OmegaSupreme
advanced basic bitch
Not a bad rule but it was completely not necessary the last generation. The PS4 and Xbox both had 8 core cpus but they were easily smoked by quad cores on pc. It's about more than just core count.
-
Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.
PaintTinJr
Member
It looks like an interesting comparison IMO, because - going by technpowerup info - you have smaller caches - L2 2MB(256k per core) vs 4M and L3 16M vs 32M - yet have those caches running 20% faster, so you could possibly be cache size bound, which would maybe explain higher utilisation.U got some weird usage in horizon tho. my 9900k does this.
Now this is ultra settings at 1080p, didn't test with lower settings as i was cpu bottlenecked here, so not really pushing it, but again cpu utilization seems perfectly fine, and with everything above the 75% taxation, its safe to say it does use more then 6 cores and 12 threads. Unless the software is wonky or my background tasks are consuming more performance neeeded if 6/12 cores is the limit the engine can use. Which again pushes me towards my point of 8 cores being useful over 6 cores even in games designed for 6 cores only.
Memory speeds and memory channel count might also change things, although I did notice
Md Ray
machine looks to be using an SSD software raid, which might be causing interrupts - that disrupt CPU utilisation.The other week I got my first look at my nephews 10th gen 6-core i5 (K version) RTX 2060 PC, and was surprised how just freeing explorer to have 1 thread per window, disabling hibernation mode, disabling the recycle bin on his nvme, relocating the swap file on his HDD, and changing his bios to detect CPU temp for case fans - instead of the MB - made reasonable gains on games. So as you say, your OS configurations might be altering utilisation too.
Md Ray
Member
Because those 8 cores were simply too weak. Each CPU core was even weaker than Intel's Core 2 Duo and 8 cores combined (of the PS4) was about as fast as an Intel Core 2 Quad Q6600 from 2008, that's how terrible they were. Console Zen 2 on the other hand is nowhere near that behind when compared to PC counterparts.
Last edited:
OmegaSupreme
advanced basic bitch
I know. That's why I said it's bout more than just core count. Those ps4,xbox cpus sucked ass even for the time. They really cheaped out there.
Md Ray
Member
Nah, I haven't installed any RAID driver when I built my PC. I've just populated all the SATA ports, plus a USB port with HDDs for storage, backup purposes.It looks like an interesting comparison IMO, because - going by technpowerup info - you have smaller caches - L2 2MB(256k per core) vs 4M and L3 16M vs 32M - yet have those caches running 20% faster, so you could possibly be cache size bound, which would maybe explain higher utilisation.
Memory speeds and memory channel count might also change things, although I did notice
The other week I got my first look at my nephews 10th gen 6-core i5 (K version) RTX 2060 PC, and was surprised how just freeing explorer to have 1 thread per window, disabling hibernation mode, disabling the recycle bin on his nvme, relocating the swap file on his HDD, and changing his bios to detect CPU temp for case fans - instead of the MB - made reasonable gains on games. So as you say, your OS configurations might be altering utilisation too.
Not every game behaves like HZD and Days Gone, mind. DOOM Eternal is optimized for 8C Zen 2 and here's how it uses the CPU, as you can see it spreads the workload almost equally across all the cores and threads as it should:
Here's HZD below for comparison's sake:
Bonus: ROTTR DX11 vs DX12:
PaintTinJr
Member
Yeah I should have looked more closely, it isn't two drives as a software raid, but 1 ssd as 2 disksNah, I haven't installed any RAID driver when I built my PC. I've just populated all the SATA ports, plus a USB port with HDDs for storage, backup purposes.
Not every game behaves like HZD and Days Gone, mind. DOOM Eternal is optimized for 8C Zen 2 and here's how it uses the CPU, as you can see it spreads the workload almost equally across all the cores and threads as it should:
Here's HZD below for comparison's sake:
Bonus: ROTTR DX11 vs DX12:
Armorian
Banned
Most games don't use more than 8 threads and that's why single core performance is way more important than huge number of cores.
Jedi FO is in fact 4 thread game IIRC, it sits mostly on single core performance.
This CPU is just weak and CP wrecks CPUs with RT. Game can use many threads but there is no improvment above 8 I belive.
