CPU; # of threads matter much?

[HeavyGroovez]

I mean...for me, I just logged cpu/gpu usage for a while and saw which one was getting pegged, and upgraded that. Seems unlikely if your cpu is at 60% usage and your gpu is "high", that it would be a cpu problem. ymmv

Problem is the way that CPU usage is reported. The render pipeline is only as fast as the slowest execution thread. Parallelisation can present the primary execution thread with the resources it needs asynchronously but that doesnt prevent latency as a result of its own intrinsic thread workload.

You have to observe all physical cores and their thread virtualisation to get a true picture of the workload.

 

[Aekero]

Problem is the way that CPU usage is reported. The render pipeline is only as fast as the slowest execution thread. Parallelisation can present the primary execution thread with the resources it needs asynchronously but that doesnt prevent latency as a result of its own intrinsic thread workload.

You have to observe all physical cores and their thread virtualisation to get a true picture of the workload.

this is true, but it's also pretty easy to view that all at once as well, I'd still bet money on "whichever one is pegged" as your bottleneck though, if neither are, or both are, it gets a little trickier

 

[Desert63]

Actually vr ed is cpu hungry. Try overclocking if you haven't already

I am so over my overclocking days . Now its all about stability of performance, not how fast my machine can sprint.
So are you saying overclocking my CPU would help? So that means a CPU upgrade would then? That has not been my experience with ED, again, maxing my CPU at 60% but usually far less - not sure an upgrade or overclock would help.

 

[HeavyGroovez]

this is true, but it's also pretty easy to view that all at once as well, I'd still bet money on "whichever one is pegged" as your bottleneck though, if neither are, or both are, it gets a little trickier

Indeed.

A good example is GPU usage at the threshold of frametime. I often get ~70% GPU usage and ~10% CPU usage representational in Afterburner yet still have frametime far exceeding the v-sync limit and causing reprojection.

 

[TorTorden_VR]

I am so over my overclocking days . Now its all about stability of performance, not how fast my machine can sprint.
So are you saying overclocking my CPU would help? So that means a CPU upgrade would then? That has not been my experience with ED, again, maxing my CPU at 60% but usually far less - not sure an upgrade or overclock would help.

During troubleshooting my pimax one of the thing I did was reset all clocks to default, I was running a 4.9 OC on my 8700k, even set my ram down to default non overclocked and didn't notice anything much at all in terms of performance like that in ED.

So now I'm just running my cpu at 4.7 and my RAM seemed to get unstable after I popped in another two DIMMS so only running those at 2900mhz instead of 3200.

If I run them on 3200mhz I even get problems getting through post.

 

[cmdrrdmc]

So now I'm just running my cpu at 4.7 and my RAM seemed to get unstable after I popped in another two DIMMS so only running those at 2900mhz instead of 3200.

I had to drop my clock by 100Mhz and my RAM wouldn't post past their rated speed anymore after adding another two sticks

 

[TorTorden_VR]

I had to drop my clock by 100Mhz and my RAM wouldn't post past their rated speed anymore after adding another two sticks

Yeah I suspect my mother board not entirely keeping up.
Not the first I have heard of things like that.

To test that I would have to swap out the mother board, and to get anything much better than I got I'm in the premium price range so would be at least $5-600.
I really don't notice a difference on these.

 

[cmdrrdmc]

Yeah I suspect my mother board not entirely keeping up.
Not the first I have heard of things like that.

To test that I would have to swap out the mother board, and to get anything much better than I got I'm in the premium price range so would be at least $5-600.
I really don't notice a difference on these.

Based on what I've read, I think it's actually because more DIMMs place greater stress on the IMC, so I don't think it's a mobo issue. But yeah, not worth the cost for that tiny bit of extra performance in case that is the cause.

 

[Globusdiablo]

Just a quick question about CPUs for VR.

I currently have an i5-6600k, which I am looking at upgrading slightly, as I am not inclined to buy a new motherboard as well at this time.
Looking at the i7-6700k vs i5-7600k and their comparison here, I am unsure what would be considered the best option for VR. The i5 has some aspects where it performs better, while the i7 has others. Which matter the most for VR?

