Intel is BACK | Core Ultra 5 125H feat. GMKtec NucBox K9 mini PC

[@_ravelin]

that intro made me think something is wrong with my pc

[@affhmd]

I dont have a pc 🗿

[@shrishpotdar343]

4:52 Haa, love the Dawid does tech stuff reference, once again great video iceberg tech :)

[@kagamine1993]

Not in a million year i expected to see dawid here but here we are i guess 😂

[@UltraVegito-1995]

I miss "Intel leap ahead" jingle so much ☹️

[@auritro3903]

now its called "Intel leap backwards"

[@barwit12345]

Ok, I officially love you now. All those times you'd intentionally avoided mentioning the number 69 and the inevitable bad jokes that would follow, but the one time you get the shot to make an OVER 9000 joke you immediately take it ❤👴❤

[@dmagik8]

Love that Dawid reference. Great video as always.

[@mtgn59]

14:07 that headshot🤮🤮🔥🔥

[@srvuk]

Glad to see the AV1 rendering test, which on many a PC would take an interminable amount of time, so to see it match H265 and beat H264 was amazing, which should now signal an acceleration towards AV1 standardisation now that editors can actually output to it in a productive manner. All in all, when comparing the price to other offerings, as a general use PC with some oomph, this may do the job for those who can spend the money. Although we should hope for more driver optimisation from the Intel Team, you might have thought that they would have tested gaming and seen the results themselves before release. I hope that those 2 LP cores can result in very low energy use when running processes that don't need to use wasted power. Taking hat top off was hardly that much of a hassle. I've taken one off another of this companies PC's and you just have to get in there and rip it up.

[@TomiWebPro]

That intro framerate's like my gaming frame rate. 🥲

[@Karti200]

Hey, silly question - could you try to run this iGPU on latest CoD duo its having XeSS support? could be nice to see how much you can get out of that even in bigger titles if you are using proper XeSS with proper hardware for it

[@krissayswhy]

for the Fortnite benchmark around 14:26 im willing to bet the issue is that Performance mode is still DirectX 11, an API version that the Arc Alchemist architecture infamously is not great at. testing the game with the API set to DX12 at the lowest settings (save for view distance) could have possibly reduced the stutter down to a more normal amount, at the cost of performance obviously since all low DX12 isnt comparable to Performance mode. just something i wanted to note, otherwise it was a great video!

[@tomlolbob3182]

Also Performance Mode is known for being inconsistent.

[@edgartheface]

I wonder how all of you are happy with this P-/E-Core stuff … recently got a laptop from work with Intel CPU, loaded up my simulation software and was amazed how quiet and snappy it ran while I had the simulation running in the background. Came back after an hour … guess what? Nearly no progress! It moved the background task onto the E-Cores and I couldn’t stop that behaviour. Asked IT-support for a Ryzen 9 laptop (first time they heard it) and no problems anymore since then.

[@wingcommanderbob8268]

just set core affinity in task manager?

[@CatSovietski]

​@@wingcommanderbob8268Every single time? 🤔

[@ranjitmandal1612]

👏

[@GeorgePerakis]

@@wingcommanderbob8268 I see the Intel stock holders are here.

[@abctips3034]

you can try out processlasso ?

[@piecaruso97]

the low power ecores are there to run the system while it's idle, so that the other cores can be powered off saving battery on laptops and reducing power usage and heat

[@faei1897]

Seeing the nipogi am0 p2 I have as my portable workstation appear on the charts makes me smile

[@MuhammadHammadAshraf]

Loved the bit about timeline scrubbing 11:50. Video editing with H265 can be horrendous at times so I'm looking to work with native AV1 support on both CPU and GPU. Thanks!

[@paul9812-t3i]

Iceberg Tech is my comfort content ❤

[@AlexRivera-jd8fi]

Dawid mention goes crazy, ya love to hear it

[@DigitalJedi]

Also of note here, the 125H only has 7/8 Xe Cores enabled on the iGPU. The fully-enabled iGPU comes in chips like the Core Ultra 5 135H and above, and should be 8-12% faster.

