News Core i9-13900K Early Review Shows Big Gains Over Core i9-12900K

[TerryLaze]

Solidworks and Autodesk Revit to answer your first Q. We can also throw in virtual machine hosts and SQL. There are many, many others once we get into modeling and scientific applications but you'll have to do your own research.

And how exactly do those stop at 16 threads?
How exactly don't you get more performance by using more cores/threads?

However, the major difference between AMD's IF and Intel's E-cores is that, on the other side of the IF, you have more P-cores that are just as powerful as the first CCX. What do you get on the other side of the latency hit introduced by Intel's E-cores?? Eunuched cores (E-cores) that only act as an anchor to the P-cores much of the time. Again, we'll see if Intel corrected the issue with the Raptor chips in about 5 weeks.

It's not major, you can't keep all the cores of an 5950x running at 5050Mhz, so loading up too many cores will turn all of your cores into eunuched cores and not just the e-cores, maybe you will be able to keep single clocks on all the cores of the 7950x if it has any overclock headroom.
On intel at least the p-cores stay p-cores ALWAYS.

 

[alceryes]

And how exactly do those stop at 16 threads?
How exactly don't you get more performance by using more cores/threads?

Ahhh, I may have misread you initial post. But only if you're referring to just more AMD P-cores (Threadripper) and not the Intel combo of P-cores and E-cores. (see below)

On intel at least the p-cores stay p-cores ALWAYS.

Except when the E-cores are actually running, since having E-cores active on Alder lake creates a performance deficit that the computational benefit that the E-cores are supposed to create in the first place can't overcome.

 

[TJ Hooker]

Solidworks and Autodesk Revit to answer your first Q. We can also throw in virtual machine hosts and SQL. There are many, many others once we get into modeling and scientific applications but you'll have to do your own research.

Regarding the two CCXs - this is true. There is definitely an extra latency introduced with the IF.

However, the major difference between AMD's IF and Intel's E-cores is that, on the other side of the IF, you have more P-cores that are just as powerful as the first CCX. What do you get on the other side of the latency hit introduced by Intel's E-cores?? Eunuched cores (E-cores) that only act as an anchor to the P-cores much of the time. Again, we'll see if Intel corrected the issue with the Raptor chips in about 5 weeks.

Have you experienced issues with CAD software with Alder Lake? Because it seems to perform quite well in benchmarks.

Core i9-12900KF, Core i7-12700K and Core i5-12600 in a workstation test with amazing results and an old weakness | Part 2 | Page 2 | igor´sLAB

So today I’ll get serious and show you where Alder Lake S can really score aside from colorful gaming pixels. Gaming what? Completely overrated if you look at at least some of today’s results.

www.igorslab.de

 

[alceryes]

Have you experienced issues with CAD software with Alder Lake? Because it seems to perform quite well in benchmarks.

Core i9-12900KF, Core i7-12700K and Core i5-12600 in a workstation test with amazing results and an old weakness | Part 2 | Page 2 | igor´sLAB

So today I’ll get serious and show you where Alder Lake S can really score aside from colorful gaming pixels. Gaming what? Completely overrated if you look at at least some of today’s results.

www.igorslab.de

Nope. I've always said Intel's P-cores are some of the best performance cores around.

I would love to see these tests done again with completely disabled E-cores. This would allow the P-cores to run at their best.

 

[TJ Hooker]

Nope. I've always said Intel's P-cores are some of the best performance cores around.

I would love to see these tests done again with completely disabled E-cores. This would allow the P-cores to run at their best.

They also have benchmarks of the i5-12400 (no e cores). The results for autodesk compared to the 12600K are essentially in line with the clock speed differences, so seemingly no impact from e cores. In solidworks however the 12400 wins (and ties with the 12700K), so it does look like the e cores are a detriment there.

