I tried gaming on Intel’s new CPUs without E-cores. It just made things worse.

[@cipriankhlud703]

Maybe it sounds silly, but the new "E-Cores" are quite fast. So maybe disabling the P-Cores would make a very interesting benchmark, given that there are 16 E-cores, so it would match more or less many games that may expect 16 threads. It should use low power.

[@Silicon-Insights]

Someone else suggested doing a low P-core count test so I think it might be interesting to test this when we have time. Thanks for the suggestion!

[@cipriankhlud703]

​@@Silicon-Insights 245K has a low(ish) P-Core count of just 6 if I got it right. So my thinking is simply that 16 cores are according to Intel's presentation "in the same ballpark" with Raptor Cove (yes, they have lower cache per core, but compute should be IMHO similar). Not sure if you are aware, but these claims are out of their (aka Intel's) presentation of how awesome are their new Atoms: (you can listen few minutes ahead if you have time) kzbin.info/www/bejne/n3mcpqGwbLSNqdEsi=pw7O3ENChQvGETFe&t=1673 I guess that these cores will be hampered by low L2 sizes and they will be bound of L3 cache (so the U5 should behave worse because not only of fewer E-Cores but high traffic between cores and smaller L3). Later: fixed timestamp in the link

[@pvdgucht]

I really hope we see a second generation on the Z890 LGA1851 platform. Would love to see same 8P 16E design but with improved architecture of the e and P cores, some decent increase in cache maybe little faster interconnect of the tiles, battlemage architecture for the integrated GPU and maybe a few more Tops for the NPU. That would probably make it much more interesting and the higher caches would probably increase performance by a lot in gaming and make the CPU even more efficient.

[@Silicon-Insights]

That would be really nice especially since there's lots of potential in Arrow Lake that wasn't fully realized, but I wouldn't bet on it.

[@pvdgucht]

@@Silicon-Insights Yeah 🥲 you’re probably right that sounds too good to be true from intel. 😆

[@impuls60]

They put the memory controller on another tile that the compute tile. MASSIVE FAIL. It has huge ram latency issues and even with lightning fast ram this inbuildt latency cant be fixed. Not with oc or fast ram. Its just a DUD. They could have used more cache like Amd does but they didnt. Most smart ppl matched an 14900k with 7000-8000Mhz ram and Intels own MLCC test app shows really low latency upto 8p cores. Turning on only 2e measured latency went up like 80%. The internet forums have known about this issue for years. Ofc the industry itself have known about this many years before the X3d parts existed otherwise they wouldnt have made such an effort to create it. Intel just dont care about gamers any more, atleast the current management. Ofc HT and other decisions like why only 8 cores? Your not excactly leading the industry with that number.

[@AntonYadrov777]

i9 is literally just a scam with E-cores, no difference to P-cores or imc whatsoever. i7 is superior in price/performance ratio because of that.

[@DragoMorke]

I think to have a better idea and camparability you should have tested 14rh gen without e-cores as well.

[@Silicon-Insights]

The focus of the video was Arrow Lake specifically and the whole CPU scheduling thing, so anything with Raptor Lake would have been outside the scope.

[@egnappahz]

Glad I found this little dark corner. Welcome to the intel cope zone.

[@Silicon-Insights]

For what it's worth, we recommended against buying both the 285K and especially the 245K. They're not good products.

[@solidsnake5051]

@@Silicon-Insights I love how Intel fans used to make fun of AMD for adding more cores, but this time it's intel doing it :D

[@eliadbu]

​@@solidsnake5051this time? Max core count hasn't changed since 13th gen. These CPUs perform inconsistent and overall regress in some aspects, no idea what they exactly done to cause this, but it ain't good.

[@RobBCactive]