[@OneTwoWolf]

9:45, correct me if I'm wrong, but i don't believe Arc has to emulate DX11? Intel's performance in DX11 is not as good as DX12 for sure, but I think only DX9 needed to be emulated initially (though I think they might've since moved past that to native support for DX9 too).

[@lightspeed1113]

You’re exactly right

[@IcebergTech]

Perhaps my information on this is wrong? I was under the impression that Arc only natively supported DX12 and Vulkan, and that everything else required emulation.

[@lightspeed1113]

@@IcebergTech I asked some people in the intel insiders discord and they corroborated the Dx11 driver is native. They pointed to the Q3 2023 Dx11 update video. There is a whitelist Dx11 driver for games they have optimized and there is a legacy general purpose Dx11 driver for titles they haven’t gotten to which is what they were using at launch and doesn’t perform as well.

[@lightspeed1113]

DX9 still isn’t native and won’t ever be but they switched from using d3d9on12 at launch, to using DXVK baked into the drivers

[@matte_bar]

14:50 the stuttering is a problem after latest update, even on an higher end system (in my case 12700k and 4070)

[@lamhkak47]

Least the Arc iGPU nominally catches up with Team Red's performance (kinda)

[@moevor]

Very solid review. Thinking about a mini-PC for work, but my most CPU heavy task is running MATLAB. Other scenarios are having like 40 windows of chrome, word, excel, PowerPoint, and outlook open at the same time.

[@solarrxenon]

that intro made me troubleshoot my internet… dang it Iceberg!!! 😢

[@ChaiSuBin]

I am early! Nice. Love your videos and reviews! Keep at it :)

[@ABaumstumpf]

You mentioned the Ryzen 7940HS in terms of powerdraw and performance - how do they compare for the specific programs/games? Cause most reviews just mention peak powerdraw or only CPU or only GPU power in specific instance. Here it seems the 125H was drawing around 30-40W - which is a bit lower than most figures i found for the 7940HS, but also not that different.

[@DigitalJedi]

On the process node thing. Intel4 is most comparable to TSMC 5nm or 6nm in terms of the V/F response, power per bit, and transistor density. Feature sizes may be closer to 7nm, but they are packed closer together. You can also see this in the power efficiency comparisons, as the CPU is similar to other 5nm chips like the R7 7700X.

[@denvera1g1]

While DDR5 is dual channel, those two sub-channels together are the same 64bit wide as DDR4. Now what this means is that you can use something called memory interleaving, which is not at all like how DDR4 dual rank made things faster, but to similar effect. This means those business computers that come with only one stick of RAM can still feel a bit more responsive having two alternating 32 bit channels as opposed to a single 64 bit channel. Effectively the bandwidth is the same, but certain types of latencies are reduced(not all types)

[@tuff_lover]

Can we just go back to faster L2 cache?

[@DigitalJedi]

What do you mean? The L2 cache of Meteor Lake is almost the exact same as with Raptor Lake, which is both faster and larger than that on Alder Lake which has a similar relationship with Rocket Lake.

[@tuff_lover]

@@DigitalJedi Bro, Socket 775 had up to 12Mb of L2 cache.

[@DigitalJedi]

@@tuff_lover Meteor Lake has that on just the 6 P-cores, and it has faster access to it.

[@lordfenix17]

Nice, not too late this time.

[@lubossoltes321]

So the 7840 based K6 with 16GB ram and 1TB SSD costs the same as the barebone K9 while being faster overall ? I guess if you absolutely need the NPU or the AV1 acceleration you could justify the K9, however for normal use I don't see a point here ...

[@georgyzhukov4269]

Letsgooo

[@dex6316]

I think calling Meteor Lake 7nm is misleading. Ever since the transition to FinFET nodes, process naming has been really arbitrary. The Intel 4 node in Meteor Lake did indeed used to be called 7nm by Intel. But the Intel 7 node (formerly 10nm+++) is comparable in density to the TSMC N7 node and Samsung 7LPP node, and the Intel 4 node’s HP libraries are theoretically denser than the TSMC N4 HP libraries (Intel 4 is an incomplete node and doesn’t have HD libraries). Angstronomics has a good article on node densities if you wish for further reading.

