 32GB of VRAM could have brought 32GB of VRAM to perhaps the $649 MSRP point.

[@richardmcgowan1651]

Profits over quality. Also people are holding onto parts much much longer. You don't need a 4090 ti to play vampire survivors. The market is shifting. AAA games out indie games in.

[@Kapono5150]

Absolutely loving my 6950 XT. I’ll be skipping the 7900 series GPU’s

[@Duplex39]

My belief is NEVER get gen 1 of a product (specially in tech), gen 2 of chiplets will determine if that is the future or not.

[@chuckchuk5157]

Question is when??

[@Sly_404]

@@chuckchuk5157 RDNA4 aka late 2024.

[@inmypaants]

Although I agree with this sentiment for large purchases like cars or similar cost items I hardly consider a GPU for $1000 to be too much money to avoid a first generation.

[@tehpanda64]

rdna 3 came up with some great ways to add a little bit of vram, and charge more money, for the same number of stream processors as the 6000 series rdna 2.....

[@vuri3798]

Yeah we really didn't see any of those savings 💀💀

[@mRibbons]

I don't think AMD even knows the value of their cards. They never launch first and are more prone to price drops. It's like Nvidia is their yardstick every gen.

[@Johnwick-ed7vo]

Great video as always, you have a great way of making people think

[@vextakes]

Ty

[@daysofgrace2934]

RDNA3 is not a failed generation, it's a 1.8X improvement in a ray tracing over RDNA2; 6900XT was equivalent in performance terms to 3060TI and 7900XT is equivalent to a 3080 in Cybperpunk ray-tracing...

[@mutawi3i]

Proper good job m8. Splendid work.

[@anarchicnerd666]

So. A question for those more familiar with the growth of AMD GPUs in the long term, coming from someone considering the upgrade to a 7900 XT at some point. What's the likelihood of the "fine wine" effect kicking in maybe a year or two down the road with the 7000 series? How much of the lackluster performance is down to the chiplet based architecture, and how much of it is down to drivers? Is it even possible that advances in the drivers in the long term and more mature software could improve how the chiplets talk to each other, therefore boosting performance? It's be really interesting to hear from 7900 XT and XTX owners how the drivers have evolved since launch :)

[@lukasr1166]

Personally I don't believe the 7000 series will have the same "fine wine" effect as the 6000 series. Because the tech is fundamentally flawed. Still they're not bad GPUs, they are bad in comparison to AMDs marketing.

[@Sly_404]

I would also be very critical. So far very little indicates that these cards will become more efficient. I think RDNA3 has some innate design flaws that give them less headroom for improvements than RDNA2 did. The XT is still an okay card if you can get them at a discount. Generally this generation is best to be skipped and if you need a card the 6000 AMD ones have better value than anything Nvidia or AMD offer this gen.

[@harbinger218]

The 7900 XT is a great card I have had zero problems. I can run everything I play at max settings at 1440p. Absolutely zero issues. I think that it is true that the architecture is posing new challenges for them as far as drivers and stuff but it isn't anything they can't handle and haven't been able to handle. It is a learning curve for them. That being said I haven't had any problems. I get well over 160 FPS for most AAA games. Also I have a rise in 5 5600X CPU.

[@anarchicnerd666]

@@harbinger218 You mention you haven't had any problems with the drivers, that's good, but have you seen much in the way of performance or stability improvements with driver updates? Or has the performance of the card been consistent for better and worse? Thanks :) It's good to know the cards good at max settings 1440p, the big reason I'd be getting it is for an ultrawide monitor

[@Magic.God.Othinus]

@@harbinger218 In part it has to with what GPU are you upgrading from. If it's a few generations old, then it's worth the purchase imo.

[@Ivan-pr7ku]

At some point RDNA3 was defiantly meant to reach 3GHz turbo boost at least, but even the chiplet approach wasn't sufficient for that. I think the high clock-rate initial target was the reason there were so few changes in the micro-architecture. This generation bears some similarities with Vega -- minor architectural upgrades, where most of the effort went into the memory subsystem.

[@Schroinx]

RDNA3 is underperforming compared to what AMD expected. That said, AMD could have re-named their SKUs, so the XTX could be 7900 XT, 79000 XT a 7800XT, the 7600 a 7500 XT and so on, and priced them competitively, as their CPUs has been. But AMD still choose to follow Nvidia's price increase...

