Intel Working on Fixing "Arrow Lake" Gaming Performance with Upcoming Patches

[3DVCash]

Its application performance is great, and gaming performance on par with Raptor Lake. (And I'm of course ignoring tests at 720p/1080p with RTX 4090, as testing unrealistic workloads are pointless.)

Agreed with everything up until this bit. Why is this still a point of contention? Low res CPU benchmarks are an easily controlled testing environment for CPU performance. Nothing more, nothing less.

And while they may not be a big deal for single player games at higher resolutions, those performance deltas absolutely can matter for online play, particularly MMOs.

 

[Deorc Mearh]

As I said in other forums - it should be "fora" ... by the way... - a few days ago I believe these cpu's are underperforming compared to their potential due to the non optimization of SO in deciding how to assign threads to cores. So far Intel had hyperthreading, this has impact on how many threads at the time a program should use and to which cores they should be assigned. I'm not sure the improvements in performance will be huge but quite sure they will be noticeable.
Anyway this new architecture works similarly to mobile processors, trying to complete tasks as fast as it can before rising power consuption too much. Sure if pushed to its limit it consumes a lot but in normal processing it's meant to be able to complete its job consuming less.
Many childish salty comments reminds me the smartass kids that - when nvidia changed its traditiona gpu architecture and designed Fermi architecture for its 400 series but TSMC didn't reduce in time its cmos size so the gpu had a lot of problems of overheating... - claimed nvidia was dead, it was about to fail, nvidia wasn't able to design gpu anymore and such idiotic and arrogant statements.
I think we know how it ended up.
Not sure Intel is in the same promising position but surely they are changing many things at technological level and at corporate structure level and they are improving overall their business. I'm pretty sure this last architecture will improve enough its performance in the very next months.

 

[JohH]

Nice words but I'll wait to see results before changing my mind on Arrow Lake

 

[DemonicRyzen666]

As long as the games are distributed to the filler tile "cores", which are not overclockable or have L3 cache, the performance won't improve much. But the temps are fine now...

Which cores are you referring toward?

4 E-cores share an L3 cache slice
1 P core uses its own L3 cache slice.

Oddly enough AMD's CCD's are slightly thicker than on Zen 5 than they were on Zen 4.
Intel has thinned the tile dies compared to the older monolithic dies because of the added layer of [EMBI] silicon interposer which is "Foveros".

The only problem I see currently with arrow lake is the L3 cache slices should all be once peice & connect to each P-core. Old Zen 2 architecture used two 16Mb L3 caches slice to split up a quadrant of cores or "CCX's" inside of the CCD's which resulted in a performance loss.

 

[Deorc Mearh]

Nice words but I'll wait to see results before changing my mind on Arrow Lake

reasonable. I simply suggest to be more prudent on sentencing over Intel's future.

 

[chrcoluk]

Too late for the reputation, this should be done before rolling out a product. You would think with RL issues, Intel would have learnt something.

 

[Dr. Dro]

I disagree with all that nonsense, their first mistake was implementation hybrid processors and the use of two different architectures, the second mistake was removing support for the AVX512 which hybrid architectures does not allow such feature because of different ISA across cores, of course, Intel lately renamed AVX512 to AVX10.2 which is already added in GCC compiler.
One more thing to add on list of shame is Intel’s mismanagement and lack of vision, I say that because they recently announced another revision of upcoming x86s which is a good thing but who knows when will it see the light, now they must move on and through away legacy which keeps them on edge.

1. Hybrid is not a problem. E-cores are fine. Raptor Lake did away with the ring domain ties, Meteor Lake introduced all-new types of E-cores and the new Arrow Lake chips are a completely different beast altogther.
2. AVX-512 is entirely possible on client segment, regardless of E-cores. Intel chose to disable it to simplify platform validation and for market segmentation purposes. Remember, Intel is very fond of their SKU spam and segmentation, while AMD just unlocked everything and enabled ECC regardless of model you buy. They clearly feel that the consumer-grade segment doesn't need it.
3. I do not think Intel lacks vision. They, however, lack execution and they keep screwing up time and time again. The "14th" re-heat fiasco, low-level firmware bugs, delays in their graphics products, etc. - things aren't looking so hot.