The reason for comparing these is that they are quite comparable in price. I could always go for the i7-7700k (which is max what my motherboard can take), but for comparison reasons the two above make more sense. Additionally I am running a 1060 6GB GPU, which I think will bottleneck if I start going for 8 or 9 gen. So going beyond 7gen would pretty much force me to buy a completely new pc, sans harddrives, and I am not there yet

All thoughts are appreciated.

I found an article years ago that showed that Elite: Dangerous actually prefers threads over cores. You're welcome to browse my history to find it. I posted here in the VR section about it. I have the 6600k as well and seriously considered updating to the 6700k at that time, but decided not to as my VR and E: D days were coming to a halt.

The 6700k should be the perfect upgrade for you.

Edit: Monitoring my cpu at the time, it was evident that my 6600k was a bottleneck running at 100% on all 4 cores during E: D VR during stress. (I did have ultra settings and extended distance rendering enabled though). My gtx1080 showed acceptable loads.

 

[Benji85]

I am so over my overclocking days . Now its all about stability of performance, not how fast my machine can sprint.
So are you saying overclocking my CPU would help? So that means a CPU upgrade would then? That has not been my experience with ED, again, maxing my CPU at 60% but usually far less - not sure an upgrade or overclock would help.

Ocing definitely helps performance in all aspects of computing there is no arguing that. Of course it will vary with each application but I have seen nice improvement in ed from overclocking both this and my last processor.

Overclocking isn't about "sprinting" imo it's about finding a good, safe decent overclock that has no problem running 24/7.
Done property it won't effect stability or the CPUs lifespan. Just get a half decent CPU cooler, make sure your case has good airflow and don't go crazy on the core voltage .
Good luck!

 

[Morbad]

I found an article years ago that showed that Elite: Dangerous actually prefers threads over cores.

Nothing prefers SMT derived logical threads over an equivalent number of physical cores of a similar architecture and clock speed. ED is no exception to this.

SMT is still highly beneficial. But a 4c/8t part is quite a bit inferior to an 8c/8t part, and usually even a 6c/6t part, all other things being equal.

 

[Globusdiablo]

Nothing prefers SMT derived logical threads over an equivalent number of physical cores of a similar architecture and clock speed. ED is no exception to this.

SMT is still highly beneficial. But a 4c/8t part is quite a bit inferior to an 8c/8t part, and usually even a 6c/6t part, all other things being equal.

Not what the article stated, but I'll take your word for it. At the time, the conclusion was a bit surprising to me too.

 

[Morbad]

Not what the article stated, but I'll take your word for it. At the time, the conclusion was a bit surprising to me too.

This the article you are referring to: https://www.tomshardware.com/reviews/oculus-rift-vr-cpu-performance,5215-2.html?

If so, you are fundamentally misinterpreting the article and it's results.

The conclusion of the article wasn't at all surprising and is completely correct, but it also doesn't come anywhere near implying that logical cores are superior to physical cores.

Also: https://forums.frontier.co.uk/showthread.php/423105-6-cores-or-more?p=6637413&viewfull=1#post6637413

An 8700K is slightly faster than an 8600K in VR mostly because it's clocked higher and partially because it has two-way SMT on top of the same number of physical cores (i.e 6c/12t vs. 6c/6t).

A 9700K, which is the same architecture, is even faster than the 8700K, even if you clocked it to the same speed as the 8700K, even with only eight logical cores, because it has two more physical cores and we are already reaching the point of saturation in the number of threads most games can leverage, even with VR.

Anyway, a part with 1n physical and 2n logical will never, ever have more real world aggregate performance than a similar part of 2n physical and 2n logical cores. Two-way SMT adds anywhere from -20% to +60% to effective IPC, with the vast majority of well-threaded apps on modern architectures seeing between +15% and +30% improvement from it. Taking a non-SMT part and adding more than that percentage of physical cores to it will improve performance more than enabling SMT on it. 8c/8t is faster than 6c/12t, most of the time, all other things being equal. This is not at all in contradiction to 4c/8t being much faster than 4c/4t, in suitably threaded apps, or four fast cores being better than eight or ten slower ones that cannot leverage more cores adequately or which are limited by a handful of demanding threads.