If you aren't testing RT in 2021 then you aren't testing jack. Once you take that into account the landscape shifts very quickly. Outlets like HWU unfortunately prefer to be wilfully ignorant of best testing methods.
Only the german sites seem to understand how to do it atm:
Raytracing und die CPU - Was kostet Raytracing im Prozessorlimit?
Raytracing stellt bekanntermaßen hohe Ansprüche an die Grafikkarte. Aber wie wirkt sich die Technologie auf die Prozessor-Performance aus?www.pcgameshardware.de
PS: Never quote GameGPU. 99% of their results are "estimated" aka FAKE.
They have to test at least some configurations. Tell me from where you know they estimate the results?
Yes, but CPU itself is much weaker than Zen2 from desktop and devs have 7 cores and 14 threads available.
7/14.
Most games ends scaling above 8 threads. There are exceptions (and there will be more in the future of course) like SotTR.
Most CPU outlets only test handful of games over and over again, this site tests almost all new games:
Most games behave like this. You need stronger cores more than more cores for games mostly. And this won't change anytime soon.
CuNi
Member
Most games don't use more than 8 threads and that's why single core performance is way more important than huge number of cores.
Jedi FO is in fact 4 thread game IIRC, it sits mostly on single core performance.
This CPU is just weak and CP wrecks CPUs with RT. Game can use many threads but there is no improvment above 8 I belive.
They have to test at least some configurations. Tell me from where you know they estimate the results?
Yes, but CPU itself is much weaker than Zen2 from desktop and devs have 7 cores and 14 threads available.
7/14.
Most games ends scaling above 8 threads. There are exceptions (and there will be more in the future of course) like SotTR.
Most CPU outlets only test handful of games over and over again, this site tests almost all new games:
Most games behave like this. You need stronger cores more than more cores for games mostly. And this won't change anytime soon.
Yes and no. I never got the logic. While saying "game uses X cores so you also only need X cores" is wrong. If a game needs 8 cores, it's smarter to get more cores than 8 faster cores, obviously as long as the difference isn't like 1ghz or something ridiculous.
People forget that pc gamers don't just have the game running. You got discord, steam, maybe eben other launchers, chrome/Firefox etc. The game performs better if it can utilize 8 cores and offload the other Programms onto other cores instead of all the programs competing for the same cores that just run faster. CPU speeds were mostly important when we were still talking about games running on 1 thread alone because then yes you'd notice a noticeable difference. In today's time where games regularly already use more cores and it starts to be even more spread out, more cores is better than one faster core.
Armorian
Banned
You have the results showing that anything more than 6/12 in not needed now, when it will be needed Zen 5 will destroy your Zen 2/3 CPU anyway in every category. More cores for programs in the background make sens but in reality anything hogging CPU much will affect CPU heavy games, Windows is not good at setting programs to specific threads and there will be conflicts. I don't use anything other than game launcher (Steam etc.) and maybe spotify, Windows isn't even close to use as many resources as console OS, without something like chrome in the background it barely do anything. But yeah if you are streaming or some shit more cores come to the rescue. But games itself don't need them.
CuNi
Member
The issue is, the CPUs with more cores usually are on better silicon and thus even run faster. If you don't OC, the 5950 has the highest official boost clock so even with your assumption that u need faster cores than more cores, the 5950 still boosts 100Mhz higher than the 5900 for example, which also boosts higher than a 5600 and equal to a 5700.
PaintTinJr
Member
The other thing also about brawny CPUs with more cores, is that - with Intel definitely - the cache size for the L3/LLC is either the same or proportionally bigger as you go up the core-count/price-scale, so in lower core utilisation workloads the shared L3 cache gives proportionally more effective memory bandwidth per core in like for like configs, where only the CPU has been changed.
I also suspect on the basis of this, if someone disabled cores - via the bios - for the higher end core-count chip, to give parity, and then overclocked, the better silicon, as you mention, would actually clock (the lower core count) cores higher.
Last edited:
Dream-Knife
Banned
People greatly overestimate how much that uses. I have Steam, discord, Vscode, Brave with 5+ tabs, wsl2, and discord running and I'm using 1-5% cpu and 9.3gb of RAM.