[@caribougoo349]

Realistically the cost savings on going chiplet isn't huge outside bean counting. Maybe $10 on the 7900 xtx. That's not including the added difficulty in assembly and design but also doesn't include the possibility to leave mcds off if the gcd is being cut down. The bigger benefit is being able to get more production out of the new expensive node by cutting the fat and putting it on a less sought after node. Eventually it'll be a necessity when they have nowhere to go but pushing bigger dies. Doing a 600mm gcd + 300mm in mcds is way more practical than doing a 850mm monolithic

[@kiop901]

I would say the same with Nvidia with 4000 series except for the 4090. Except for the 4090 the rest of Nvidia 4000 series cards are bad in performance and price

[@03chrisv]

I always thought that was a weird statement. The 4070 and 4070 Ti actually have better price to performance than the 4090. You literally get more frames per dollar with those cards, yet people think the 4090 is the only good 40 series card. Interesting.

[@zjanez2868]

​@@03chrisv probably because of the vram gimping and the rx 6000 series being even better cards are more expensive than ever, yet apparently 12gb is already causing issues at higher resolutions the 90 is also in a completely diffrent price and buyer segment there are some people that just want/need more power, money being no object. those buy the flagship and for them the absolute performance increase matters the most the mid range should offer good value, but there rx 6000 series is just better

[@Pouria_1664]

amd loves giving people no reason to buy their products.. what's up with them ..? no one knows.

[@jochenkraus7016]

Afaik Nvidia uses more advanced manufacturing at TSMC so it would be sad if RTX4000 didn't improve performance and efficiency. On the other hand I've read that TSMC increased prices and not all units (computing, Cache, IO) profit from the smaller structures. Maybe that's why both are disappointing: Nvidia at price to performance. AMD at performance and efficiency compared to RX6000.

[@DragonOfTheMortalKombat]

@@jochenkraus7016 If TSMC price increase was so huge then why do Ryzen 7000 not disappoint ? STOP blaming TSMC for GPU makers' fault. Chiplets from AMD and small die sizes from Nvidia should've solved the problems of 5/4nm pricing easily. It probably did............ for them making them more profit while we get nothing good.

[@eternaldarkness3139]

Where I am in Canada, the MSI gaming trio 7900XTX is almost 25% off From $1550 (1125usd) down to $1200 (875usd) making it priced at RTX4070Ti levels. PC parts costs here compared to US, are typically MSRP+50%

[@wallacengineering8096]

Its kinda the same thing as Nvidia. RTX 4090 huge boost over the 3090. But then as you move down the line, the generational uplift becomes smaller and smaller. 4070 Non-Ti isn't hugely faster than 3070 Non-Ti. Same story for AMD now. RX 7900-XTX hugely faster than RX 6900-XT, RX 7900-XT a fair bit faster than RX 6800-XT. But then the 7800-XT will be compared to the 6700-XT, and so on. Because in reality, the RX 7900-XTX is just the XT, and the RX 7900-XT is actually the RX 7800-XT. You gotta ignore stupid naming schemes. I love my Red Devil 7900-XTX. It is a badass card, but Im well-aware that the second "X" is just imaginary. It didn't need a new name.

[@Generationalwealth94]

Wtf is it with AMD and their naming schemes anyways lmao, I just found out there's like a Radeon 9850 XT from like 2003 whose name will overlap with the GPU that comes 2 gens from now, it's completely ridiculous

[@ryanspencer6778]

Nvidia and AMD are in a competition to see who can give customers the worst value for their money. Nvidia: you get a gpu die one tier lower for a price one tier higher. AMD: you get basically zero generational improvement in performance and efficiency. They're basically rolling out the red carpet for Intel. If Intel has a normal GPU release next year, then it'll look like a massive jump. That's how bad Nvidia and AMD are doing.

[@F1neW1ne]

Yes we all agree 7900XT should have been called 7800XT and priced at $749.99. Would have been a smash hit all day long. AMD dropped it to $800 and included a game so it's there now kind of.

[@vextakes]

I agree, however should’ve been around $650-700. Because we see nowadays the 6950 xt is $620 and the 7900 xt is barely faster than it

[@F1neW1ne]

@@vextakes That is last years model and on clearance pricing so I guess if you were spending 600 - 700 ... sure get one.

[@DannyzReviews]

It's RDOA3 now😂

[@ProfPistolPete]

I'm glad I bit the bullet and got a 6950xt last month rather than continue to wait for the rest of the 7000 series. Would recommend if you have a beefy psu and some experience undervolting/tuning Radeon cards.