So they will get rid of rails and push the CPUs to limit throwing "efficiency" out of window, on what already is a power hungry CPU falling short of competition. A polished turd is still a turd.

Well let's be fair. The 9800X3D, while still manageable, was the one guilty of that. It's a pretty bad regression vs. the 7800X3D in this area.

 

[Darmok N Jalad]

Not to mention there was huge regression in the last 6 months for all hybrid designs. The 14900K went from just behind the 7800x3d to pretty far down on the charts in newer reviews.

Something is going on with microsoft/intel and the thread scheduling.

Well, just look at how much change has been thrown at the scheduler. From all P cores, some with HT, some not, to P+E with HT, dual CCDs, dual CCDs with one with extra L3, P cores and P cores with less L3, and now P cores with no HT with E cores. Windows went from a relatively simple scheduler to one that has to mind all these combinations of things. It's a wonder threads end up in the right place at all.

 

[chrcoluk]

Watched a bit of the video, an additional comment, seems pretty whacked, that the Intel rep keeps referring to reviewers as the community, and that its their feedback that is shaping their products and they take onboard for any fixes. Its as if consumers are invisible or something. A lot of it around the middle of the video.

Never been happy with how close reviewers are to these companies, and the extra loops consumers have to jump through.

But anyway it seems we have a few weeks and by then we should have something released, then I guess new testing and feedback from people.

Well, just look at how much change has been thrown at the scheduler. From all P cores, some with HT, some not, to P+E with HT, dual CCDs, dual CCDs with one with extra L3, P cores and P cores with less L3, and now P cores with no HT with E cores. Windows went from a relatively simple scheduler to one that has to mind all these combinations of things. It's a wonder threads end up in the right place at all.

You should see Android.

There is a different scheduler configuration across half a dozen app states.

 

[watzupken]

People act like x86 is some kind of amp slurping monolith. It isnt. Lunar lake showed the whole "x86 cant be efficient" line was pure BS.

OTOH, removing legacy support IS one thing that will kill off X86, without legacy x86 has little future.


He's in marketing, one step removed from absolute scum like used car salesman. No respect is deserved for anyone in that particular profession.

No.

Lunar Lake definitely is an improvement from Intel’s recent CPUs, but it comes at a significant cost to Intel, (1) opting for cutting edge TSMC 3nm, and, (2) integrating memory on chip that likely reduces latency. Compared to Intel’s own Intel 10 and 7, I believe TSMC’s node is more power efficient since the foundry has always been making low power ARM based SOCs primarily. The flip side however is that they had to sacrifice performance because of the hard TDP cap and in multi-threaded performance because you can tell that even Qualcomm’s first proper attempt at Windows based SOC can outrun the top end Lunar Lake in some workload which may be running with an emulation layer. The sad reality is, Lunar Lake is good, but Intel is not going to make a lot money selling it. The lost of the 40% discount from TSMC and the integrated memory likely eroded most of the profit margin.

 

[Taraquin]

Good that they are working on it, but I see no reason why you should pick it over Ryzen 9000 which is overall cheaper due to being able to run fine on a 120usd B650 mb with cheap 6000c30 ram which can be tuned nicely, while you need a 250+ mb and 7000+ ram to make Arrow lake comparable.

 

[Ruru]

Not an emergency edition KS SKU yet?

 

[TumbleGeorge]

The right way: Get HyperThreading back to work!

 

[SPARSTE96]

There must be something wrong with arrow lake, i knew it. but i don't think any patch can make them have a better performance than 14th gen....

 

[JustBenching]

Well, just look at how much change has been thrown at the scheduler. From all P cores, some with HT, some not, to P+E with HT, dual CCDs, dual CCDs with one with extra L3, P cores and P cores with less L3, and now P cores with no HT with E cores. Windows went from a relatively simple scheduler to one that has to mind all these combinations of things. It's a wonder threads end up in the right place at all.

Can't it handle specific optimizations based on CPU being used? Cause 22h2 worked great on my 12900k, 24h2 is a massacre. Not in every game to be fair, but a few games here and there show a big difference. Yesterday I tested a clean install of win10 vs win 11 24h2, look at the difference. And this isn't even the biggest offender, R&C is.