Aren't most lower core chips the same as their higher core chips just with cores disabled to increase yield?
CuNi
Member
I have 4 chrome windows open with each like 50 tabs. But I know I'm a edge case so won't bring this in as a argument.
Yes cores are disabled for yield, but the exact same reason why they have to disable cores also means the whole chip isn't perfect and that's why lower core CPUs usually also have lower frequencys.
Also there is a difference in chip between high end and low/mid end, at least on AMD. I haven't checked Intel yet. AMDs 5900 and 5950 has 2 CCDs so 2x 32MB L3 cache, while 5800 and lower only has 1 CCD which means also only 1 32MB L3 cache.
There is a lot more going on under the hood than just disabling some cores and/or having more cores. Architecture needs to account for that.
Kenpachii
Member
10200 = memory clock on gpu, stock is 9500 i believe. GDDR6x
3200mhz on the cpu front = 12gb used. i got 32gb, could be background processes, or simple windows reserving more memory space. which does happen when u put more memory into your PC for faster operations when its available.
Last edited:
RoadHazard
Gold Member
Console games don't get to use all the cores for game code, there's an OS running in the background etc. Although I guess that's true on a gaming PC too.
PaintTinJr
Member
I whole heartedly agree, they weren't what we wanted as gamers following the Cell Be or Xenon, but when you consider the situation, and where we are now...
The PS5 (at least) has a full 8 core brawny CPU for game code, with audio and I/O offloaded onto the Tempest Engine and the IO complex. An IO solution with its own ESRAM so it can DMA directly to unified memory - much like Cell BE SPUs could work as satellite processors, un-salved from the primary PPU controlling core - and the CPUs new brawny cores can also support heavy use of 256bit AVX2 instructions AFAIK - for CPU Ray tracing - going by Cerny's comments in the Road to PS5, but this was only possible because PlayStation and Xbox switched lanes with their processor designs in moving to APUs, and took the 7 year generation hit of Jaguar cores - because Intel weren't competitive on pricing, POWERPC wasn't guaranteed to survive as a mainline architecture, and Jaguar cores were all that was technically available for APUs at that time.
I presume the choice of an APU was driven by three factors: Firstly, memory prices were no longer falling as they had - with age - so having one unified memory pool reduced potential hardware costs for the whole generation, at day one. Two, games have lots of redundant data using discrete memory pools for CPU and GPU - that is very wasteful in processing latency, wasteful in northbridge bandwidth, and very wasteful in memory use which can all be eliminated with unified memory in a APU solution, and gain performance too. Thirdly, the Cell BE and Xenon in the PS360 gen were in effect used as single core, two way general purpose CPUs for game logic on their primary PPU - the 360 had two additional PPUs, but in hindsight the tasks those additional PPUs did could(and is now) be done more efficiently with an ASICs, and because using those PPUs for game features used more of the x360 memory bandwidth, which was likely better used by the GPU, given its tiny EDRAM that made it difficult for it to match the (720p)HD ready resolution of that console generation and already strained the unified memory.
So, IMHO spending more money wasn't an option to reach where we are now - where a 12 core CPU and/or RTX IO board with 6 core might be needed by the end of the generation to match the PS5's CPU use.
Last edited:
YeulEmeralda
Member
I recall the PS4 reserving a gigabyte of RAM for the OS.
Armorian
Banned
3 gb for this slow fucking os.
While HUB are correct that you only need a set amount of performance for gaming there are going to be some cases where slack devs expect 14 threads and even if your 12 threads are fast enough the lack of 2 threads is going to hurt performance.
I also feel that it does somewhat miss an element. For me I intend to get a a 5800X 3D if AMD make that sku available because I expect that with a B550 motherboard will be cheaper than 6c zen 4 + B650 + DDR5 and I don't think the performance delta will be worth the cost.
I also feel that it does somewhat miss an element. For me I intend to get a a 5800X 3D if AMD make that sku available because I expect that with a B550 motherboard will be cheaper than 6c zen 4 + B650 + DDR5 and I don't think the performance delta will be worth the cost.
Md Ray
Member
Exactly. In such cases, there's a chance your 1% and 0.1% low min fps can suffer even if the avg. is good enough.