[@briangrant9942]

hmm don't need a beefy psu for the 6950xt I just have a 850 gold and it's enough

[@srxt6758]

​@@briangrant9942there are some people in other countries than US who dont have money for 850w. You considering 850 not beefy where there are guys still running 400w and 450w in their systems lmao. And yes 850w is beefy for budget gamers.

[@mastroitek]

​@@briangrant9942 850W is quite a beefy PSU in my opinion. But in the end I think even a good 700W could be enough, and if one intends to do some UV then 650W is more than enough.

[@briangrant9942]

@@srxt6758 sorry to say that's not my problem it's just my opinion.

[@vicroads3465]

I’m running a 7900xtx on a 550w lol

[@Ericozzz]

They increased the price and lowered the performance jump. I think both nvidia and amd thought they could push the price because they also thought mining would last a little longer, enough for them to cash in with the scalpers. No one will buy midrange anymore since the used market is flooded with mid range from mining rigs (5700xts, 3070s, 6700-6800xts). I grabbed myself a 6700xt last year after mining crashed and cards went back to what was already an inflated msrp, If I knew the price would drop even lower so fast I would wait to get maybe a 6800. Coming out of a repaired rx590 that was underdelivering performance it was a huge jump, so this gen is a skip for sure and anyone that is at 1080p like me should just buy used midrange that is dirt cheap now.

[@ThunderKat]

What increase cost is probably the frequency and amount of cables running on the GPU board, the complexity and the need to avoid interference increase the cost of thing. Maybe is best to ask the 3th party companies about the cost of the board to see what change but so hard to get into those things, they always keep things a secret. There is also the printer for the silicon, who know what kind of trouble they had to go throw to improve into those small sizes including quality of the silicon alone.

[@MrSherhi]

What im missing on current GPU market is nice analysis of not just "15 games average raster performance" but also with consoles mixed in and then do a price/performance comparison

[@Sly_404]

I can point to endless number of YTs that argued, give AMD 6 months and we will see some super driver that lowers consumption and increases performance. Now we are 7 months in and nothing happened. Just because RDNA2 had enough headroom left to allow for post launch improvements, doesn't mean anything for RDNA3 having a significantly different architecture.

[@lukasr1166]

They were basically huffing copium. you can't get such big performance upgrades from driver updates as RDNA 3 needs to meet the performance AMD advertised. You just cant. AMD missed the 1.7x performace of the 6950 xt with a big margin, they are not making up for it. They havn't even released FSR 3 yet. RDNA 3 is just bad tech I'm sorry to say.

[@SweatyFeetGirl]

@@lukasr1166 it always said UP TO, and there are games that are up to 70% faster... doesnt mean it will be 100% of the time be 70% faster... you can have 1 game out of 5000000000 gazillions perform 70% faster, giving u the right to advertise it as such

[@lukasr1166]

@@SweatyFeetGirl That is what I like to call deceptive marketing. Legal or not, it does not represent what kind of performance you'll be getting generally. Nowhere did they say what kind of performance you can expect in most games. It is deceptive and makes the product look bad.

[@Sly_404]

@@SweatyFeetGirlAnd how did that turn out? Generating marketing material that isn't reflective of the product but only selected best case scenarios? This right there was one of the reasons why the 7900XTX launch reviews been so critical.

[@SweatyFeetGirl]

@@lukasr1166 that doesnt matter at all, what matters is that some games do perform up to 70% better.... only a dumbo would believe that every game will be that much faster. it's like visiting a shop that says our products are up to 80% off ! and the only product being 80% off is a 2$ product to begin with..

[@MDFGamingVideo]

Neither RDNA 3 nor the nVidia 20-series were failures. ANY time a company brings a big, new innovation to market, they are breaking new ground. Sometimes they may be ahead of their time, software needs to catch up, or lessons are learned from the initial release to make subsequent iterations far superior. But first generation success is very rare. Ryzen CPUs are a good example. They started out good, but then got GREAT. And Intel had to stop selling us overpriced crap, stand up, and represent. I'm hoping the same thing happens with GPUs, because nVidia's prices and products...remind me a LOT of Intel before Ryzen. 🤔 But...to tack the label of "failure" on something new is for the young. Patience, grasshopper. 😉

[@onomatopoeia162003]

Not to mention Intel was very complacent too. Took Ryzen to move Intel.