[@GeorgePerakis]

Yes we get it you bought stock in Intel and you need to find a way to cope. Density isn't the only way we measure node performance and you know this.

[@rj7250a]

Density is the most important aspect, moore's law was measuring transistor density. More density means more IPC, because you can put a wider decoder, larger branch predictor, larger micro-ops cache, more ALUs and AGUs, larger register file. It is even more important in mobile chips, because you can not just crank up the clock speed.

[@badgermemelord6207]

@@GeorgePerakis Imagine being this butthurt by someone pointing out facts

[@DigitalJedi]

@@GeorgePerakis I work in lithography (ASML, then Samsung, now Intel 14A team) and while it is true density is not the only measure of node performance, it is quite often the most impactful to the performance of chips like this. Intel4 is comparable in density to N5, and has a V/F response somewhere between N4 and N6 depending on where in the curve you measure.

[@computerscience1101]

​@@badgermemelord6207 there is no 10nm+++. The main comment is misleading

[@Crustenscharbap]

I would be interested how was is the R9 7940HS at Davinci. It also can encode AV1.

[@saturnotaku]

Kind of disappointing the HDMI port is 2.0 instead of 2.1. Unless there was some significant cost saving, I'm not sure why it would be limited like that.

[@saricubra2867]

HDMI is way worse than display port anyways.

[@joseph3164]

I’m very glad you included apex legends into your test suite of games!!

[@gyrz19]

Needs another USBC instead of that second network connector. Also, no idea if you have a external doc, but I would love to see a video on how well an external usb 4 connection GPU performs with one these.

[@Deathgazer]

Would it be possible to swap that top fan with say, a 40mm 5V Noctua 4pin, or any others for that matter?

[@jurisjancevskis9076]

what is up with the intel naming scheme

[@GeorgePerakis]

The people in charge of product naming are part of Intel's legion of contractors I'm afraid.

[@lharsay]

Intel killed all their Celeron, Pentium and Core i branding. Now they have Intel Processor (sic!), Core and Core Ultra.

[@saricubra2867]

Trying to copy Apple Sillicon products... (those make more sense).

[@jurisjancevskis9076]

@@lharsay celeron & pentium was already incredibly stupid, killing iX branding is even dumber

[@PeteKay]

before I heard you reference D I was thinking how similar your delivery is.

[@AndersHass]

NPUs have been in mobile SoCs for a while (all the image processing they are doing), which is probably why many mobile systems have been tested in the machine learning benchmark. Only recently it has started to be added to Windows PCs with all the AI hype and copilot. Though their actual usage is fairly limited so NPUs probably only matter with future generations compared to the ones released now. Like the new snapdragon that is going to be released soon sounds like it will be first one that can run copilot locally, where future Intel and AMD will also be powerful enough as well.

[@IdeasAreBulletproof]

Can't wait for the Core Ultra to be added to the Framework Laptop

[@Bilal.Hussain.Shaikh]

Why can't Intel use 8 performance cores and 6 efficiency cores and 2 lower efficiency cores

[@saricubra2867]

Same reason why my i7-12700K (or the 12700 family in general) only has 8 perfomance cores and 4 e-cores. They just don't care. Apple with Apple sillicon also made a "mistake" with the M2 Pro (8 perfomance cores and 4 efficient cores).

[@DigitalJedi]

E-cores are 1/4 the size of the P-cores and come in groups of 4 due to the shared L2 cache. You can have any number of P-cores, and multiple of 4 E-cores, and any multiple of 2 LPE-cores. In this chip there are a total of 6 P-cores and 8 E-cores on the CPU tile (2 P-cores are disabled). 8 P-cores would consume the entire tile, and alternatively going to more E-cores could allows for a 4+16 setup or a 5+12. LPE-cores are on the SoC tile and there is only room for 2 on that die as they have to share space with all of the system agent stuff like the memory and PCIE controllers. There is a tradeoff that needs to be made then. P-cores provide less multi-core performance than using the same space for E-cores, but provide a single much faster thread each (along with a slower hyperthread). In the case of Meteor Lake, it was determined that most applications people will run on it will either need fewer than 6 (or in this case 4) very fast threads, or will scale well to a larger number of threads, 22 maximum, or in this case 18.

