i9-7980XE & 7960X Review: Delidded Thermals, Power, & Performance

[@GamersNexus]

Publishing this from a plane that's on the way home. There is one missing chart in this video (oops! Sorry!). You can find both over time charts in our article, published at this link: www.gamersnexus.net/hwreviews/3066-intel-i9-7980xe-7960x-thermals-power-review The article also has 'auto' setting temperatures, which are not comparable for TIM vs. LM (as voltage is variable), but will give a better idea to the thermal positioning of the CPUs out of box. See above link. 7900X delid video: kzbin.info/www/bejne/najUiptsmJiLmdE Yes, it's dark when I'm on camera. Shot/edited this one alone while traveling, but we'll be back to normal tomorrow.

[@PeterNjeim]

Forgot to mention absence of ECC memory support on x299 yet exists on x399. These can't be used for workstations, they have the same usefulness as the i7-7640X and i7-7740X.

[@GamersNexus]

Sure they can be used for workstations. You don't need ECC for all types of workstations. They're more enthusiast-workstation hybrid parts for that reason, among many others, but not every workstation needs ECC. Ours don't, for example.

[@bossromeo9158]

500watts wow wow weee woow

[@AbyNeon]

Your 1950x OC benchmarks are off by 15%, you might want to check your overclocks , as many gets at least scores above 3400 way below 1,4 volts

[@GamersNexus]

We're not really running OC 1950X benchmarks. Those are just stock benchmarks. The OC was explained in the 1950X review -- its performance is bad because XFR is better than 4.0GHz all-core in most applications.

[@richiec7700]

Good guy Intel. Making 1200-1600 Watt power supplies useful again.

[@mapesdhs597]

I thought I was helping with quad GTX 580 3GB for CUDA. :D www.sgidepot.co.uk/misc/3930K_quad580_13.jpg

[@SlowHardware]

Ha

[@octoman_games]

40 amps! @ 500 watts?! HOLY SHIT! AMD is no longer the "Toaster" maker.

[@WeirdSeagul]

toasters easilt draw that much. mine is rated at 900W and after quick googling that is not that high

[@mmmako]

Even blenders go way above that.

[@atearen]

Actually measured my toaster, achieved 600w.

[@Nobody-vr5nl]

n00bie don't look at it gf's hair stuff. Even that kind of thing draws more power.

[@reav3rtm]

Intel Core is 1st "good" Intel CPU architecture. Intel was traditionally smashed by AMD in perf/price ratio over the years since K5 (AMDs first truly own CPU). Achieving comparable performance with smaller clock speed and smaller power draw.

[@Hardwareunboxed]

Awesome video as always GN team! Steve how do you move around so easily with that giant pair of brass balls? Delids a $2000 CPU, right gotcha :D

[@Nobody-vr5nl]

Hardware Unboxed honestly deliding isn't hard. All u really need is a cheap bench vice.

[@Hardwareunboxed]

haha yeah I've delidded many CPUs mate, none have cost more than $1000 though. That wasn't my point. There is still risk involved and if you get it wrong, well ya just brunt $2000 US. Not only that but for consumers you also void your warranty even when successful.

[@D87468]

2K$ cpu, pulling out 450-600w+ on BGA "sandwich" and maybe some other sneaky things (Broadwell-E + XMP, Haswell FIVR), without warranty - very stupid idea.

[@KarrasBastomi]

Those cpus are engineering sample from intel..... So, it's free....

[@mduckernz]

Karras Bastomi Mhmmm, but if you fuck up, they ain't gonna send you another one.

[@unclerubo]

Remember when people bashed AMD for making "space heaters"? Now AMD is making the more efficient CPUs. People have to learn not to be fanboys...

[@DivineF0xGoD]

Klaus Morgenholz intel still has better cpu's you just have to replace the TIM which also voids the warranty so get kinda fucked there

[@unclerubo]

I am not saying that the 1950x is faster than the 7980XE, it is obviously not. But it costs half as much, is much more efficient, and the 200% price difference is far from providing 200% the performance. Neither of those are CPUs for me, as I don't need nearly as many cores, but AMD is clearly providing the more balanced product here. And just in case someone is going to call me an AMD fanboy, my main PC is running on a 5820K.

[@EvilWiffles]

I don't imagine these chips are aimed for people who care about budget but want the best possible performance possible, regardless of cost efficiency.

[@unclerubo]

I don't think that big studios with virtually unlimited budgets, like Pixar would buy these. They would buy monsters like higher count Xeons or Epycs to help with their render farms. But good old Mike, who runs the local wedding filming shop and has a single system would probably get the processor that gives him 80% of the performance, if not more for half the cost.

[@mapesdhs597]

Klaus, that's spot on, see my longer post elsewhere for example details of one such studio.

[@psikogeek]

So....... Intel is selling a do-it-yourself processor?