[@Technicellie]

Chiplets seem to only be worth it on bigger dies, since the waste if you have a defective die is smaller. And since those smaller dies have less stuff, they also have less stuff that can be defective. So the amount of die material to throw away because of them being defective rises exponentially to the size of the single dies. Of course that doesn't count in the possibility of dies to be partially deactivated if they have multiple units of the same thing in parallel (i.e CUs, CPU cores, parts of the cache, RAM/VRAM busses, etc.) But some dies won't be able to be salvaged if the damage is done to a unit that only exists once and is necesarry for it to function (like the scheduler, the Display output engines (I think AMD calls them Video Core Next), encoder and decoder, etc.) Please don't take that info as pure facts, I am not an engineer nor do I have expertise in chip manufacturing in any way. I just read and watched a bunch of stuff because I find it interesting and wanted to add my take on why AMD goes chiplets in bigger GPUs now. So please see it as just that: A take.

[@shreshta_yt]

im sorry vex, i completely disagree. the bad day 1 perf was because of driver problems, those have now been fixed and now RDNA 3 is doing great. and as for your RX7600 rant, AMD does not want this to be a attractive GPU, NAVI 23 (RX 6600, RX 6600xt, RX6650XT) has a oversupply that AMD wants to get rid of, so they are making them seem like a better value to hopefully sell them. though the A750 will give them a hard time

[@vextakes]

All my performance numbers I grabbed were from HUB’s 4070 ti video, which was a couple months after 7900 xt and xtx launched. So they should be pretty accurate from why ur saying

[@shreshta_yt]

@@vextakes , are you shure they did not just reuse launch data? If it was new data then that's embarrassing for me and AMD, and also bad for all those people to who I recommend the 7900xt to. There is also the case that AMD wants to just clear out RDNA 2 stock, so maybe they are intentionally making the RDNA 3 cards look like a bad value

[@Accuaro]

6:23 To add onto this, chiplets mean smaller silicon, so you get more yields per wafer. It gets really expensive when they're monolithic and huge, so you get less profits per wafer. It's easy to pass on the cost to consumers though.

[@dotxyn]

Not only can yield allow for cost savings, but AMD also uses different process nodes for some of the chiplets, using N5 and the less expensive N6 together. The 7600 only uses N6.

[@SeventhCircle77]

The 7900xt should’ve just been the 7800xt at $700 tbh.

[@wrusst]

If i recall they thought it would run at 3ghz which didnt happen. AV1 is nice though and having a super cheap card but agree seems a bigger issue. That navi 3.5 seems to indicate theres a hardware refresh coming. Also 40 series outside the 90 especially over vram feels a slow generation also amd only thing that intresting is software innovation

[@chapstickbomber]

RDNA3 runs 3GHz no problem, it just uses more power than they'd prefer.

[@wrusst]

@@chapstickbomber you say no problem but even with power limit increase it rarely if ever hits 3 on a 7900xt. Also what's the point of designing something to hit 3ghz and not do it ? It's clear the card if designed correctly should hit 3ghz goal within there power goal . It's not not like you cant give the option to the consumer to have a an option to go 3ghz if they chose to ignore efficiency again it's not there so pointless feature . If all the cards had this option it would be a selling point and would give a better gen on hen increase. People have noted the results Vs the transistor increase seems to be off a fair bit 🤷‍♂️

[@chapstickbomber]

@@wrusst not my fault AMD didn't want to run 550W 🤷

[@wrusst]

@@chapstickbomber you said 3ghz no problem now your saying AMD doesn't supply 550w which is a problem? I suggest you choose one

[@chapstickbomber]

@@wrusst I use 550W or more on mine, the hardware can handle it fine. N32 at 350W will also scream over 3GHz, but they aren't going to sell it that way because they want to compete on power (lol weak)

[@АндрейШевченко-к6й]

I don't think that RDNA3 is a "Failed" Generation. This generation at minimum better than RDNA1. Why so: 1) Yep, improvements are small if we compare "RDNA2 1CU vs RDNA3 1CU" - a bit better performance in high resolution and almost +0% in 1080p. But switching from 6nm node to 5nm gives a minor upgrade in power efficiency too. 2) New features like Wave Matrix Multiply Accumulate (WMMA), FSR3. 3) Competition with nGreedia: Part 1 - AMD increase amount of CU in their GPU models, as example - RX6600=28CU and RX7600=32CU, RX6900XT=80CU and RX7900XT=84CU. Part 2 - Price policy, AMD trying to follow current market trends instead of nVidia's "are you buying or not?". As example RX6600 msrp = $329, RX7600 msrp = jumps from $299 to $269. On top - RX7900XT released with msrp = $899 but today we can easily find such cards with $799 price brand new! And RX6900XT have been launched with $999 price tag.