[@truesightgrabber]

Is it possible to run in 24/7 as Server ? How long it will survive ?

[@ranjitmandal1612]

Great video

[@minunat9]

Hey buddy, what capture card are you using?

[@SteelTumbleweed]

Apex is worth retesting with dx12 command line.

[@ItsDeeno69]

why are u comparing the ultra 5 to ryzen 9 why not ultra 9 and also why didnt u tested the npu in geekbench ml

[@HowManySmall]

sad you didn't test fh5 with xess

[@badgermemelord6207]

Also worth pointing out that this is just an i5 125h, which has the cut down 96eu igpu, the 155h and up have an even more powerful 128eu igpu

[@DigitalJedi]

Not 96Eu. There are 112 here as the 125H has 7 of 8 Xe Cores enabled. Still though, those larger GPUs should be 8-12% faster.

[@mindinversions4487]

18 threads = 9 cores? Slightly confused.[Ok, I've worked it out in my head that there are 4 pcores with hyper threading, everything else is non hyp[er threaded, and presumably Intel is just adding all the numbers together. Kinda surprised there aren't more X's involved...

[@IcebergTech]

Easy formula: if you know the number of total cores and threads, subtract the cores from the threads, and what remains is the number of P-cores. So, 14 cores 18 threads = 4 P-cores. Or look up the chip on ark.intel.com, whichever works 😁

[@DigitalJedi]

E-cores don't have hyperthreading. In this case there are 4 P-cores (8 threads) and 10 total E-cores (10 threads) for a total of 18 threads.

[@theradcat11]

To fix stutters in fortnite try to use DX12 instead of Perf. Mode

[@Teh-Penguin]

It still seems crazy to me that you can legitimately game on a mini pc.

[@appleapple3855]

Noice, but you have to use Vegeta's reaction for that

[@anas_voice_over]

DAWID MENTIONED AYEEEE

[@aghostyboi3075]

I just can’t get over the fact intel decided to make it all the way to 12 and say “nah we’ll just skip 13”

[@kaydenmadency6448]

i want a iceberg and dawid collab

[@Razzbow]

I hope Kraken has competition.

[@yarost12]

LPE cores are there so they can park everything else and just keep the IO die powered, unlike AMD 2/3 chip design where at least 2 need to be powered at all times.

[@betag24cn]

intel is back? when did it left, sadly, it never left

[@WendelTC]

Doesn't Fortnite have a dx12 mode?

[@toma01001]

You have hands?

[@IcebergTech]

No, I kidnapped a sailor and pressed them into service as a hand model for me. Although I'm an iceberg, I'm still British. It's what we do.

[@moist_ointment]

Calling Intel 4 "7nm" is a huge pet peeve

[@hyperturbotechnomike]

I assume only Windows 11 is officially supported on this kind of triple core type contraption. I'll give it a pass, since i don't want this OS. Addendum: The manufacturer provides drivers through altered Windows ISO files and not standalone executables. And only provided via google drive. Scary.

[@Sunlight91]

Unless you need a specific professional application just use Linux.

[@hyperturbotechnomike]

@@Sunlight91 I use Windows 10 LTSC 2021 for my workstation and entertainment systems. Debian and SUSE for the servers.

[@Sunlight91]

Now Intel just needs to fix their pricing. Zen5 isn't far away.

[@pumkining]

Gtx 970 review? 👉👈

[@micahottaway8455]

Intel is finding out how much of a pain in the butt driver support for GPUs can be.