[@railshot888]

psikogeek no and it's weird these high margin chips from Intel requires a delid or a boss custom loop for heavy use. Terrible terrible thermals... and Damn 40amps 500 watts lul

[@psikogeek]

I have to agree that it is really, really " *weird* " like you say.

[@peterpimmelmann3330]

for only 2000$

[@Legiro]

@Aron Lee, since those processors have maybe only OC above Xeon solutions (and such price point) - it is do-it-yourself in reality. Intel just didn't care enough, it seems. Because to tap in that overclocking potential, you need to delid and replace TIM, and then you have to somehow solve the issue of cooling the motherboard components too (considering absurd power going through them), because passive solutions provided by motherboard manufacturers aren't enough. The question arises - why would people bother, if they can switch to Xeon platform at that point. Maybe that was the whole point by Intel? I just don't get where simply bad proposition for the price fits in the picture, where lack of the solution wouldn't. Hoping someone uneducated buys in? At that price point, with the offer being targeted at professionals who know what they need?

[@psikogeek]

"breaking your cpu by delidding will void the warranty." From Intel's point of view, that is a feature not a bug.

[@crimsun7186]

So, you could say that Intel was jealous of how Vega uses so much power and decided that they could do it better.

[@DivineF0xGoD]

CrimSun power usage on a CPU is understandable no reason for GPU to suck that much power for such little performance

[@smokeydops]

LUL

[@ElZamo92]

DivineF0xGoD dude, usually GPUs are much more power hungry than Cpus. for example, a midrange GPU like the GTX 1060 (a very power efficient one at that) has a TDP of 140W, while every consumer grade CPU on sale right now has a TDP lower than 100W.

[@TwistingDarkSun]

I think this is more closer to a 2017 Fermi than a Vega

[@itselectric6199]

Ana Sevi Vega is actually relatively power efficient for the GCN architecture with lower clocks (around fury x levels) and massively lower voltage. You could probably get it to around 200 watts, where fury consumed over 400 watts. Problem is, IPC hasn't really changed, so you'll get the same performance.

[@morzone]

It's sad that Intel called them "sandwiched together" yet when you look at thermals of their offerings, they seem to be eating their own words in term of power efficiency.

[@Kureiji-Desu]

Just wanna know who was the genius behind that statement.

[@JarrodsTech]

Great in depth details :) still no regrets with my 1950X, these things are expensive!

[@Decenium]

im jealous, would love a 1950x just to crush the hell out of livestreaming :P

[@Arlyon9999]

Jarrod'sTech I want a 1950x but I can't justify to myself upgrading from a 6850k. Hehe

[@TecnoCR]

2000$ +`95 Cpu temp + 44 pciLines +440 wats = i9-7980XE VS 1000$ +70 Cpu temp +64 pciLines + 150 wats = TR 1950X

[@GamersNexus]

Well, no, not entirely. The 7980XE runs a lot cooler on auto. These are fixed at a higher voltage to show the differences between LM and TIM, as auto voltage is variable and can't be used to compare head-to-head.

[@mungojerrie86]

BUT MUH IPC!!!!!111111

[@SWEbear021]

also no ECC memory and u need to pay for RAID xD for that money just buy epic this is trash for that price:P

[@greebj]

1000$ + 95 CPU temp + 44 pcie lanes + 250 Watts = i9 7900X is the better comparison Which the TR 1950 is >25% faster than in MT which matters more for these CPUs than ST and which you don't need to delid and void warranty on if you want to OC 7980xe is faster than TR 1950x by 10% @ stock, ~30-50% @ max OC ... if you can get there due to the mega power draw

[@saladwithsalad]

Southeastern777 Why are you so sure?Just because you can't afford one doesn't mean he also can't afford to get one. I can buy a 1950x but i don't fucking need it.

[@bobohbuboh8604]

Good job Intel, welcome to your Bulldozer era

[@osgrov]

This is exactly the type of quality content I'm paying you guys for. Proud Patreon backer, keep up the great work! :)

[@simiolonga]

In order to get good t° you need to delid it and void its warranty! BRAVO, INTEL.

[@railshot888]

Sausageape the typical Intel replies would be "who gives a crap about warranty and delid kit is only $30" something stupid like that.

[@foreignsgamingtech]

The recent content has been so constructively critical that it brings a tear to my eye. Great review!

[@videocardzrule354]

I just purchased one of these. Here we are in 2020! And processors are in plenty in all flavors too! I paid $300 bucks for my 7980XE! It should last me years. Mine is not delidded. I am coming from a 5.3Ghz i7 8086K. So, I will lose a little top end HP. While gaining massive bottom end torque with this new 7980XE big block CPU lol. The 7980XE is still a monster, and still one of the many CPU kings available in 2020! So cool! I can’t wait to run it.