[@Alovon]

Honestly I think the problem isn't necessarily with Chiplets. but moreso with the core architecture. RDNA3 does inflate the TFLOP rating and "shader-count", like how NVIDIA did with Ampere. But unlike NVIDA, AMD is doing dual-issue with their shaders. Meaning software has to dedicatedly utilize the extra shaders, and even then they only moreso act on throughput rather than major increases raw shader grunt, it can work faster, not necessarily harder. This helps with RT which with how AMD has done their RT Pipeline, does benefit from working faster, but in regards to Raster it doesn't help much. We can see this as much with the RX7600 vs the 6650XT, both have the exact same CU counts, but one is rated for 22 TFLOPs and the other only 11 TFLOPs. yet they are within margin of each other outside of RT which the throughput of the dual-issue design does help utilize more than RDNA2. But in raster workloads, RDNA3 is effectively doing half of what it could theoretically do in an optimal workload.

[@not_a_samsquamsh6777]

I can't remember where I heard it, it was either on WAN or moore's law is dead. There was a story about a persistent hardware bug in RDNA 3 that couldn't be fixed. Relating to a sort of time-out. Hence, why performance was shown higher at the launch event than it really turned out to be for consumers. The driver level "fix" for an unfixable issue dropped performance.

[@willia451]

Chiplets. With Infinity and beyond.

[@karmaplus87]

I currently have an RX 7900XTX. I'm really happy with it. It takes games at 1440, and 4K chews them up and spits out buttery smooth 100+ fps at any settling on any title. I'd rate.

[@Mr.-Toad]

you were right!

[@zdspider6778]

3:57 Because the "4070" is a 4060 in disguise, so of course it's gonna have a low power draw, typical for a 60-class GPU. All 4000 series cards are 1-2 tiers shifted, except for the 4090. And yes, the 7900 XT should have been called 7800 (non-XT).

[@vextakes]

I agree. Tho it still shows the insane power efficiency gains Ada has vs RDNA 3

[@LukeBroadhurstfilm]

Something else to point out us chipsets bring lesser performance gains over monolithic. It’s new territory, it’s something that will go through growing pains. I think that’s why AMD didn’t carry chiplets over to the rx7600 because the performance uplift would had been far to small.

[@vextakes]

Yeah, it’s kinda like they spent a bunch of time engineering chiplets and they don’t actually bring anything in the way of performance, other than maybe bringing costs down. Next they gotta be working on their performance

[@inmypaants]

If there’s no performance (or little) loss and margins are improved, it’s still a massive success for AMD. These launches aren’t always about market share in a low margin vertical, it’s often about investor perception. AMD were the first to market with chiplets and that was important given their Ryzen position. Share price for AMD remains strong and they will likely build on that with investments into AI provided that bubble doesn’t burst.

[@TommyBlazeHD]

I have a RTX 2070. I'm looking hard at the 7900XT or 6950XT. Any thoughts? Should I expect a large improvement since I'm coming from a 20 series card?

[@JaseHDX]

you’d gain around 80% so yeah

[@Shadowbot074]

I have a 3070ti and am planning to upgrade to a 7900xtx. The uplift is enough for me and the vram should last years. And at 1440p I hope it’s the case

[@mikkolindstrom1797]

They can still salvage this with dropping prices.

[@inmypaants]

Watching this on my 7900XTX and couldn’t be happier with the purchase and card overall. I think the generation could have been more exciting and it would have been awesome if AMD did price these cards in a way to actually take market share, but that hasn’t happened. Instead we got a gen on gen performance increase for similar money. I’m honestly okay with it though because the card does the things I want and I was happy with the price I paid overall.

[@chuckchuk5157]

One of the better tech youtubers in the scene. quick and consice explanations keep it up. Sidenote anyone knows rx7700 rumors specially release dates?