Intel Core i5-12400 Workstation Review - How does real work succeed without glued-on E-cores? | Part 2 | Page 4 | igor´sLAB

Today we want to question where the upcoming Core i5-12400 can still score apart from colorful gaming worlds and whether it will remain just as frugal and efficient. Gaming is actually overrated when…

www.igorslab.de

 

[JamesJones44]

Except zen 4 7950x will have 16 full performance cores crunching 2 threads per core vs intel’s 8 full cores + 16 useless cores. No desktop needs 16 e-cores to handle background tasks. At most you need 4. The other 12 are there to pump up benchmark scores.

I agree with most here, 95% of the desktop users space will never use 16 or 24 cores. However, I am curious as to how 24 physical (8p, 16e) cores compares to 16 performance physical. Remember SMT is 1) highly dependent on workload 2) Not really a full thread so your not getting a full 32 threads from either platform, but you will get 24 full threads with e cores. IDK if will have much of an effect, but at a minimum we'll see if hybrid really can juice multi core numbers for desktop.

 

[alceryes]

They also have benchmarks of the i5-12400 (no e cores). The results for autodesk compared to the 12600K are essentially in line with the clock speed differences, so seemingly no impact from e cores. In solidworks however the 12400 wins (and ties with the 12700K), so it does look like the e cores are a detriment there.

Intel Core i5-12400 Workstation Review - How does real work succeed without glued-on E-cores? | Part 2 | Page 4 | igor´sLAB

Today we want to question where the upcoming Core i5-12400 can still score apart from colorful gaming worlds and whether it will remain just as frugal and efficient. Gaming is actually overrated when…

www.igorslab.de

Thanks! I missed the i5-12400 in the list.

 

[alceryes]

I agree with most here, 95% of the desktop users space will never use 16 or 24 cores. However, I am curious as to how 24 physical (8p, 16e) cores compares to 16 performance physical. Remember SMT is 1) highly dependent on workload 2) Not really a full thread so your not getting a full 32 threads from either platform, but you will get 24 full threads with e cores. IDK if will have much of an effect, but at a minimum we'll see if hybrid really can juice multi core numbers for desktop.

Remember, in many instances, having 8 E-cores active is still not enough to counter the performance deficit incurred by having E-cores to begin with, compared to E-cores being disabled altogether and letting the P-cores run wild. We'll see how much Intel fixed this issue when the Raptor comes around.

 

[TerryLaze]

Remember, in many instances, having 8 E-cores active is still not enough to counter the performance deficit incurred by having E-cores to begin with, compared to E-cores being disabled altogether and letting the P-cores run wild. We'll see how much Intel fixed this issue when the Raptor comes around.

The effect is slightly slower cache access and is barely noticeable in workloads.
It's a fraction of a single e-core so any amount of e-cores is a benefit.
If you run a cache bandwidth benchmark then the results might look huge but in anything normal it's barely a factor at all.

The big performance issue is with games or software that run on the e-cores alone making it look like the e-cores hinder the p-cores performance, that's a completely different thing though.

Alder Lake – E-Cores, Ring Clock, and Hybrid Teething Troubles

This will be a short post about how Alder Lake’s ring behaves when E-Cores are active.

chipsandcheese.com

 

[alceryes]

It's a fraction of a single e-core so any amount of e-cores is a benefit.

While this may be true in some application workloads, in gaming workloads, E-cores are a detriment to performance on the overall average, in the review results I've seen.
Some results even show consistently better performance on four P-cores and zero E-cores vs. four P-cores and eight E-cores.

 

[shady28]

Nope. I've always said Intel's P-cores are some of the best performance cores around.

I would love to see these tests done again with completely disabled E-cores. This would allow the P-cores to run at their best.

Der8auer overclocked a 12400 to 5.24Ghz all core (can do that with certain motherboards) and ran a few tests.

I suspect this is what a 7600X will look like, in games at least, as the clock on that test is about 7600X level and the IPC on Zen 4 close to Alder Lake.

Short story is, the 12400 @5.24Ghz trades blows in games with a stock 12900K. Same story that AMD had for their 7600X. It also had more consistent 1% to average FPS numbers.

View: https://www.youtube.com/watch?v=8diXDeTDCbo&t=698s

 

[Wolverine2349]

The e-cores suck. But those P-cores are awesome.