[@folterknecht1768]

Shoving down nearly 500W thorugh the VRMs on a daily basis for several hours ... motherbord failure rates will go through the roof, resulting in system down time eating up all your gained production time.

[@itselectric6199]

Folterknecht I don't think any board without active coloring on the VRNs would be able to handle that kind of heat long-term. It's just too much for passive cooling on an ATX board. If the board size was SSI-CEB with 24 phases, maybe it's possible.

[@Filmer1eX]

Except that no one overclocks a system meant to be used professionally. Not trying to defend intel in any way mind you, just pointing out that the power usage isn't nearly that high at the settings most people will end up using these CPUs (that being stock).

[@mexicanmanjohn]

have to watercool the motherboard and cpu for sure

[@hydrochloricacid2146]

Filmer1eX I'd argue that the entire point of these chips is overclocking. You're trading ECC for overclocking in HEDT. Otherwise it offers no improvement over xeons

[@kagex5426]

Not true. The Xeons have a far lower Turbo Frequency compared to the X-series and are more expensive.

[@keybraker]

AMD TR 1950X 16C/32T (BS) : 143.91 Watt AMD TR 1950X 16C/32T (OC 4@) : 274.29 Watt INTEL I9 7960X 16C/32T (BS) : 217.71 Watt INTEL I9 7960X 16C/32T (OC 4.6@) : 492.00 Watt

[@SupermassiveGaming]

Everytime I hear liquid metal I think of Terminator 2

[@mewimi]

+Supermassive Gaming Every time I SEE liquid metal I think of Terminator 2

[@mapesdhs597]

I wonder if a LM supplier could get away with calling a product T1000? :D Btw, read the T2 book, fascinating stuff about how the stuff "works". Detail not present in the movie.

[@GamersNexus]

Spreads the same way!

[@Astfgl]

That is the only correct response to hearing that term.

[@crisray6789]

Only the most thorough reviews here, Awesome!

[@DrSpychology]

Damn, I love threadripper but this is even worse than I expected. Obscene power consumption, horrible temperatures, insane prices for a tiny bit more performance. Also, I hope no one actually takes a 2000$ gamble on delidding these CPUs.

[@erejnion]

Well, it's not a tiny bit in some areas. People may actually jump on this if the CUDA thing is reproducible for any project.

[@r3dleh107]

tom from OC3D, der8auer and you are actually my favourite youtube reviewers. great job

[@randomdudewalkingdownthest8158]

I'm pretty sure the unavailability of ecc ram on x299 platform is enough of a deal breaker for many studios

[@jominator3187]

I've been waiting for this specific review all day! - Thanks Steve, and your peeps at GN!

[@timothygibney159]

Holy shit. A nuclear power plant in both electricity and heat. For stable workstation use it seems a Xeon would be more sensible for these reasons. Yikes

[@vegetablesaregoodforyou7008]

or threadrippers and epyc. workstations would often prefer using ECC memory, the x299 platform doesn't support it, but x399, epyc and xeons do

[@JusttheEdge]

Timothy Gibney If you lower clocks to Xeon levels the power will drop. ECC is the only benefit of Xeon.

[@mapesdhs597]

Scalability is also a XEON benefit, and low noise in the workstations that typically have them (my Dell T7500 is absolutely silent). If I had the relevant cash though, I'd go for a dual EPYC board. Mind you, for such budgets there's a lot of options in refurb previous-gen XEON blades when they get dumped by datacentres each year.

[@timothygibney159]

Xeons are stable and very well supported. This is why corporations buy them. It seems the i9 was rushed and in a hurry Intel invented a chipset and glued abunch of i7 cores and called it an i9. Both Dell and HP workstations use them for power users. I bet even underclocked the Xeons would use less power and be cheaper too. I don't get the point of this other than to say ME TOO to AMD.

[@mapesdhs597]

I really do wonder who will buy the XE, it seems such an odd product; premium price, but without any interesting premium features.

[@ugotznuked]

I was WAITING for this level of detail, thanks guys ❤️

[@hellboybihac]

Finally! The only review that matters.

[@chukku2175]

4:00 Waaait a minute.. there is some mistake hidden somewhere here. The "Water" Temperature should be the same with TIM and with LM. This may seem counterintuitive but: the thermal interface (LM vs. TIM) is only responsible for the heat transfer coefficient. And that coefficient only determines, how much temperature difference you "need" between CPU and Water to transfer a certain amount of Heat (in Watts) away from the CPU. So for example the 7690x needs to have 93°C to "press" 165W through the TIM but only 79°C do the same with LM. But in BOTH CASES the amount of heat transfered from the CPU into the Water stays the same (165W). So there is no reason, why the water should get warmer. Or to look at it from another angle: With a Liquid Temperature of 54.7°C the radiator of the AiO is emitting roughly 40% more Watts into the room than with a liquid temperature of 45.5°C (assuming a room temperature of 22°C and assuming that the fans stay at the same speed). Where are these extra Watts coming from, when the CPU is running at the same clockspeeds and the same voltage? If anything the water temperature should be lower with the TIM because more heat is radiated directly from the cpu into the Mainboard instead of into the water. Is the temperature sensor positioned inside the asetec pump or near the radiator? Maybe its readings are off because it is influenced by the warmer surroundings?

