How many CPU cores are you running for gaming?

How many cores are you running for gaming?

  • Total voters
    73
xDiVolatilX

xDiVolatilX

2[H]4U
Joined
Jul 24, 2021
Messages
2,636
How many CPU cores are you running for your gaming PC? Reason I ask is because it is genrally easier to hit higher clocks on the P cores with more E cores disabled.

How many do you feel is necessary? How many is a waste?

I think 8 is the sweet spot right now but I personally run 12 cores 8p and 4e with HT off. What do you guys run?
 
Last edited:
I ran a 12-core 3900X and then a 12-core 5900X before switching to an 8-core 5800X3D. It was the first time I ever upgraded to a CPU with fewer cores, but the main game I play (World of Warcraft) makes such amazing use of the extra L3 cache that I just couldn't pass it up. The game also rarely uses more than 4 cores, so even with "only" 8 cores I still have plenty of CPU available to handle the game as well as background tasks.
 
Currently 5950X (16-core) with SMT disabled and static OC.
 
Last edited:
13900K, E cores disabled, HTT disabled, Windows 10. All cores locked to 5.8ghz.

STILL to this day enabling hyper threading degrades gaming performance on a well optimised PC. If you game with lots of background apps running, then yeah sure, enable HTT, but if you truly want the best frame rates possible, it's HTT disabled. Always has been.
 
Full 16 cores plus hyperthreading. My gaming machine is multi-boot also for use with software development.
 
spine said:
13900K, E cores disabled, HTT disabled, Windows 10. All cores locked to 5.8ghz.

STILL to this day enabling hyper threading degrades gaming performance on a well optimised PC. If you game with lots of background apps running, then yeah sure, enable HTT, but if you truly want the best frame rates possible, it's HTT disabled. Always has been.
Click to expand...
I might give this a try. Have you benchmarked it by any chance? I'm asking because it takes a while and I don't have the time. Running 8 P cores at 5.8GHz might be the high end sweet spot for 13900Ks I ran it briefly but running stock now on the KS which is still high st 5.6GHz P cores but I only run 4 E cores so 12 cores considering going 8 cores because this is a 100% gaming only machine.
 
fprish said:
How does one assign how many cores are used during gaming?
Click to expand...
In the BIOS you can control how many cores (specifically E cores because no one turns off P cores lol) you can shut off. For example I have shut off 12 E cores leaving only 4 E cores active plus the 8 P cores of course for a total of 12 cores 8 P 4 E.
 
xDiVolatilX said:
In the BIOS you can control how many cores (specifically E cores because no one turns off P cores lol) you can shut off. For example I have shut off 12 E cores leaving only 4 E cores active plus the 8 P cores of course for a total of 12 cores 8 P 4 E.
Click to expand...
I see, but this says during gaming. So every time you turn on your PC to game you have to go into BIOS and do this?
 
I leave them all on, 24 in my case. Games run good so I see no compelling reason to disable E-cores. Either the GPU or my monitor refresh is the limit in games, not the CPU. Then when I fire up something like video rendering, the E-cores are there to speed it up.
 
Look at the CPU usage here (top left afterburner monitor) This is 8 p cores and 8 e cores hyprethreading off. Spiderman remastered uses all 16 cores, I'm curious what happens when i enable the other 8 e cores also for 24.

16 is almost maxed out though in this game which tells me that games in the future ar
Screenshot 2023-12-28 190553.png
e likely to utilize more cores if spiderman is a sign of things to come.
 
Last edited:
This is with DLSS and frame generation both off. Native 4k 144 hdr enabled on a stock 13900KS and stock 4090 (both underwater tho) and 6400mhz ram in case you're curious about the frame performance in spiderman remastered at 4k native resolution with no software boost just native performance.

This is a MEGA cpu intensive game. If it has my 13900ks (16 cores enabled) at over 90% that's one hell of an achievement .

I will later test 24 to see the difference.
 
Currently 12 cores on one die with no HT is probably the ideal, but might soon be 16 cores that is the ideal. Current sweetspot is 8 cores on a single die with HT on. Hope we get 10 or 12 full speed cores per die in the future as it should use available cores before moving to hyperthreading or similar. Multi CCD chips like the 7900x and 7950x are not ideal for gaming due to the latency when moving between CCDs, but are great for productivity. Scheduling seems to struggle with different types of cores though in a lot of titles, so either need per game optimization or just single core type all around.
 
My current gaming desktop uses a 12700K, so 8P/4E = 12 cores, 20 threads.

I just leave the E-cores on because I can't unlock AVX-512 on this particular CPU stepping by disabling them, so the benefit is minimal. Might as well leave 'em on, I'm hitting overclocking walls trying to go past 5.1 GHz all-P-core stably as it is.