This is a sign of problems in design and/or hardware implementation and/or software. Previously memory stalls could be mitigated by HT without context switches, now that high power fast P-core is nerfed. The efficiency argument falls flat when you remember Intel RPL used 10nm rebranded as Intel 7, while ARL skipped 5nm/4nm in favour of TSMC 3nm. It's supposed to be better but simply isn't as a detailed look at gaming reviews where code/data isn't fitting in L1 like in productivity benchmarks. Reviewers are incorrect saying Intel tiles is new, not only did they sell Meteor Lake but Sapphire Rapids and Refresh, with the Xeon 6 launching to server. That ARL looks and worked like an experiment is a sign of systemic design and execution problems, following on the QA shoddiness revealed by RPL BIOS power validation failures and the chips blown by over-volting even on conservative server boards used for Minecraft and other games, which block on memory triggering a condition only exacerbated by creative settings on retail boards. Again reviewers ignore Intel's track record on security, which caused the need for expensive mitigations in software. Bending over backwards to ignore the fact that ARL was supposed to deliver general uplift through IPC and be more efficient with a process advantage, while Zen5 recycled an IOD and chose 4nm to keep costs down, while Turin dense launched on TSMC 3nm performance process within weeks means Lunar Lake is Intel's leading consumer CPU. But Hawk Point is cheap and good, with Strix Point giving more cores, so LL has competition even if the ARM hype lost its shine when reviews mean the Snapdragon became the ex-elite as Qualcomm's story met 3rd party performance reviews. "Nice screen and body", on this Asus laptop shouldn't fool people into ignoring problems.

[@Silicon-Insights]

I would quibble on the tiles thing, sure Meteor Lake was first and Arrow Lake does reuse the SoC tile (and I believe also the GPU tile), but desktop is a very different beast from laptop. And Sapphire Rapids wasn't really a true implementation of tiles, it's like saying Zen 1 was an implementation of chiplets. So as far as the desktop is concerned, this is the first time Intel has ever used tiles. I mean yea Arrow Lake has problems, but it wasn't really just born out of incompetency. There are lots of reasons why Arrow Lake just isn't where it needs to be, and lots of it is Intel's fault, but another large part of it is Intel trying to pivot to make it right. Didn't work out very well though.

[@hrayz]

What video card was used? Did i miss you stating it?

[@Silicon-Insights]

Yea I can see in the script that I forgot to mention the GPU, we used Dell's RTX 4090. Basically the most stock version of the 4090 out there.

[@jordan-mn6yy]

Intel doing better in cpu heavy games that few play but at the cost of higher tdp and always a dead platform. The trade offs to benchmarking better in obscure titles = totally not worth it and at those high fps count nobody will notice anyway

[@abaj006]

It feels like they have optimized it for Cinebench, and nothing else.

[@AngryChineseWoman]

(off-topic) 8:49 Crazy how the stock 285k sucks at the witcher 3 at 1080p but takes the lead at 1440p

[@Mazdabater07]

I mean not really. No one buying one of these cpus is going to be gaming at 1080p. So I am totally fine if its 1080p performance sucks if it runs well at 1440/4k

[@pangoomis]

Hope the upcoming 9800X3D will mark first positive video from Silicon Insights :P

[@Silicon-Insights]

We wish we were getting one, but our AMD contact said CPUs were in short supply. They're not as generous as Intel.

[@R6ex]

@@Silicon-Insights 😭😭😭

[@Kingtoby1981]

Only for gaming. For video editing, workstation, and games, I might go with the AMD Ryzen 9950x

[@egnappahz]

@@Silicon-Insights Yeah our safest bet right now is on the CPUs that brick themselves.

[@AntonYadrov777]

@@Silicon-Insights bruh, there is a very distinct difference between "generous" and "desperate". Too bad corporate copium makes one unable to see that.

[@the12gaugeshotty]

Not sure where the idea that the new Core Ultra are bad for gaming. The fact that they no longer use hyperthreading and perform identical to their previous generation, at the same time drawing less power is a good thing. I'm not upgrading from my 13600K for a while but the new Core Ultra are great for gaming.

[@RobBCactive]

It's benchmarks where they fail to match early '22 CPUs nevermind Zen4 chips like 7700 in gaming, of course the 7800x3D crushes it. It still uses more power, it's just better than the awful RPL efficiency. So it's numbers, price and instability, which makes ARL a hard pass when ADL is available on offer and AMD have stable alternatives. Turning E-cores not helping is a problem not a feature that says E-cores are good, because migrating threads could be a reason for poor latency and performance in default configuration. Disabling HT might be an issue because without context switches threads mitigated memory stalls, so those large L2 caches aren't blown as often. AMD chips perform much better using SMT increasing power used by the core to do useful work, so even Ry5 5600x can tolerate 2 cores being disabled in some game tests with it slightly improving performance in some games. Linux handles both Zen5 & ARL better.

