Intel "Arrow Lake-S" Engineering Sample Posts Over 25% 1T Perf Gain Over i9-13900K, Falls Behind in nT

[InVasMani]

For gaming in general for most consumers since we're still stuck firmly in the 8C/16T era of console ports a 4P + 4E is actually sufficient and won't really hold things back much in terms of game performance on average. Like even in the case of Direct Storage gaining greater traction it's intended to decompress on the GPU resources SMT/HT isn't really beneficial for that scenario in the same way it is for 7-zip.

The biggest thing Intel could do if they wanted to make a cheap economical gaming chip is just sell a 4P + 4E chip with a larger cache because that's all the X3D options are in essence from a gaming perspective. Everything pretty much hinges on current console specs heavily. We won't see more serious deviation from game to game until we transition to a console utilizing more cores out of necessity of particular developer driven game design change reasons.

The only real reason we have more than 8C/16T in general is due to productivity and multitasking to avoid background tasks getting in the way somewhat at times. It's not really leveraged for gaming much due to consoles being the largest target audience and minimal effort game ports.

I said it few months back Intel could hypothetically do such a chip as means to compete with X3D. I mean AMD basically just targeted a game chip aimed around console limitations that most developers will closely adhere to.

Lesson learned for Intel always offer a game orientated CPU that matches console core and/or thread specs closely and can provide the best game experience tailored to them because that's what developers will be aiming at.

I actually think a 1P and 8E or a 2P and 8E would be pretty ideal with a big slab of cache. Just brute force the P core/cores and bolster it with a bunch of MT from the E cores and tack on a big chunk of cache to both. It would be very affordable and probably still be about 90%- 95% as good as a current X3D chip or possibly even better depending on the amount of cache involved and just how big a role it plays for games.

 

[stimpy88]

It's interesting that the Intel boys are loving E-cores and are all for the removal of HT. So, you want a downgrade for consoles? Right now they have 8 full-fat cores with 16 threads, and that is the standard, even if you don't like the manufacturer.

Making a PS6 with only 8 cores, 4 normal, 4 castrated for "reasons" and all having HT removed so only 8 threads are supported would be a colossal downgrade, even if those cores were running a 6GHz. Plus, now all games would have to be re-compiled, and backwards compatibility would have to be emulated for the vast majority as performance issues would be prevalent in such a design, and soon that 6GHz will wish it was 12GHz pretty fast!

I don't get why Intel kids hate HT now, Intel invented the bloody thing in the first place! Go and benchmark your system right now, a mix of productivity and gaming, write the results down then go into your BIOS and turn HT off and repeat your benchmarks then come back here and post your honest results. I promise you that many benches, the vast majority would lower by at least 20% with some being affected more.

I don't want to live in your dystopian world where I cream my pants for a 1P 8E CPU, even if it's running at 12GHz!!!

You do understand why E cores exist in the first place don't you? You do understand that?

 

[mahirzukic2]

You do understand why E cores exist in the first place don't you? You do understand that?

Sadly and apparently and they don't.

 

[N/A]

Skymont E-core IPC is on par with Raptor P-core. 8E is a 12th gen i5 12400F equivalent or even better.

 

[stimpy88]

Skymont E-core IPC is on par with Raptor P-core. 8E is a 12th gen i5 12400F equivalent or even better.

And what's its clockspeed?

 

[Tomorrow]

Skymont E-core IPC is on par with Raptor P-core. 8E is a 12th gen i5 12400F equivalent or even better.

That's what Intel claims, but it's yet to be proven.

 

[Darmok N Jalad]

The challenge with P+E is that there are still times when Thread Director or the OS gets its bungled up and won't load all cores when you think it should be able to. I see it at work on my Adler Lake laptop daily, and it's usually MS's own Excel! I think there are just times where it won't ever use all the cores because the task I just gave it requires waiting for threads to be complete before the next can be queued, and the uneven core strength just can't handle that kind of timing. At that point, I actually wonder if the task should have been run on the 8 E cores instead, and not the 2 P cores. It's maddening to watch Task Manager and see 2 cores maxed out with the other 8 doing almost nothing. I suspect Zen P+C cores would fare better, but who knows.

 

[InVasMani]

Looks like a 5950x with light years better ST. The ST performance uplift jump is pretty substantial. It's kind of a side grade probably relative to Raptor Lake, but not too bad given ST is typically more important than MT to most users. It's not exactly lackluster in MT at the same time. Depending on which SKU this is and other factors like pricing and how power draw looks in practice this isn't too bad. Seems a better option for most than a 14700K if priced cheaper unless they really need or want heavier MT.