Right now, 8 Golden Cove/Raptor Cove P-cores or Zen 4 cores seems to be the most bang for the buck, especially factoring in Micro Center combo deals. That may change if future games scale past 8C/16T better, but I'm not exactly looking forward to the cost of upgrading to Emerald Rapids/5th-gen Xeon Scalable or Threadripper 7000 here.
 
evhvis said:
Currently 12 cores on one die with no HT is probably the ideal, but might soon be 16 cores that is the ideal. Current sweetspot is 8 cores on a single die with HT on. Hope we get 10 or 12 full speed cores per die in the future as it should use available cores before moving to hyperthreading or similar. Multi CCD chips like the 7900x and 7950x are not ideal for gaming due to the latency when moving between CCDs, but are great for productivity. Scheduling seems to struggle with different types of cores though in a lot of titles, so either need per game optimization or just single core type all around.
Click to expand...


Yes and its a shame neither Intel or AMD have more than 8 homogeneous cores on modern arch on a single die/ring bus/CCD on modern archs. It puts me between a rock and a hard place as I would like such CPU but stuck with 8 cores.

The scheduling issues are real and its like pick your poison or be stuck with 8 cores. Technically no more are needed for gaming but games are getting more threaded all the time such as that Spiderman screenshot example. Its like to just accept e-cores at least they are on same ring bus unlike AMD's extra P cores which are so far away.

Unfortunately the leaks for Zen 5 and Arrow Lake its still going to be a maximum of 8 P cores for Intel on their new node and on AMD side just dual 8 core CCDs yet again. :(:(:( AMD going to have 16 core CCDs for Threadripper and EPYC space, but those will be their own e-cores Zen 4C and Zen 5C and not the full Zen 5 cores. :(:(:(:( ANd likely only available in HEDT and mobile space.

Yes Intel does have more than 8 Golden Cove cores on a single die with Sapphire Rapids but besides just being expensive and requiring , the performance sucks compared to client Golden Cove with much slower L3 cache plus its a mesh arch which has crippled latency for gaming. Plus that platform has limited choices and loaded with too many Enterprise grade features detrimental to things like gaming even if money was of little to no object so its not some good option for gamers. Its basically only a good option if you insist on Intel and want tons of PCIe lanes and other enterpriser grade features like ECC RAM uhh.

Latency with 4 KB Pages: https://chipsandcheese.com/2023/03/12/a-peek-at-sapphire-rapids/​


Worse IPC and of course latency than client/consumer Golden Cove


Core i5-12400 per Cinebench 2024 spanks Xeon 2455X in single thread despite a 200MHz clock speed disadvantage proving IPC is gimped on SPR Golden Cove compared to Alder Lake Golden Cove.


https://www.cpu-monkey.com/en/cpu-intel_core_i5_12400

https://www.cpu-monkey.com/en/cpu-intel_xeon_w5_2455x

Its a shame really cause its not that these companies cannot make such a CPU and could have a couple years ago. They just do not. Well maybe AMD is contrained by TSMC and would be hard to get more than 8 on a single CCD.

But Intel has no excuse. Since 4 e-cores take up same die space and roughly same power draw as 1 P core and they have 8 then 16, they could certainly also make 10 core Golden Cove and 12 P core Raptor Cove dies. They could sell them as alternate i7 and i9 like 13750K and 13950K. Its a shame they do not. And designate that the 50 and 000 i7 and i9 neither is better than other but depends on use cases and preference. But they don't and sadly unlikely they will.
 
Let's not forget that Intel did make a CPU with 10 even, full-performance cores a couple generations back. Didn't it also have AVX512?
 
uOpt said:
Let's not forget that Intel did make a CPU with 10 even, full-performance cores a couple generations back. Didn't it also have AVX512?
Click to expand...
It didn't. It was the next generation (Rocket Lake) that had 512, which necessitated the cut back down to 8 cores because of the increased die area.
 
  • Like
Reactions: uOpt
like this
Grebuloner said:
It didn't. It was the next generation (Rocket Lake) that had 512, which necessitated the cut back down to 8 cores because of the increased die area.
Click to expand...


I do not think that was the reason at all for only 8 cores on Rocket Lake. It was because of Intel 10nm process and yield problems and they had to backport desktop Rocket Lake to 14nm and the new arch had less space on the less dense 14nm thus only room for 8 full performance cores.

A shame to because maybe Rocket Lake would have still been a 10 core part on 10nm and its latency would not have been crippled and would have competed well or maybe beat Zen 3 IPC. At 14nm not only 8 core limit but it crippled its latency and made it a flop where as on proper 10nm likely would have been pretty good kind of like mobile Tiger Lake is though mobile is so much different hard to compare these days.
 
uOpt said:
Let's not forget that Intel did make a CPU with 10 even, full-performance cores a couple generations back. Didn't it also have AVX512?
Click to expand...


Yes they did 10th Gen Comet Lake the 10900K and 10850K all on a single ring bus. And it had its own die I believe where as the 10700K and below had a single die with lower SKUs and parts disabled I think? Shame they do not have something current gen like that.

And that was a great CPU and even can somewhat punch above its weight and well tuned held its own or beat Zen 3 in many games. But pitty its outdated badly by now and IPC so far behind and worse yet stuck on PCIe Gen 3.