[@KrunchyTheClown78]

Intel is stumbling badly in the uber high end segment.

[@sledforlife2327]

Watched Paul's and Linus's videos this morning, but I knew yours was going to be worth the wait. Great review and video!

[@4sm0deus]

@Gamers Nexus: Der8auer and OC3d TV had a discussion about extreme VRM temps on x299 MOBos, which occurred when stress testing cpus with prime 95, because the power draw goes through the roof. Perhaps something similar happened in your tests.

[@fabianbuckreus185]

I work for a VFX Studio and have the I9-7980XE on preorder. It wasn't an easy decision but in the end I picked it because there is a lot of stuff in vfx that is not able to take advantage of a high core counts and you want a higher clockspeed on single cores, like for example particle simulations! But other parts in the same workflow like rendering are HEAVILY multithreaded. That's why I think this cpu is the best from both worlds. I hope I made the right choice! :D So far it looks like it.

[@WeirdSeagul]

ahh i will order 50 of these for my company and manually delid them all so i can use them xd. jk cool cpu. at this point might aswell get a server with a xeon or epyc for your cpu accelerated work aswell as the advantage of ECC (threadripper has it tbf) and pcie lanes

[@roikhanulyahamidan9978]

Frothar also 50 of them is still much slower than 100 unit of Threadripper though

[@JWhite-Fishing]

I don't know much about Skylake X and its AVX situation but speaking from experience, my Broadwell-E CPU when running Anything above P95 26.6 draws way too much power. After 26.6 P95 introduced AVX instructions and on BW-E this caused some serious issues. The same can be said for Haswell. I wonder if you were also experiencing the increased power draw due to AVX kicking in. ASUS motherboards (no experience with other manufactures) are told to dump more power into the CPU when running AVX (at least on X99 anyway). I'd suggest re-testing with P95 26.6 (Non-AVX) to see if this is the case. I actually had to set an AVX offset on my 6800K due to some games on Cryengine using the AVX instruction set. At 4.3Ghz it is not stable using AVX at all...

[@TB-zm3zc]

@Jonathan White, BINGO

[@subsgob00m]

Since Ryzen scales so well, I imagine an OC'd 32-core Epyc processor will most likely consume around the same or less power than the OC'd 7980XE while providing 14 more cores and potentially running cooler on top of that for potentially around the same $2000 price tag. I'm curious what the performance will end up being, but it sounds like that dollar for dollar, AMD is king across the board for production purposes. I'm hoping they get what they need to push the IPC higher on Zen 2 with better overclocking to net a 10-20% improvement on single thread performance. That would really solidify AMD in the market even more so than they have with this year's round of processors. I'm really wanting to make the jump to AMD, but the single thread performance is a must for things like Lightroom and Photoshop since the multi-thread performance is quite lacking (I'm looking at you Adobe...).

[@tuomasheikkila3628]

Speaking of the cost of these CPU's, I think we've hit a price level where actual server processors would be far more feasible in multithreaded professional workloads than something like the 7980XE. Platform costs notwithstanding, I'd say that something like the Epyc 7551P at 2100$ would make a hell of a lot more sense than this. More cores, more PCI-E lanes, ECC RAM, server grade components for about the same price. Still, these are extremely impressive chips for what they are, I'm just not so sure who these are targeted at.

[@AMDRyzenEnthusiastGroup]

These two CPU's are nothing more than a gimmick, aimed at crazy people who like to waste their money. Intel simply wants to claim that their CPU's are faster, even if they'll fry your motherboard & trip your breakers. I can't see anyone with any common sense, buying these.

[@UsingInvisibleInk]

nappydrew I’m sorry but that is very ignorant of you to say.

[@AMDRyzenEnthusiastGroup]

No need to apologize for your opinion. Not hurting my feelings. If you like it, buy it. None of my business what you do with your cash.

[@Kojiro3210]

1950x @ 4.1ghz 1.4685v is just under 350w in cenebench 15 and scores in the 3500's 4.0 ghz @ 1.365v stays under 300 for the most part and scores in the 3400's 64gb 3200 14-14-14-14-34 Gskill Ram

[@PakoSt]

Hope you will get some rest! Outstanding content in everyway you look at it

[@flyjum]

Why is the 4.0Ghz TR chip slower than the stock TR chip? 13:10 is the chart I am referring to. I feel like the two are swapped as a stock TR runs a bit over 3k points not near 3200.

[@saladwithsalad]

XFR boost higher than 4ghz.