If the 12% IPC uplift is true and the P cores can hit 6GHz, that could make Raptor Lake a monster 8 core chip as you disable the e-cores and use 8 only P cores with 35MB L3 cache and much faster than Zen 4 at same clock speed and with games unable to take advantage of more than 6-8 cores anytime soon if ever, could have the best monster 8 core chips for a long long long time at 6GHz

 

[TerryLaze]

While this may be true in some application workloads, in gaming workloads, E-cores are a detriment to performance on the overall average, in the review results I've seen.
Some results even show consistently better performance on four P-cores and zero E-cores vs. four P-cores and eight E-cores.

Sure as I said:

The big performance issue is with games or software that run on the e-cores alone making it look like the e-cores hinder the p-cores performance, that's a completely different thing though.

But this isn't a general problem, e-cores don't affect the p-cores.
It's a software issue just like "game mode" for the first ryzen systems where you had to turn off half the CPU for games not to lose performance, this is the same thing and games are going to catch up just like they did for ryzen.

 

[alceryes]

But this isn't a general problem, e-cores don't affect the p-cores.

Yes, they do. From contention for the same L3 cache to lowering bus speeds, E-cores directly affect P-cores performance. Even if core affinity is being used to lock the E-cores out of anything but menial, background tasks, they are still lowering bus speeds. Only when E-cores are disabled in BIOS are the P-cores allowed to run at their maximum ability.

It's a software issue just like "game mode" for the first ryzen systems where you had to turn off half the CPU for games not to lose performance, this is the same thing and games are going to catch up just like they did for ryzen.

We'll see. Supposedly, Intel has made improvements in Raptor Lake that somewhat (totally?) mitigate the performance deficit from having E-cores active.
Also, the introducing of E-cores, and realization of just how weak they actually are is 100% a hardware issue. The current software 'fix' is to just disable that waste of sand in BIOS. 😉

Also², I called 'Game Mode' complete BS years ago too. I'm not for or against Intel or AMD. I'm against poor design cash grabs and deceiving the public.

 

[The Historical Fidelity]

The e-cores suck. But those P-cores are awesome.

If the 12% IPC uplift is true and the P cores can hit 6GHz, that could make Raptor Lake a monster 8 core chip as you disable the e-cores and use 8 only P cores with 35MB L3 cache and much faster than Zen 4 at same clock speed and with games unable to take advantage of more than 6-8 cores anytime soon if ever, could have the best monster 8 core chips for a long long long time at 6GHz

So buy a 13400K cause everything else has a useless E-core tax on it

 

[Wolverine2349]

So buy a 13400K cause everything else has a useless E-core tax on it

Yeah but no other options with 8 P cores. The 13400 only has 6 P cores and also less L3 cache.

 

[TerryLaze]

Yeah but no other options with 8 P cores. The 13400 only has 6 P cores and also less L3 cache.

But why do you need more than 6 cores but have no use of the e-cores?
It doesn't make any sense, if you want high multithreaded performance than the e-cores add to that and if you don't want it then why do you want it?!

 

[Wolverine2349]

So buy a 13400K cause everything else has a useless E-core tax on it

Is there even going to be a 13400K. With Alder Lake there was no 12400K. Just a 12400 and 12400F with the F just having iGPU shut off which meant locked multiplier and much more restrictive overclocking.

 

[alceryes]

But why do you need more than 6 cores but have no use of the e-cores?
It doesn't make any sense, if you want high multithreaded performance than the e-cores add to that and if you don't want it then why do you want it?!

Unfortunately, the only chips produced without E-cores are chips that have been binned down to that level. They failed to operate with the larger cache and higher frequencies of the big boys.

It would be great if Intel produced a top-tier i9-13900KP (or some such name) with just performance cores that have the same cache and run just as fast as the standard 13900K will. I don't think they would ever produce this because they'd just be giving third party reviewers more fuel to bash the E-cores even more.