Just watch the first 5 seconds and it's enough, there is a 10+% difference between the 2.

Win 11

Spoiler

Win 10

Spoiler

 

[EatingDirt]

Not to mention there was huge regression in the last 6 months for all hybrid designs. The 14900K went from just behind the 7800x3d to pretty far down on the charts in newer reviews.

Something is going on with microsoft/intel and the thread scheduling.

The 14900k has seen no regression, it's just that the AMD CPU's have finally gotten some love from Microsoft's scheduler in 24H2, and because of that AMD CPU's have seen an improvement. There's now 3 X3D CPU's that are faster in it in gaming (7800, 7950 & 9800), and the standard 9xxx series occasionally gets some wins as well. There's nothing 'going on' with thread scheduling of the 14900k.

 

[N/A]

The right way: Get HyperThreading back to work!

Noo, relocate the memory controller back where it belongs and make it a 10 core, and a separate variant 10P+20E. 8/16T equals a 10 core in Mp score but much more code friendly and shorter pipeline, not coming back. Anyway intel has a 50% gap to 9900 X3 not just 5% to 14900K. Taking the 24h2 fiasco into account that's 100% i guess.

 

[Bwaze]

“…multifactor issues at the OS level”

Remember, reviewers have given Intel plenty of fighting chance by benchmarking exclusively in Windows 11 23H2 - because in the version that is now rolling out, 24H2, AMD gains additional performance across several generations, while Intels “Arrow Lake” just appears to be broken. So even when benchmarking newly released AMD Ryzen 7 9800X3D they benchmarked it in 23H2, citing apparent bugs in 24H2 - but not explicitly saying the bugs are mainly instability and lower performance in new Intel CPUs.

Isn’t that a bit of a preferential treatment of a company that still holds 70% of market share in desktop CPUs, even when releasing measurably crappier product? Should they be given such treatment, especially given their history of unlawful practices of achieving and maintaining this market dominance?

So Intel has a lot of work to do to even reach a crappy last gen performance level in new Windows version - because sooner or later everyone will get that update, in the end even the reviewers.

 

[dj-electric]

“…multifactor issues at the OS level”

Remember, reviewers have given Intel plenty of fighting chance by benchmarking exclusively in Windows 11 23H2 - because in the version that is now rolling out, 24H2, AMD gains additional performance across several generations, while Intels “Arrow Lake” just appears to be broken. So even when benchmarking newly released AMD Ryzen 7 9800X3D they benchmarked it in 23H2, citing apparent bugs in 24H2 - but not explicitly saying the bugs are mainly instability and lower performance in new Intel CPUs.

Isn’t that a bit of a preferential treatment of a company that still holds 70% of market share in desktop CPUs, even when releasing measurably crappier product? Should they be given such treatment, especially given their history of unlawful practices of achieving and maintaining this market dominance?

So Intel has a lot of work to do to even reach a crappy last gen performance level in new Windows version - because sooner or later everyone will get that update, in the end even the reviewers.

"reviewers" means? The majority bulk of 9800X3D reviews I have seen out there were actually on 24H2, as people were given about a 3 week window to prepare test systems for its arrival.

There's no right or wrong here. I agree with a review using 23H2 and agree with one using 24H2. Those are different perspectives and use cases users can extrapolate information out of suiting their own kind of use behavior.

 

[_roman_]

I can't go into all the details yet, but we identified a series of multifactor issues at the OS level, at the BIOS level, and I will say that the performance we saw in reviews is not what we expected and not what we intended. The launch just didn't go as planned. That has been a humbling lesson for all of us, inspiring a fairly large response internally to get to the bottom of what happened and to fix it.

So Intel admits the intel processor is windows only hardware? Where are the optimisations for other operating systems and software?

I buy generic hardware which run on current, past and future software. Not a windows 11 only processor.

They had real gains. Ecores is a boondoggle

My software will crash on "fake" E-Cores. I called them early "Fake" Cores.

In my point of view those E-Cores belong to a plugin card like a sound-card, as they are a different architecture in the first place. Like an accelerator card.

-- When you only run Windows 11 it will not really matter if the architecture is fully backward compatible. It will not really matter which instructions are supported by the processor on how many cores. The user will most likely not bother with compilers.