[@RockRocky09]

Nice review and summary, subed!

[@ocidax]

Clear winner: AMD. The power consumption on those OCs Intel chips is insane. No worth the extra money and the stock performance advantange is not that great compared to TR.

[@johnsimbana5660]

Lord OX yes amd is the winner, is the best for the price

[@TheySeeMeTrollen]

So premiere is the perfect use case for the skylakex arch. the extra L1/L2 cache seems to agree with rendering. good to know.

[@SuperTurbo9001]

Weren't the x299 boards having vrm heat issues with the 7900x already?

[@AMDRyzenEnthusiastGroup]

Yes. The X299 platform is an overpriced gimmick, that was rushed to market because of TR. Anyone who actually really needs a HEDT, would be batshit crazy not to buy a Threadripper. For die-hard Intel fans, it's much wiser for most ppl to wait a week for coffee lake. The 8700K legit looks decent.

[@FabiVoltair]

Great review as always

[@2000cobraguy]

First, you work too hard. Great review. It is difficult to fathom how much actual work goes into your reviews, especially behind the scenes. This stuff does not magically appear. Second, on the X299 platform, I noticed that no one has actually done a VROC build, benchmark and review. Just a lot of complaining about how Intel is "greedy". How do the Samsung M.2 OEM drives work on it? (They are on the "approved" list) Does VROC eat up PCIe lanes? How hard is it to get the standard "un-locker", and/or the premium "key"? How much do they cost and where to get them? Part numbers? Stuff like that. Thanks. Well done.

[@mapesdhs597]

Steve, big studios that need max speed normally use multi-XEON blade servers (perhaps an area where AMD could gain ground with EPYC?). A company I know well, last time I asked 2 years ago, had 7000 cores and a 10GB/sec SAN (I won't mention the name, but they're larger than ILM). Curiously, such companies that seek max speed (they're now working with uncompressed 8K and 9GB/sec PCIe RAID) are also the least flexible - multiple offices worldwide, hundreds of employees, etc. There's pressure to standardise and suchlike. These are the very companies which don't tend to use consumer tech, and they don't buy or build oc'd setups, at least not for in-company official use (individual staff might have their own more exotic systems at home). The scale of rendering operations tends to be so large, they buy more render nodes in bulk and deploy them fast (the above company uses Alfred for render management, Centos for the nodes, so its all compatible with RHEL for the HP IFFFS setups). Way back I offered to loan them a 5GHz 2700K PC with a Quadro 6000 just to test out (would have blown their 1P Dell box away completely), but the admin eventually said there was no point as there was no way of absorbing anything custom like that into their procedures, even though he personally would love to buy and use such things (way better interactive performance for Maya). How's that for an irony then? Movie companies which can most afford the XE are the least likely to use it, and the lack of ECC kills it anyway. Btw, in the late 1990s I was offered an admin position, which I turned down (for various reasons); if I had worked there though, around that time they were beginning RAM upgrades for the whole setup, the budget for which was approx. $6M. Shows the scale of such operations. Smaller studios might be more interested in this sort of thing, but even then it probably makes more sense to go XEON or even EPYC, or there are previous-gen refurb XEONs they can go for. These SL-X models just pose too many thermal and cost/noise problems. Reminds me of the XEON in Apple's dustbin Mac Pro that would overheat and force throttling down to a measly 2GHz when the cores reached 100C under load. It's really solo professionals and very small studios (less than a dozen people) who might be interested in the XE, but even then the high cost may put them off, especially since that cost (plus the cooling cost and consequent noise) would be a serious impedement to risking oc'ing, in which case a TR cluster becomes way more sensible on cost/power/noise/heat grounds anyway (or again, previous-gen refurb kit; XEON blades dumped from datacentres can be very attractive, eg. I was once offered a stack of S1366 dual socket XEON blades, 12c/24t and 32GB RAM per blade, for just 300 UKP each). Someone I'm helping atm was typical, solo pro artist, budget just not high enough for this class of CPU at all (sadly not TR either); infact the entire system I'm sorting for him is about a third less than the cost of just the XE on its own. :D This is what intrigues me about what Intel is releasing here, who the heck is going to buy them, and will they really oc them that much given the risks? The latter was already an issue when the 5960X came out. At least Intel had that Tuning Cover option that was insurance in the event of failure, but only for certain models. Ian.

[@ChucksBasix]

It wouldn't surprise me if AMD made a comeback in datacenters and similar... Many older Blade systems from IBM used Opteron processors from AMD.