I think on average though it's going to be more serviceable with the hefty ST uplift. If things shift a bit more heavily to heavier MT it might be a little more dubious in relative terms, but if priced better doesn't really matter. Also whatever tweaks have been done to IMC and/or IGP and things along with chipset features. Will have to see how it stacks up against the new Zen chips, but isn't looking terrible.

I mean people were complaining vocally that they wanted more P cores essentially for higher ST and this is pretty much that coincidentally though they went about things differently. It looks like a solid change and concession by Intel and if this is merely the i5 SKU this is looking rather nice probably to people looking to a build a system in the near future.

 

[TF-GrayWizard]

Sadly its a CPU-Z bug

Reproducible by running the version 19 AVX beta test then switch it back to the version 17 test but don't run it then select a CPU to compare to and the ver 19 results show up.

further proof

purely hypothetically if the 14% IPC is translated into say a 13900KS 1T (stock is around ~950 1T afaik) and so it would likely be in the range of ~1080 score which would be cool to see a CPU break the 4 digit mark on 1T CPU-Z anyway.

 

[Minus Infinity]

Arrow Lake-S will have HT. It's only Arrow Lake-U that's confirmed not to have HT.

Where'd you hear that? You're the only person saying that, that I've come across.

 

[Sunny and 75]

Arrow Lake-S will have HT. It's only Arrow Lake-U that's confirmed not to have HT.

Where'd you hear that? You're the only person saying that, that I've come across.

Guess all will be cleared up at the 2024 edition of the InnovatiON event.

on chiller

XD! The good old W-3175X! XD!

Spoiler: Never gets old

 

[bitsandboots]

The "i" went back to Steve Jobs...

Still stuck with "7" in the naming scheme from the Windows 7 days.
But hey iPod was like 8 years before Windows 7 so I think we can get another 8 years out of the 3/5/7/9 naming scheme before they realize it's dated.

 

[MrMilli]

Where'd you hear that? You're the only person saying that, that I've come across.

Multiple sources confirmed that Lunar Lake removes HT but Lion Cove supports it.
It's speculation on my part that Arrow Lake-S will have it but that seems very reasonable to assume.
Certainly Lion Cove in server parts will have HT.

 

[stimpy88]

Multiple sources confirmed that Lunar Lake removes HT but Lion Cove supports it.
It's speculation on my part that Arrow Lake-S will have it but that seems very reasonable to assume.
Certainly Lion Cove in server parts will have HT.

I just don't get how Intel is going to make up for the +-30% performance loss. I hear that they have improved IPC by 20% but where is the other 10% coming from, certainly not from clock speed?

 

[N/A]

As you said for Intel CPU 80% result without SMT is 20% loss or 25% gain for 15% area which is physically removed now though. Pcore 6C x6.0Ghz occupies 39 mm2 instead of 45, when a low-power island cluser of 4E occupies 8 mm2 area and that's 10 threads operating at 6C x 6Ghz plus 4C at 4.5Ghz instead of 6C at 5.9Ghz+25%, or 20% more perf/area.

 

[Prima.Vera]

I'm in a market for a new CPU when the single thread score will be more than 1700 points. Until then... good luck.

 

[stimpy88]

As you said for Intel CPU 80% result without SMT is 20% loss or 25% gain for 15% area which is physically removed now though. Pcore 6C x6.0Ghz occupies 39 mm2 instead of 45, when a low-power island cluser of 4E occupies 8 mm2 area and that's 10 threads operating at 6C x 6Ghz plus 4C at 4.5Ghz instead of 6C at 5.9Ghz+25%, or 20% more perf/area.

View attachment 354213

I will need to see independent testing to see if they were really able to make up for the massive loss of performance. There sure will be perf regressions under some conditions, and just testing the top 10 games and Adobe software are going to be exactly what Intel would demand the tech-tubers, and the paid shills/advertisers like Cutress do to push positive stories and reviews.

 

[Nanochip]

No, we absolutely do know that HyperThreading is dead in Intel CPUs going forward.

It’s not dead in Xeon.

 

[Noyand]

I will need to see independent testing to see if they were really able to make up for the massive loss of performance. There sure will be perf regressions under some conditions, and just testing the top 10 games and Adobe software are going to be exactly what Intel would demand the tech-tubers, and the paid shills/advertisers like Cutress do to push positive stories and reviews.