Other than that I consider it a great product.

I do not think so.

First the high power consumption. The one or two processor generartion a new mainboard flaw.
Since Intel 12th generation it's a mixed core processor flaw also.

The hole E-Cores makes software optimising a nonsense game. I do not want to optimise for a intel 80486 because of some E-cores (I hope you get my point 5th gen? 6th generation intel i core generation. // for the younger generation intel 80486 is a very, very old processor from around 1990)

I do not see any decent software support for those E-cores in linux. Linux is only the kernel. I do read all those new options I'm offered with every new kernel build. (make oldconfig) The other parts from gnu linux is from different projects. When I look how many years that 12th generation intel processor is already available with that e-cores feature, it's very bad software wise.

I doubt those BSD and other choices have any optimisations for those e-cores at all.

-- The lack of AVX512 makes it a no buy for myself anyway. I need that instruction.

 

[JustBenching]

Most people don't want to pay more for identical performance, with a 3W power difference being negligible, not "much lower power". But in all-core, the 9950x uses 15W less power (also not a huge deal) so I don't see the reasoning?

If you are talking about gaming, it's not 3w power difference. Tom's hardware has the 9950x consuming 40% more on their 12 game average. It also draws a lot less power (while being a lot faster) in mixed workloads like autocad. Too bored to spam you with links and graphs but if you want me to I will.

I do not think so.

First the high power consumption. The one or two processor generartion a new mainboard flaw.
Since Intel 12th generation it's a mixed core processor flaw also.

The hole E-Cores makes software optimising a nonsense game. I do not want to optimise for a intel 80486 because of some E-cores (I hope you get my point 5th gen? 6th generation intel i core generation. // for the younger generation intel 80486 is a very, very old processor from around 1990)

I do not see any decent software support for those E-cores in linux. Linux is only the kernel. I do read all those new options I'm offered with every new kernel build. (make oldconfig) The other parts from gnu linux is from different projects. When I look how many years that 12th generation intel processor is already available with that e-cores feature, it's very bad software wise.

I doubt those BSD and other choices have any optimisations for those e-cores at all.

Huh, you seem pretty unbiased. Ok bud.

 

[Rahmat Sofyan]

new socket, new CPU, new chipset, and same old problem... welcome to "intel inside" and goodluck with degradation feature problem

 

[SL2]

I don't care about Arrow but I'm curious how or if they can improve performance.

Kind of a clueless statement from Hallock tho. Intel's not able to run a few benchmarks on their own? They just do the launch and wait for reviews, hoping for the best? It doesn't add up lol

 

[dyonoctis]

He's in marketing, one step removed from absolute scum like used car salesman. No respect is deserved for anyone in that particular profession.

If you think like that, you have a very superficial understanding of marketing. You know when people say, "It's a product looking for a problem that doesn't exist," before fading away in irrelevance? In some markets, that is what happens when marketing does a poor job because marketing isn't just about selling a product; it's also about figuring out what kind of product people want.

Marketing also decides where and how a product will be sold. This also explains why a company doesn't sell some of its products in a specific country: The US best-selling Ford F-150 is sold in limited quantity in select countries in Europe because trucks are just not that hot in the EU compared to the US. Small SUV/Hatchback is what is hot in the EU. Marketers are the people figuring that stuff out, because an engineer (it seems like you would rather have an engineer handling all that stuff) doesn't have the time, or the will to deal with that kind of stuff.

Price slashes, bundles, and special offers are also marketing

 

[Assimilator]

He didn't design it, he didn't build it, and he didn't do QC for it.

This is possibly the most stupid non-argument I've yet heard on these forums, and I've heard a few.

His job is to sell a product and say whatever he can to achieve that goal. You're shitting on a man for doing his job as expected, to the letter.

Saying "I don't know about the product I'm trying to sell" is doing a shit job.

Be better.

Be an adult.

trucks are just not that hot in the EU compared to the US

Primarily because men in the EU are secure in the size of their manhoods.

Marketers are the people figuring that stuff out, because an engineer (it seems like you would rather have an engineer handling all that stuff) doesn't have the time, or the will to deal with that kind of stuff.

Uh no, market researchers are who do that job. Research is not marketing.