[@mapesdhs597]

Only thing that bothers me atm is the lack of mbd choices. Tyan/Supermicro don't seem to have an option that would be suitable, ASUS/Asrock have nothing, and Gigabyte's MZ31-AR0 is hard to find. EPYC looks like an excellent choice for a really powerful workstation, but atm there's nothing akin to Intel-based WS boards (whether consumer like the P9X79-E WS, X99-E WS or workstation like the Z10PE-D8/D16). There isn't any X399 board that's built akin to the Intel WS models, which is a real shame. Atm X399 boards seem aimed solely at gamers (which given the modern lack of wide SLI/CF dev support is a bit weird). Thus, when it comes to prosumers, AMD has the great CPU options, the mbd choice is lacking.

[@christopherbohling5719]

Steve I will give your patreon $100 if you take a day off and take a nap

[@disk0__]

Check the article for additional sleep www.gamersnexus.net/hwreviews/3066-intel-i9-7980xe-7960x-thermals-power-review

[@gabrielofpark]

Nice vid Steve but holy smoke those power draws are insane.

[@OGAM_Visuals]

Threadripper does better in Cinebench at stock speed than OCed to 4GHz?

[@GamersNexus]

Yeah, because its XFR is boosting it higher than 4.0GHz. See the TR review.

[@OGAM_Visuals]

OK, thanks! Also, how cool is that you answered! Also, great job. You're setting the benchmark for everyone else. Thanks!

[@sparkyenergia]

It has to do with XFR. When you overclock you loose XFR so you only run at the overclock speed of 4ghz. By not overclocking you allow XFR to hit it's high clocks with thermal headroom. In the case of the 1950X that's 4.2ghz. It can only do that on a few cores though.

[@D87468]

In ST. But, this score - 3036 in MT, where XFR not in use (2 core) and 3.7 all? Which lower than stock - 3170. Thermal throttling or error? OC@4 must be around ~3400.

[@mik995px]

Hardware unboxed got the 1950x @ 4ghz reach like 3425 at multithreading cinebench though.. something doesnt add up. further testing might be needed :P haha

[@toddsimone7182]

Steve at 13:00 the Cinebench chart I think you have the TR 1950X's swapped. The stock chip scores higher than the 4Ghz OC?

[@OscarRobbing]

Your 1950X OC results are very strange, ie. in Hardware Unboxed's Cinebench tests, their 1950X stock was a bit lower than yours at ~3100, but when overclocked got up to ~3400. Yours gets lower for some reason...

[@AMDRyzenEnthusiastGroup]

No 2 CPU's are the same. Silicon lottery is a very real thing, my friend. Not to mention different variables, for different testers, ike cooling solutions, PSU's etc.

[@rako81sna]

Well Prime 95 is using different workloads. Some heavier loads starting kinda late, so the power consumption goes to the roof then.

[@sreekanthbhagavatula]

For the amount of detail you provide, your subscriber count should be higher than Linus.

[@kineticman123]

Hello I had some questions in light of this video. Can you suggest how often we should reapply liquid metal TIM after first application? I ask this question generally and not specific to these specific processor. What was the relidding process? Did you use some sort of super glue or instant gasket maker? If so, how does that affect the PCB if you need to delid again for reapplying the liquid metal TIM? Also, what exactly happens over time? The liquid metal dries up? Sorry for so many questions.

[@brendanbyrne3028]

12:25, how is the OC'd 1950X scoring lower than at stock?

[@IK694]

Can we get a step by step delidding and liquid metal application video, curious about putting the sealant around the IHS

[@EdgyNumber1]

Nearly 500+WATTS!! No company could justify that cost - nor any enthusiast unless they don't mind chewing through motherboards and their VRMs...! Btw, I notice in one test Vega and Threadripper seem to be quite a powerful stock combination.

[@superbibibobo]

The review I've been waiting for

[@TemesPlayer]

Hey Steve! I think one of the better gaming-related use-cases for these HEDT CPUs is in an encoding PC in a 2-PC streaming setup. Could you perhaps test how well these flagship i9 and Threadripper CPUs can handle the slowest of the OBS x264 encoder presets, and if there is a thread limit in the OBS x264 encoder?

[@caiocc12]

Prime95 has different tests parameters which are executed one after another (I think it is FFT size). Some of them consume more power than others, that's when the power usage spikes and the computer turns off to protect itself :)

[@Mandurath]

So, 7980XE and a Vega 64 and this winter, the extra electricity used for work can also apply to heat the office.

[@Pewpiepiepewper]

Watching you apply liquid metal is like watching someone with parkensons knit. Good lord. Great video though.

[@dickr9345]

Pewpiepiepewper Why are you watching people with Parkinson's knit? ;)

[@Pewpiepiepewper]

That shit is entertaining

[@mapesdhs597]

More knot than knit, hehe.

[@ChucksBasix]

You would probably be in the same boat if you was applying LM to a $2000 CPU.

[@rodneyrogers1319]

Curious how these “Flag ships” would compare against dual system set ups. Ie is heat for all the hardware still an issue, are their better gains? Xeons are usually cooler since they have lower clocks right? Just tremendous cache? I’m just a wannabe enthusiast trying to learn to be like the cool kids.