[@vextakes]

Ty. Lol idek what 7700 xt is looking like

[@AndersHass]

The main GPU in RDNA 3 is one piece, it is the cache that are added as the smaller pieces. It could be possible in the future they add multiple GPU chips together kinda like what Apple has done with their own M Ultra GPU (and what AMD has done with Ryzen CPU cores). Cache does seem to not shrink on a smaller process note unlike the cores so another benefit is saving money on just using a bigger process note for the cache then use smaller for the cores. There are cache performance issues with RX 7900 XT and XTX compared to RX 7600 which likely is because of chiplets vs monolithic.

[@AGuyFromHazard]

I'm buying a 7900 xt over the 4070 ti mostly because of the 8gb more vram

[@BenedictPenguin]

honestly it could've been other way around if and only if they called the navi 31 20GB "7800 xt" instead of the 'xtx scheme' then the navi 32 can be release as a successor of the 70 class with 16 gigs and performance similar to 6950 xt (based on rumored specs) imagine how much better that would've be? a ~500 usd GPU with the power level of previous flagship card, basically back to rtx 3070 destroying rtx 2080 ti in price per performance

[@gureguru4694]

its like AMD calculated the exact amount of needed compute of the 7800XT to be exactly like 5% faster than 6800XT to the dot. it doesn't make any sense.

[@exoticspeedefy7916]

In regards to compute units, RDNA3 actually has twice the TFLOPS per CU because of the dual issue function. Remember that doesn't equate to gaming performance

[@atnfn]

I've seen in some tests that in CAD programs RDNA3 cards are often faster than nvidia quadro cards. I guess that's the reason.

[@Gustinelli]

RDNA 3 and Rtx series 40s are memes and fillers generations.

[@pf100andahalf]

I realize that this is an amd video, but one thing that should never be forgotten is that both amd and nvidia misnamed their cards. The 7900 xt is really a 7800 xt. The rtx 4070 ti is really a 4070, etc.

[@DragonOfTheMortalKombat]

True. They could've called it a 7800XT and priced it at 700$. Even now 7900XTs go for 750$ only. 900$ was never sustainable.

[@chinezut1942]

No,7900xtx is 7800xt and 7900xt is 7800,look at the previous generations 6800xt competed with 3080 and 6800 with 3070&3070ti(was priced with 579$ ,between 499$/599$).They should be priced at 799$ and 659$

[@kevinroosa1315]

I feel like they pushed the card a little too soon. The drivers are slowly improving things on Ryzen 7000, but that doesn’t really help the case for performance in the past and even right now.

[@SpecCrazy]

I think one thing that we look across it is a count of Texture Mapping Units (TMUs). These days games have more and more graphic details that why it need texture units more and more. Not just more Compute Units to process faster. As i though the power of compute processing is more than enough but bottleneck on TMUs that a important part to use to mapping graphics textures. Look at to 4090 that have 512 TMUs and 7900 XTX have 384. Combine with clock speed that make it process faster too.

[@ziokalco]

Although architecturally RDNA3 differs a lot from RDNA2, what the end user experiences is more in line with a refresh. Even though the argument of much better launch MSRPs exists, it is only true because most of RDNA 2 launch MSRPs were horrendous

[@iordanchis2437]

Today I've learned something new. Chiplets vs Monolithic die. Thanks man! Now I'm hopeful for future AMD GPUs.

[@kaseyboles30]

IF it's only 15% or less ahead of the 6800xt that sells for $500, then doesn't matter the model number, it's a $500 class card.

[@Hoigwai]

You can see this was a trend with GPU performance. While NVIDIA did do some work to improve power efficiency on the 4080 and lower the singular method they used on the 4090 was brute force. 3090 328 cores to 4090 512, 1.5X more cores to be able to dunk on the competition, to get roughly 63% more performance. If AMD had put 120 cores in the 7900XTX it might have come much closer to the 4090. The problem we're also dealing with is both companies are courting that sweet sweet AI money right now and we also dealt with crypto + pandemic making them think that they could make anything and we would buy it. Also, the price that we are seeing now for the AMD 6950XT should be the price for the flagship for all the cards at launch. There is so much gouging by both of them it is disgusting.

[@defeyterjelle]

Some sources indicate that RDN3 is not using its full potential. A few "weeks" before launch they ran into a problem, when RDNA3 was used for a few hours they got some "artifacts". They hoped to get it fixed with drivers at launch, but they din't . And it probably never will. Other than that, they're still good cards

[@dgillies5420]

AMD fixed the encoders in RDNA3. VP1 and h.264 encoders are finally good !! Their ray tracing is more competitive than before. A 7900xtx is about as good as a 3090 so it's 1.5 generations behind. It's not 2 generations behind (6950xt = 3070 ti) but we thought they would catch up faster!