[@Silicon-Insights]

If you notice, we never said the performance was bad, but middling or disappointing or poorer than expected. That's because the performance is, objectively, sufficient and even good, but it's not pushing the boundaries on the whole, and at times Arrow Lake is slower than 14th Gen Raptor Lake. Usually when new CPUs come out we expect more gaming performance, even just a little, but Arrow Lake is almost Bulldozer-like and showed a regression in many titles.

[@Maxkraft19]

What does disabling the E-core actually do. I mean its doesn't magically rewire the chip. Is the scheduler still nocking on the door to the E-core and moving on once no one answers. Also Windows isn't doing dynamic simulations to figure out the most accurate core loading. If there was they would need to updated for news cpus. It is almost certainly using some preprogrammed pattern or logic. Microsoft would have to program the scheduler to have an optimized versions with E-cores disabled. Without this programing by Microsoft the os will push too many threads on the cpu. I want to be clear. I am not saying your are wrong. Just that we might have to dig more into what happens when e-cores are off to know.

[@Silicon-Insights]

They are completely disabled via the BIOS, Windows does not recognize them, it is the same as if they were fused off or nonfunctional. We didn't use an app like Process Lasso to tell Windows not to use the cores, they just weren't there at all. And truth be told, I don't know how exactly the Windows scheduler works, but it seems very unlikely to me that merely reordering the cores would cause performance issues. This is a very basic thing that I'm sure Intel thought of, they created Thread Director specifically for hybrid CPUs in the first place. With the E-cores disabled, an Arrow Lake CPU is basically just the same as a pre-hybrid chip.

[@Maxkraft19]

It isn’t simple. If it was AMD and Intel would not have made software to make windows work better.

[@Maxkraft19]

I guess I should be more specific. When the core is disabled it is not necessarily like it’s fused off. In fact there is no reason for us the believe that intel even optimized the chips micro code to run with certain cores disabled. Again I am not saying you’re wrong. Am saying that with out specifically knowing what the schedulers are doing, I am not sure can answer this question. There are just too many unknowns in the way.

[@Silicon-Insights]

I'm sure there are microcode level optimizations to be made but Intel has been doing hybrid for about five years now. But it's hard to think that these optimizations would elevate the P-core only results to being better than what you get with E-cores. That would be a crazy level of optimization for something that has nothing to do with the CPU scheduler. Not to mention the fact that the 13900K was already a top gaming CPU when it came out. The 285K's gaming problems are almost certainly to do with lower clock speeds and lack of optimization rather than some sort of issue associated with the hybrid architecture. We see little evidence to believe that.

[@commandostryker304]

having E-cores offloads the less intensive processing away from your P-cores so that it can be fully utilized by the game but E-core count should be always less than P-cores the 8 to 16 E-cores is unnecessary for gaming the 4 E-cores is more than enough

[@ryanvtec3885]

12700k is perfect

[@emnoir]

for some reason turning E cores in Bios really cuts off total cache speed of the cpu. atleast for my 13700k you should leave E cores, and disable them in task manager, Also if youre going to do more tests Turn off Hpet and turn off Virtualization in bios. enjoy 30%+ perfomance boost.

[@Silicon-Insights]

E-cores have their own cache so turning them off removes their cache. I'm pretty sure just disabling them in Task Manager would also have the same effect, because you need to use the cores to benefit from the cache.

[@laszlozsurka8991]

So is this the missing performance? You just have to get rid of E-cores?

[@abaj006]

That sux, so for productivity you have to enable e-cores, then when gaming have to re-boot and then disable e-cores.

[@Silicon-Insights]

You might want to watch the video because this is actually the opposite of the conclusion we came to.

[@egnappahz]

Yes, it's that easy! If you believe hard enough you can stick to buying the newly released succesful intel CPU

[@Silicon-Insights]

Or if you watched the video, you'd see that we said there's no quick and easy solution to Arrow Lake's disappointing gaming performance.

[@thedesk954]

Please do 1 P core 16 E core 2 P core 8 E core 1 P core 7 E core Benchmarks

[@jibbawatt1209]

You are only showing frames per second as a benchmark. If you run any type of latency tester, having the e-cores on will add almost a full millisecond of system latency. In my case is was about 1.3 ms to system latency. Games just feel better with the e-cores off.

[@Silicon-Insights]

That could be true but the main focus of the video was on frametimes/framerates anyways. Looking at latency could be an interesting thing to do in the future, we plan on coming back to Arrow Lake next month when the performance patches are finalized.

[@joaovitorsilvagohl682]

Actually do the oppsite to get better performance