[@mapesdhs597]

Where cluster arrangements are viable, yes indeed tying multiple systems together can be more cost effective, but it's a tricky equation to get right, so many things to consider. Sometimes the admin complexity is also an issue; reliable XEONs over uncertain behaviour from XE (even ignoring the lack of ECC) would sway the movie corp admins I know to the XEONs every time, or perhaps EPYC at some point I hope (32c for 2100 UKP is pretty sweet).

[@Hybris51129]

These are glorious results for the high end customer now we can only hope that Intel will do something about their TIM.

[@df1299]

What about under volting performance on the HEDT intel chips with Liquid Metal? If you can run at stock clocks but under volted maybe that would be more eye opening to intel...

[@alexmarin7897]

The real champion of this Skylake X lineup is actually the 14core 7940x which is higher clocked as well. Shame there are no youtube videos on it.

[@kalmtraveler]

with the 9980xe out now... I'm taking another look at the 79x0 chips for upgrading my HEDT rig - for a machine that will be 50/50 gaming + productivity would you guys suggest the 7960x or 7980xe ? money is not a concern here, just want to make sure I pick the chip that will make the most sense for the use case since tearing down a full custom loop HEDT rig is a full day's worth of work.

[@alanschamber]

So... AMD ThreadRipper 1950X, consumes 144 Watts at stock, that is 3.4 GHz for all cores, and up to 4.2 with XFR. Meanwhile, the 7980X consumes 214 Watts, with a base clock of 2.6 GHz. Blender stock AMD TR 1950X: 15.3 Intel 7980X: 13.3 Performance result: 15% higher for Intel. But with 48% more power usage. Is it just me or we are facing the new performance per watt kings at AMD. By far.

[@alanschamber]

Oh, yeah... AND PRICE. Once you consider price, I believe there shouldn't be a single person buying those intel chips.

[@Realtime1501]

Well Ryzen 14nm process is essentially a mobile cpu process

[@chincemagnet]

Extraordinarily impressive. 4.5-4.6 GHz on an 18c/36t CPU, If you can keep your system from catching on fire.

[@Realtime1501]

David system? X299 uncooled is enough to ignite the atmosphere

[@Decenium]

I actually would love to see a livestreaming test with these high end cpu's. OBS with 1080p and the "placebo" cpu preset will probably require a lot...

[@Brianpeppers15]

Good work!

[@OscarRobbing]

I love how AMD is far and away now king of thermal and power efficiency, when their previous architecture was the exact opposite - pretty much as inefficient as possible.

[@EpicListening7]

Great in depth review, Shampoo advertizer! Nice atmosphere you have in there! You found yourself a girl, my friend?

[@Dazzxp]

Holy crap, intel have released alcoholics of electricity out on the power grid. RIP AMD FX you have been dethroned on power consumption. RIP motherboards your maker is here, Clock this thing to 5.6GHz and it sucks 1Kw to itself. Energy star rating -F for fking hell

[@Atilolzz]

isnt using Q Tips bad because littoe cotton fabric could be stuck in the liquid metal?

[@cheddle88]

Why are the 1950x firestrike physics scores so low??? did you run 2133mhz ram speeds or something??? you should be seeing 29,000-32,000 points at 4.0-4.2ghz all core.

[@zettle2345]

The more bench marks I see with Premiere, the less I see the use of it. imo Hey Steve, would you recommend people wanting this chip, to make sure they have more than just 1 8 pin connector on the Mobo they buy?? ;)

[@kurosumomo]

Looking at the power consumption and performance, I wager it would make a lot more sense buying two Threadripper 1950X, splitting the task, and still remain under the power consumption of a single OCed 7960X/7980XE chip. The biggest bonus for the Threadripper family is ECC support, which Skylake-X does not. While the OC perfromance (which no business running these CPUs 24/7 will do) of Intel is commendable, the power draw and no ECC support is not, adding pricing to that and I think the AMD option is the better investment for businesses and home users.

[@minecraftermad]

how about putting thermal sauce around the die and liquid metal on it? could this improve temps?

[@sumfoo1]

I don't understand the thermodynamics of this, how does the better conductor on the CPU reduce water loop temp. The total power consumption of the CPU hasn't changed and technically more of that power would be transferring to the liquid loop. So how does the loop temp go down? What your test indicates is better heat transfer from the radiator not on the CPU side. Q in = q out and q stays constant. Loop temp should be set by the radiator unless somehow the delidded processor somehow uses less power at lower temps ?

[@GamersNexus]

We talked about that in the 7900X TIM/LM video a bit. There is some discussion in the comments.

[@VikingPickles]

OK, so the 7980XE is a beast. I'll be getting a 1950x though.

[@rudrashivagaming4377]

Steven Davidson what smite 7960x or 1950x??

[@VikingPickles]

I don't quite understand your comment.