Looking at the recent news, ARL might only need one thing to beat RPL: not being fucked on an architectural level. It turns out the instability issues of 13th and 14th gen are deeper, it's not just about the CPU being pushed too hard, the design itself might be faulty, and not fixable with microcode updates

 

[chrcoluk]

If this is relevant to what actually happens, I feel it will make it a better balanced processor.

Single core performance is still king in mainstream consumer workloads, raptor lake and alder lake are too emphasised on cinebench type performance which contributes to the amount of energy they can absorb and heat produced.

If you was to offer me today e.g. a 13700k with no HTT support but 25% better single core, snap your hand off.

No, we absolutely do know that HyperThreading is dead in Intel CPUs going forward.

Yep, although I think this sample isnt a 8/16 chip.

Here is some notes from some maths I did.

695 per logical core multithreaded
538 per logical core multithreaded - 22% loss vs 13900k
882 per logical core multithreaded normalised value 27% increase single core vs 13900k - 39% loss for 538 score - must be less cores HTT isnt that good.

It's interesting that the Intel boys are loving E-cores and are all for the removal of HT. So, you want a downgrade for consoles? Right now they have 8 full-fat cores with 16 threads, and that is the standard, even if you don't like the manufacturer.

Making a PS6 with only 8 cores, 4 normal, 4 castrated for "reasons" and all having HT removed so only 8 threads are supported would be a colossal downgrade, even if those cores were running a 6GHz. Plus, now all games would have to be re-compiled, and backwards compatibility would have to be emulated for the vast majority as performance issues would be prevalent in such a design, and soon that 6GHz will wish it was 12GHz pretty fast!

I don't get why Intel kids hate HT now, Intel invented the bloody thing in the first place! Go and benchmark your system right now, a mix of productivity and gaming, write the results down then go into your BIOS and turn HT off and repeat your benchmarks then come back here and post your honest results. I promise you that many benches, the vast majority would lower by at least 20% with some being affected more.

I don't want to live in your dystopian world where I cream my pants for a 1P 8E CPU, even if it's running at 12GHz!!!

You do understand why E cores exist in the first place don't you? You do understand that?

HTT aside from being a complete security mess, are not actual cores, its just an extra thread allowed on the core where the second thread might be able to process something if the first thread is waiting for an i/o response. It works "ok" in specific workloads and is useless in "most" workloads. Its also power inefficient.

E-cores are actual physical cores, even though they not as fast as p-cores they will always offer more performance than a 2nd logical thread even under ideal HTT conditions. Also with e-cores being a different performance class on the windows scheduler, it can automatically move background processes to e-cores whilst keeping foreground processes on the p-cores which offers scheduling advantages. Same as what phones do.

Its nothing to do with fanboyism but rather progress. Not sure why you using terms like intel kids.

Not sure I want a 1P + 8E gaming CPU though, I prefer to keep top end chips as 8P, but no need to go above 8P in my opinion. Anything over that should just be e-cores, and reduce down to 4E from 8E/16E, add cache in the saved silicon and you have a beast of a CPU. Also the N100 which is an e-core only chip is proving very popular in the NUC community, most powerful portable Intel chip I have ever used.

 

[Sunny and 75]

Looking at the recent news, ARL might only need one thing to beat RPL: not being fucked on an architectural level. It turns out the instability issues of 13th and 14th gen are deeper, it's not just about the CPU being pushed too hard, the design itself might be faulty, and not fixable with microcode updates

This^

 

[InVasMani]

Looking at the recent news, ARL might only need one thing to beat RPL: not being fucked on an architectural level. It turns out the instability issues of 13th and 14th gen are deeper, it's not just about the CPU being pushed too hard, the design itself might be faulty, and not fixable with microcode updates

Intel could probably reduce ring ratio and cpu ratio to resolve it the problem at the cost of performance. It won't go over well though since they market to operate higher. Intel's pretty much stuck in a recall scenario no matter what. Consumer will be up in arms no matter what about this situation even if the former is a solution to the outright degradation failure from pushing the hardware itself too far beyond it's reliability means.

 

[Sunny and 75]

Intel could probably reduce ring ratio and cpu ratio to resolve it the problem at the cost of performance. It won't go over well though since they market to operate higher. Intel's pretty much stuck in a recall scenario no matter what. Consumer will be up in arms no matter what about this situation even if the former is a solution to the outright degradation failure from pushing the hardware itself too far beyond it's reliability means.

Sure.