[@jabonorte]

Can't believe that, even after all these years, Intel drivers are constantly underdeveloped. It's almost as if they always want their graphics chips to fail. Had this been an AMD computer you would have given it a pass on issues that are clearly driver-related because you know it would get fixed in an update or two. Intel seem to think they have an Android product, where you only update the software once a year. Still looks like a nice machine, if gaming isn't your immediate need

[@Matt-oq4jq]

great video, but it really shows just how bad these chips are for intel: performance regression in every space except for gaming is a gigantic negative for everyone except for people who review mini pc FPS benchmarks, lol. there's a reason why OEMs are prepared to go 50/50 AMD/Intel for the first time ever after meteor lake. ESPECIALLY when these chips are slower in gaming than two-generation old AMD parts... what a nightmare.

[@06dpa]

Intel try to not make your naming scheme less understandable (impossible)

[@aavvironalex]

Test Quake Champions.

[@OneAngrehCat]

Strix Point is coming out along with Zen 5 in a month. Don't be toooooooooooooooooooooooooo hasty with "not giving Intel passes anymore". Actually, I hope your passbook is full of passes.

[@SpoonHurler]

"Non-consensual" 😂😂

[@PSTechCR]

The fortnite error is most dx12 graphics cards or the unreal 5, I've been having this issue as a lot of other players, and I have an Nvidia 3090

[@JesseO-nh3us]

Dont let the intel fanboys hear you say that intro.

[@betag24cn]

its ok, there are no intel fanboys left

[@saricubra2867]

I'm not impressed, the singlethread is almost my i7-12700K at 5GHz. Meteor Lake will be destroyed by Zen 5 mobile.

[@DigitalJedi]

Id' say it's pretty impressive to see a 28W TDP mobile chip catching a 125W TDP desktop chip. Of course neither is even drawing full TDP for single-threaded tests, but that's not a bad showing. Zen5 and Arrow Lake are both going to be huge jumps though. At this point I'd wait it out too.

[@xPandamon]

I'd still go with AMD, just cause they aren't as scummy as Intel.

[@danielfernandezaguirre]

intel is just wasting sand at this point, first e-cores and now even more useless lp cores? that's sad

[@DigitalJedi]

MTL isn't really meant for desktops. The E-cores do a lot for multi-core performance, more than a P-core would in that space (and so same amount of sand). LPE-cores are intended to allow the main CPU tile to power down while the system remains active on those 2 cores, helping to save battery when doing simple tasks. Video playback for example can happen without the main CPU cores being on continuously, which allows MTL to get a few more hours out of your battery than RPL did. In this case they traded about 3mm^2 of sand for longer battery life on a chip over 130mm^2 total. Oh no!

[@buff9267]

stupid new naming scheme

[@SuperStareGry]

AMD variant is not only cheaper but better. Why would you buy this i do not know.

[@tendosingh5682]

Absolutely no reason to get this over AMD.

[@ZERARCHIVE2023]

I'm team red

[@K3vyB]

Back to being a laughing stock, sure (:

[@charlesg5085]

Its crazy to see intel being slammed so hard by amd. They are losing across the board. They have had too many bad ceos. 4 core 4 thread cpus for nearly a decade stagnated the company. Sad...

[@ruwn561]

Those CPU are rubbish for virtualisation. it’s a hard pass Amd still kicking intel.

[@sv1595]

too early

[@jimlee3725]

Intel is gone.

[@Khloya69]

It’s pronounced CPU zee.

[@tyler6602]

Who cares

[@DigitalJedi]

Zee and Zed are both valid pronunciations of the character 'Z' or 'z'. It largely depends on where you are from. You will often hear Americans say Zee, while Canadians and the English will more often use Zed.

[@Khloya69]

@@DigitalJedi but zed isn’t valid when it’s used name of a brand/app/really anything. That’s always Zee, zed just doesn’t sound cool and defeats the point.

[@DigitalJedi]

@@Khloya69 The name is the name of the 4 characters 'C' 'P' 'U' and 'Z,' making both valid.

[@Khloya69]

@@DigitalJedi wrong.

[@windowsnt63official]

This is the worst intro of my life. It’s just false advertising for AMD.

[@itsatrap4986]

intel is toasted aka cooking. i9s are worse than Ryzen 7 7700X.