[@rudrashivagaming4377]

Steven Davidson I mean intel 7960x or amd threadripper 1950x? Which is better?

[@VikingPickles]

pretty sure the 7960x is better, but the chances of a better chip launching on x399 is much greater than on x299. So 1950x now, 3950x in a few years.

[@erejnion]

Dat render time with CUDA. That's a market there.

[@hypershadow5g]

Why does GN use LFFT's when just testing thermals?

[@MrDjMichii]

is a 360mm AIO enough for 24/7 18x 4,2GHZ OC ?

[@bluedragon219123]

The reason Intel uses TIM instead of soldered isn't being cheap, mostly. It's about having the CPUs degrading faster. If your old Sandy Bridge, which uses soldered, can still go strong without throttling then there's no reason to upgrade. But if it does then there's a reason even with minimal improvement in performance(I'm ignoring Coffee Lake of course). Most "Techies" will use an aftermarket cooler and slow down it degrading. But Dell or HP? Those CPUs will degrade faster since they, usually, not a K series and will use the stock cooler. Not a good combo. I could be wrong on all these but it does make sense. :)

[@atretador]

question: can you use regular thermal paste(to cover the PCB)to prevent LM from...you know, blow shit up?

[@GamersNexus]

Better to use nail polish or high-temperature tape.

[@mungojerrie86]

7980XE and 7960X are Vega 64 and Vega 56 of CPUs.

[@hydrochloricacid2146]

Павел Мастрюков well, vega 56 is actually worth buying though.

[@mungojerrie86]

Well, 7960X is also kinda sorta worth buying.

[@hydrochloricacid2146]

Павел Мастрюков its 80% more expensive than the competition, while not being substantially faster, and using more power. Price gouging aside rx vega 56 is about the same price and faster, but using more power.

[@Zecrid.]

HAHAHA! almost 500 watts from the 16 and 18 core CPUs overclocked that is amazing... in a biblical sense.

[@EspHack]

you need to include vega when talking gaming performance now, hardware unboxed did that and its really interesting how different combinations of vega+intel or vega+amd makes the amd cpu get more fps than even the supposedly "better at gaming/single core perf" intel stable of cpus when running vega

[@SaccoBelmonte]

So basically an overclocked 1000USD TR 1950X would give you the stock scores of a 2000USD 7960XE for half the price and less power consumption / heat. Don't get me wrong, the 7960XE is an absolute beast. But it's price and power consumption is not attractive for the average content creator enthusiast. Only a rather mid/large company would consider one of these while TR is still very appealing for small businesses. So AMD is the only truly bringing 8 and 16 cores to the masses. The only move Intel really should to do is to cut prices in half. Or at least 35% or 40% to make the consumer happy....as of right now there is a noticeable and well deserved anger towards Intel, they have to change the way they make business because the general public is truly pissed.

[@twoface8717]

how come, that in your tests, the Threadripper CPUs are much closer to the Intel CPUs? I watched some other reviews where the TR CPUs are shown to be much worse. Or maybe Î just missinterpreted some other reviews. I realy like how close the 1000$ Threadripper comes to the 1700$ i9.

[@Runoratsu]

It would be cool if you could add a C++ compilation task to your benchmark tests!

[@yourtechstation]

I wanna see what the power draw is for a system with the 7980XE and Vega 64. Also will it heat the Gamers Nexus studio in the winter?

[@mapesdhs597]

If the lights go out, is that a bunch of dudes with that combo at a LAN party, or a NK first strike? :D

[@Maxvla]

OC'd 7980XE (500-600W) + dual Vega64 Liquid edition in crossfire (400W x 2) + 128GB DDR4 (~50W) + high end CLC (20-25W) + misc power for SSD, etc... You're looking at needing a PSU that can handle sustained loads of 1500W ($400-500). Bonus fact: this system would cost ~$1600 per year in electricity ($.12 per KWH) if at full load at all times.

[@NoahNasi]

What about the 7940X?

[@kgrach]

easier way to spread TIM is with a safety blade. Qtips can leave fuzz and can be really bad. Hot spots created by fuzz can kill a die.

[@AG-pm3tc]

AMD is again the efficiency king... I never thought i will get to that time again 😂

[@shailoism]

Whats the fastest single threaded CPU right now?

[@luckythewolf2856]

Hey Steve you sound sick. If you are hope you feel better :)

[@petrefrangulea2522]

the 1920x OC is missing from most of your charts

[@treborheminway1196]

I think it would help your presentation to remove some of the cpu's in your graphs. If your comparing high end CPU's, it just complicats the viewer experience to find the specific cpu's your talking to when burried in a sea of mid and low end cpu's.

[@imnotusingmyrealname4566]

Did you know that if overclocked to 6.1 GHz on all cores the 7980XE draws 1000 Watts? (credit to der8auer)