 

[JamesJones44]

Remember, in many instances, having 8 E-cores active is still not enough to counter the performance deficit incurred by having E-cores to begin with, compared to E-cores being disabled altogether and letting the P-cores run wild. We'll see how much Intel fixed this issue when the Raptor comes around.

SMT can be detrimental to performance as well, just depends on the workload. Like anything in technology these days you have to pick the products that best fit the workload you are intending to use it for.

 

[Wolverine2349]

Unfortunately, the only chips produced without E-cores are chips that have been binned down to that level. They failed to operate with the larger cache and higher frequencies of the big boys.

It would be great if Intel produced a top-tier i9-13900KP (or some such name) with just performance cores that have the same cache and run just as fast as the standard 13900K will. I don't think they would ever produce this because they'd just be giving third party reviewers more fuel to bash the E-cores even more.

Yeah more fuel to bash e-cores, but more fuel to praise the P-cores and how good they are. Imagine the praise the P-cores would get if they crush AMD's Zen 4 cores. Then they could market if you want 8 good cores or less without the hybrid arch Intel is an easy choice over AMD.

Though if you want more than 8 good cores, AMD is the choice. However for latency sensitive things, there is still an 8 core wall as it has to hop from one CCD to another. Where as Intel, it hops to a lousy core even though latency is not much. There is no CPU that exists except maybe Comet Lake and Haswell/Broadwell-E ring bus that had no latency penalty and more than 8 good cores.
First CPU on better than SKylake like arch that will have that will be Zen 5 as they will have 16 core CCDs and who knows from Intel.

Though for now and probably for a while for high their gaming and not doing background tasks even 6 cores can cut it and 8 is more than enough. More probably not going to do anything for a while as SMP programing with games is like almost impossible that it seems making use of more cores gets so hard. The jump from 1 to 2 and 2 to 4 were easy. But jump from 4 to 6 and beyond games just struggle to use so much.

So faster and faster IPC gain 6-8 core CPUs at least for now are best for gaming assuming no intensive background tasks. Though high end gaming build 8 core slight overkill is best.

 

[alceryes]

SMT can be detrimental to performance as well, just depends on the workload. Like anything in technology these days you have to pick the products that best fit the workload you are intending to use it for.

This isn't just an SMT deficit. It's specifically an E-cores deficit. The example I mentioned was four P-cores and zero E-cores vs four P-cores and eight E-cores.

 

[alceryes]

Though if you want more than 8 good cores, AMD is the choice. However for latency sensitive things, there is still an 8 core wall as it has to hop from one CCD to another. Where as Intel, it hops to a lousy core even though latency is not much. There is no CPU that exists except maybe Comet Lake and Haswell/Broadwell-E ring bus that had no latency penalty and more than 8 good cores.

Specifically for the 7900X, I'm wondering if the IF latency between the two CCDs can be slightly offset by the second CCD having a full 32MBs of L3 cache for only 4 cores (as opposed to 8 cores having to share the 32MBs on the first CCD).
That may be the one I'm getting. I don't think I'll wait for the X3D varients.

 

[alceryes]

Yeah more fuel to bash e-cores, but more fuel to praise the P-cores and how good they are. Imagine the praise the P-cores would get if they crush AMD's Zen 4 cores.

True. But I don't think it's gonna happen with the numbers I've seen. Both camps have great cores (P-cores, at least).

 

[alceryes]

Though for now and probably for a while for high their gaming and not doing background tasks even 6 cores can cut it and 8 is more than enough. More probably not going to do anything for a while as SMP programing with games is like almost impossible that it seems making use of more cores gets so hard. The jump from 1 to 2 and 2 to 4 were easy. But jump from 4 to 6 and beyond games just struggle to use so much.

So faster and faster IPC gain 6-8 core CPUs at least for now are best for gaming assuming no intensive background tasks. Though high end gaming build 8 core slight overkill is best.

Some current games already show a performance benefit with more than 8 cores. Unless you want to be behind the curve, 8 cores, 8 P-cores, should be the absolute minimum to aim for, for max performance in a new system you plan to keep for 5+ years.