[@TooBokoo]

I mean, in pure rasterisation, the 7900xtx beats the 3090 quite easily. I just dumped my 3080ti, which is about on par with the 3090, minus vram, and my 7900xtx destroys the 3080ti in raw framerate.

[@dgillies5420]

@@TooBokoo Respect! I bought a 7900xtx for my son's graduation (he's a CS major). First chiplet GPU on earth. Blazing fast, relatively cheap, smallest card with that performance, and looks BADASS. I hope AMD will rock the GPU world the way they rocked the CPU world with chiplets.

[@giucafelician]

Hello VEX, I like your reviews... I like how you explain that this time I have an observation to make! It is something that deeply disturbs me. The fact that you omit an essential thing when comparing AMD cards! AMD differentiates the cards in the 6800 series with an XT... to have an equivalent to NVIDEA's Ti. So... you have a simple 6800 or a 6800 XT... then they went up in the range with 6900 XT and 6950 XT! Correct? With the new 7900 series... they changed the name... On the 7900, normal version they put an XT and on the XT version they added an X for un XTX. So when you compare a 7900XT with a 6800 XT it's not the same thing! It should be compared with 6800! When you compare 6900 XT... it should be compared with 7900 XTX to be correct! Yes, it is confusing... but if we are passionate, we must be careful in comparisons! And you are an influencer who pays attention and is objective... except here you forgot about this difference in name! And let's not forget that the 7900 series actually came to compete with the 4080 series... remember "Nvidia Unlaunches RTX 4080 12GB After Backlash"??? So yes, even the 7900 XT would have been able to fight with what NVIDEA wanted to sell... in the 4080 12GB! If we look at the 4070 Ti have 12GB... it is the 4080 12GB version that NVIDEA withdrew before the launch when it saw how weak it was in performance compared to the 7900 !!I wish you success, you are a formidable influencer! Success! kzbin.info/www/bejne/b5-1eJRtrayhZrc www.techpowerup.com/299859/nvidia-cancels-geforce-rtx-4080-12gb-to-relaunch-it-with-a-different-name

[@kusumayogi7956]

RDNA 3 is mediocore upgrade Not like polaris to rdna 1 and rdna 1 to rdna 2. They were huge upgrade. I believe amd actually can make 400-500 watt gpu that compete with rtx 4090 but they dont because some reasons.

[@Dj-Mccullough]

They "competed" just fine in the 6000 vs rtx 3000 series. but they Did stumble HARD with the 7000.

[@jezzaboi2168]

It would be fitting to see amd follow the set-up of the Apple m1 ultra gpus. With multiple dies next to each other that acts as one. This seems in line with the chaplet design they have wit ryzen and rdna 3, but wouldn't come to fruition for a few generations

[@porkyp1845]

I like that subtle animal crossing background music

[@i_watch_everything]

More like RDNA 2.5

[@scottireland4110]

Something I find weird is that the 7900xtx is not the next 6950xt as the 6950xt was meant to compete with the 3090ti, which makes 7950xt or 7950xtx a replacement or at least thats why I hope lol. Note this is a personal thought.

[@Sly_404]

The problem is, they can’t make a 7950 if they can’t make RDNA3 more efficient. They simply have little headroom to increase power draw. AIB models of the 7900XTX partially already draw as much power as the 4090.

[@scottireland4110]

@@Sly_404 yeah RDNA 3 kind of sucks, and is a experiment? Idk what was the appropiate word. But yeah, hope they get 7700 series and 7800 series good value or at least better than the 6000 series

[@PeterParker-status]

Don’t forget monolithic chips can only get to a certain size before physicals says “oh no no no” so even Nvidia is going to have to go chip-let by the 60 cards, Nvidia really doesn’t want to do it because they get such a sweet deal from Samsung buying monolithic chips. Listening to moors law he said his contact at nvidia is pretty complacent about it right now, they believe they will figure it out before they have to go chip-let and amd had a problem that they didn’t discover till the last minute, the kind of problem that only shows after prolonged gaming, so they lock off part of the architecture, or something like that I’m paraphrasing, but if true we could see a very big generational jump with there 8000 series. At least I hope we do, I’d love some competition at the top, it could only be better for use.