 

[Globusdiablo]

This the article you are referring to: https://www.tomshardware.com/reviews/oculus-rift-vr-cpu-performance,5215-2.html?

If so, you are fundamentally misinterpreting the article and it's results.

The conclusion of the article wasn't at all surprising and is completely correct, but it also doesn't come anywhere near implying that logical cores are superior to physical cores.

Also: https://forums.frontier.co.uk/showthread.php/423105-6-cores-or-more?p=6637413&viewfull=1#post6637413

An 8700K is slightly faster than an 8600K in VR mostly because it's clocked higher and partially because it has two-way SMT on top of the same number of physical cores (i.e 6c/12t vs. 6c/6t).

A 9700K, which is the same architecture, is even faster than the 8700K, even if you clocked it to the same speed as the 8700K, even with only eight logical cores, because it has two more physical cores and we are already reaching the point of saturation in the number of threads most games can leverage, even with VR.

Anyway, a part with 1n physical and 2n logical will never, ever have more real world aggregate performance than a similar part of 2n physical and 2n logical cores. Two-way SMT adds anywhere from -20% to +60% to effective IPC, with the vast majority of well-threaded apps on modern architectures seeing between +15% and +30% improvement from it. Taking a non-SMT part and adding more than that percentage of physical cores to it will improve performance more than enabling SMT on it. 8c/8t is faster than 6c/12t, most of the time, all other things being equal. This is not at all in contradiction to 4c/8t being much faster than 4c/4t, in suitably threaded apps, or four fast cores being better than eight or ten slower ones that cannot leverage more cores adequately or which are limited by a handful of demanding threads.

Thanks for digging that up. What I took from that article and debate at that time was that core and thread efficiency is mainly down to software utilisation/optimisation. And as you also imply, benchmarks don't necessarily match real world use.

But this is all way beyond my pay grade and you certainly seem to know what you're talking about, so I won't argue further.

 

[Vtskr]

I'm at the point where i do not actually play ED, but rather just play with settings and fpsVR I have O+, i5-8400 and GTX 1080-ti and so far it looks like ED is CPU bound. Even at compromised beacon i don't have solid 90 fps so motion re-projection kicks in. All that at max settings with 200% SS of cause. I've ordered i5-9600k which also has 6 cores but runs at 4.8Ghz (OCed) instead of 3.8Ghz.

 

[Milner]

I'm at the point where i do not actually play ED, but rather just play with settings and fpsVR I have O+, i5-8400 and GTX 1080-ti and so far it looks like ED is CPU bound. Even at compromised beacon i don't have solid 90 fps so motion re-projection kicks in. All that at max settings with 200% SS of cause. I've ordered i5-9600k which also has 6 cores but runs at 4.8Ghz (OCed) instead of 3.8Ghz.

Hopefully that does the trick!

 

[TorTorden_VR]

Hopefully that does the trick!

A 2080ti wouldn't run elite at 2x regardless of CPU.
I would dial that back down some.

 

[Vtskr]

According fpsVR my GPU frame time is below 11ms, and CPU frametime is 15ms. So if CPU frametime drops below 11ms it should result in solid 90 fps. What i also noticed AO is real resource hog. I'm not even sure what AO does, but it has huge impact on performance

 

[HeavyGroovez]

According fpsVR my GPU frame time is below 11ms, and CPU frametime is 15ms. So if CPU frametime drops below 11ms it should result in solid 90 fps. What i also noticed AO is real resource hog. I'm not even sure what AO does, but it has huge impact on performance

AO Off

AO On

With regards to the benefits of ever more physical cores and virtualisation we are - as Morbad stated - past the saturation point as even the best engines do not fully utilise the parallelisation potential of enthusiast grade architectures at the moment, let alone the HEDT monsters.

 

[Vtskr]

Thats a great example of AO in action. I just can't seem to notice how it works in ED