As for AVX512 not concerned with having that though for some it may mattrer,

Even with no AVX512 on Golden Cove and Raptor Cove or it fused off, I will gladly take that for a 10-12 P core variant.


Not sure why Intel fuses it off. Do they really not want people disabling e-cores or are embarrassed that it cannot work with e-cores on and want to hide it. Or just do not want to support it anymore?
 
Wolverine2349 said:
I do not think that was the reason at all for only 8 cores on Rocket Lake. It was because of Intel 10nm process and yield problems and they had to backport desktop Rocket Lake to 14nm and the new arch had less space on the less dense 14nm thus only room for 8 full performance cores.

A shame to because maybe Rocket Lake would have still been a 10 core part on 10nm and its latency would not have been crippled and would have competed well or maybe beat Zen 3 IPC. At 14nm not only 8 core limit but it crippled its latency and made it a flop where as on proper 10nm likely would have been pretty good kind of like mobile Tiger Lake is though mobile is so much different hard to compare these days.
Click to expand...
I'm re-reading the Anandtech writeup on it and it seems to be a bit of both. I knew it was backported, and the additional architectural enhancements, including AVX512, added a lot of core area to the CPU, combined with the additional elements to make it work when backported to 14nm, reduced the amount of cores available. I found it pretty interesting that the die size was not just larger than 10C Comet Lake, but even greater than that of Broadwell-E/P's 10C LCC silicon.
 
Wolverine2349 said:
Not sure why Intel fuses it off. Do they really not want people disabling e-cores or are embarrassed that it cannot work with e-cores on and want to hide it. Or just do not want to support it anymore?
Click to expand...
The e-cores don't support it, which means that when the OS moves a process between cores and the new core is missing instructions, the process crashes. I think that there have been scheduler improvements in Windows and Linux in the meantime to address this limitation, but regardless, you can't have both. Because of that Intel actually intended to have 512 fused off from the start, but the marketing (no 512) and engineering (hey, 512!) teams weren't in sync and so it was discovered that it was not in fact fused off at the start and disabling e-cores led to AVX512 support as long as BIOSes did as well. Naturally this bug feature was killed off in later production. The silicon is still there in Raptor lake, but not enabled.

There is a suggestion that Intel's future AVX10 standard will allow mixed bit length compatibility across silicon (like ARM has) coming to all Intel cores, but we'll see.
Wolverine2349 said:
As for AVX512 not concerned with having that though for some it may mattrer,
Click to expand...
It matters to me :). Once I can get my finances re-sorted, I'll be saving up for an Emerald Rapids/TR7000 or (more likely) Granite Rapids/TR8000/TR9000 HEDT system as my next main rig.
 
Grebuloner said:
The e-cores don't support it, which means that when the OS moves a process between cores and the new core is missing instructions, the process crashes. I think that there have been scheduler improvements in Windows and Linux in the meantime to address this limitation, but regardless, you can't have both. Because of that Intel actually intended to have 512 fused off from the start, but the marketing (no 512) and engineering (hey, 512!) teams weren't in sync and so it was discovered that it was not in fact fused off at the start and disabling e-cores led to AVX512 support as long as BIOSes did as well. Naturally this bug feature was killed off in later production. The silicon is still there in Raptor lake, but not enabled.

There is a suggestion that Intel's future AVX10 standard will allow mixed bit length compatibility across silicon (like ARM has) coming to all Intel cores, but we'll see.

It matters to me :). Once I can get my finances re-sorted, I'll be saving up for an Emerald Rapids/TR7000 or (more likely) Granite Rapids/TR8000/TR9000 HEDT system as my next main rig.
Click to expand...


Yeah why did Intel intend to fuse it off. What is worng with just leaving AVX512 as an option for those who disable e-cores as it works without them. And it can just be disabled automatically when e-cores are on at stock.

Any reason other than Intel just being greedy and wanting you to pay up for HEDT to get AVX512? Like any possible conflicts or support issues with e-cores even if you disable or if you do not could AVX512 get activated by accident with e-core son and crash the system or corrupt data?
 
Wolverine2349 said:
Yeah why did Intel intend to fuse it off. What is worng with just leaving AVX512 as an option for those who disable e-cores as it works without them. And it can just be disabled automatically when e-cores are on at stock.

Any reason other than Intel just being greedy and wanting you to pay up for HEDT to get AVX512? Like any possible conflicts or support issues with e-cores even if you disable or if you do not could AVX512 get activated by accident with e-core son and crash the system or corrupt data?
Click to expand...
Initial reliable reporting on the issue

Intel being a butt about ending 512 support

I think there is a lot of behind the scenes mayhem we may never be privy to, but ultimately, to big Xeon land we were sent to have it. With common architectures between desktop and server, core designs are essentially cut-and-pasted between dies, so extraneous silicon will continue to exist at the lower levels.

All of this is in contrast to AMD, of course, who just use the same two cores (big and not so big) with all the same instructions, for all their CPUs.
 
You must log in or register to reply here.
Back
Top