[@Zakmakoto]

RDNA 3 is "again" an architecture shift or an experiment from AMD to pursue price/performance ratio or simply viability for an architecture that has a bottleneck (in GCN/Vega's case) Factoring the whole card bill of material vs the GPU if of course one thing, but let's not forget most cards are AIB built (even those we call founders edition) so this is supported by the AIB not AMD, and their only mean of acting on pricing is their main component, the GPU and of course reference design of the card. One thing you missed and that is crucial is the interposer (what they describe as interconnect on their video), infinity fabric is more a protocol, and the interposer of Ryzen differs greatly from the one you find on RDNA3 or even VEGA cards. While they're getting better at implementing chiplets since HBM endeavors it still has a non negligible cost that can only be absorbed by high end GPU, or the interposer price will outclass smaller ones we see on 7600 for instance and won't be worth using.

[@sandrocarletto9386]

the 7600 doesnt draw the same power as a 6600 xt, its just the 7600 shows full board power while the 6600 xt only shows core i believe

[@velesgaming4102]

Totally agree it was a bit of a flop... I was planning on upgrading my 6800 non xt.. And change my mind when I saw all the benchmarks... ended up getting a 6950xt Liquid devil For $350 it had a factory defect In the water block That took 30 minutes to fix. 😂

[@vextakes]

Incredible deal, good on u

[@morgan3392]

I do really hope that RDNA4 is a big step up and is a learning generation. Even if it's a couple years late, it'd be nice for AMD to claw some of the market share away from NVidia.

[@valentinvas6454]

I knew RDNA 3 won't be so amazing when some early benchmarks showed the 6950 XT beating the 7900 XT in certain games. That just shouldn't be the case at all when the 7900 has better specs at everything. More cores, newer architecture, newer node, wider bus, faster memory. On paper these cards should wipe the floor with RDNA 2. Let's hope RDNA 4 will turn out better. I also just noticed that they didn"t seem to increase IPC since RDNA 1. That was in 2019. If RDNA 4 comes out in 2024 they should really do something about that as well.

[@vh9network]

I think you should base your data off TechPowerUp's Relative Performance chart not Hardware Unboxes 1440p avg game benchmarks or Igor's Lab speculated guesses (W7800 is not a RX 7800 XT). There's a 24% difference between RX 7900 XT and RX 6900 XT There's a 36% difference between RX 7900 XTX and RX 6950 XT

[@vextakes]

Lol, so HUB was 1% different? Seems pretty good

[@deus_nsf]

Problem is, Ada Lovelace is hardly ground breaking either, and is absolutely ridiculously priced, so...

[@CrysResan]

What seems like a cool idea(not even sure if it's possible) is if you had a base chip/ PCB unit then modular heatsinks/fans and VRAM that you can buy yourself and slot in. So you could decide yourself what you want.

[@wtflolomg]

Hot takes... RDNA3's biggest uplift was ray tracing... not enough to bring it up to par with Lovelace, but they did catch up to Ampere. Of course, Nvidia doubled down with DLSS3 and fake frames, and AMD still hasn't delivered FSR3. The bottom line is that AMD got greedy. The 7600 is single chip because that's the way it was planned. Chiplets and tiers and the resulting configurations depend a lot on yields and expected outputs and sales. AMD overpriced to play Nvidia's game, when they should have played the longer game of marketshare - meanwhile, the RDNA2 cards are gaining ground at the low end (which is the deepest market). I believe there was also some unexpected problems using 3D VCache tech on the new architecture... If AMD had been smarter, the 7900XTX would have been the 7800XT, and the 7900XT would have been the 7700XT. They would have targeted a sweet spot that Nvidia wasn't engaging with Lovelace, and outclassed anything Ampere stood in that space at that pricing. We all know the 7900XT isn't even close to that tier. It also would have allowed AMD to refresh this year with a new halo card when they finally got FSR3 off the ground. Missed opportunity. In the end, neither AMD nor Nvidia has anything to really offer this generation. Nvidia's offering is just as milquetoast compared to previous generation, worse in the low end, for an obscene price increase. Reject both until they come around. Last gen is "good enough" until Blackwell and RDNA4 launch.

[@MarksChannel52]

Think like the 6000 series we have to wait 8 to 10 months of updates before we will actually know.

[@vextakes]

Lol it’s almost been that long