9700K vs 3700X

[demetri]

In real VR test cases I can tell you that with all the i9-9900K cores running 4.8GHz wirh 4 x 8Gb 3200MHz DDR4 memory I can max out one or two of it's cores easily just by tweaking the super sampling.

The key problem is still lack of parallel processing by the game engines because there is plenty of extra CPU horsepower being wasted.

Why does super sampling affect CPU usage though? It should be a GPU-intensive load

 

[RCHeliguy]

Why does super sampling affect CPU usage though? It should be a GPU-intensive load

All I can say is that it does. When I increase the Super sampling enough I can cause a CPU bottleneck and the GPU will start dropping in % used as ASW kicks in. Until that point the CPU goes up linearly with the GPU utilization.

In some games I can reach SS of 2.0. In Dirt Rally and PC2, I could get an extra .1-.2 SS when I boosted my MB memory from 2166MHz to 3200MHz. So memory can be a bottleneck too.

BTW with a 1080Ti I can get PC2 up to about 1.5-1.6 before ASW kicks in. With Dirt Rally 1.0 it is closer to 1.3. I only get SS of 2.0 with non-racing sim games.

 

[Deleted member 197115]

Their 9700K usage never went above 49% in that PC2 test, so I guess it was not the extra HT cores that made the i9 faster there. Could be higher boost clock or 33% larger L3 cache that made the difference

That's average, doesn't mean individual cores taking load were not taxed out.
But the difference is not that great indeed and it's $100 cheaper though, small saving but still.

 

[demetri]

That's average, doesn't mean individual cores taking load were not taxed out.
But the difference is not that great indeed and it's $100 cheaper though, small saving but still.

Yeah, but per core performance should be pretty much identical for these CPU running at the same clock speed, so if an i7 core was maxed by something out so would be an i9 one.

Anyway, back to the actual topic. Just watched this

Steve made a pretty good point of 3700X being not as cheap as 3600 and barely passing it and very much not as fast as 9700K for gaming basically making it an odd choice for a gaming-centric system. And that's even with the Intel considered to be more expensive which is not the case for me.

 

[demetri]

All I can say is that it does. When I increase the Super sampling enough I can cause a CPU bottleneck and the GPU will start dropping in % used as ASW kicks in. Until that point the CPU goes up linearly with the GPU utilization.

In some games I can reach SS of 2.0. In Dirt Rally and PC2, I could get an extra .1-.2 SS when I boosted my MB memory from 2166MHz to 3200MHz. So memory can be a bottleneck too.

BTW with a 1080Ti I can get PC2 up to about 1.5-1.6 before ASW kicks in. With Dirt Rally 1.0 it is closer to 1.3. I only get SS of 2.0 with non-racing sim games.

I think @Andrew_WOT here runs the same setup and he's using 200% steam VR supersampling in AC with the same specs PC and Odyssey+. Andrew, I wonder if you noticed the same pattern of CPU becoming a bottleneck with higher SS settings

 

[Deleted member 197115]

I think @Andrew_WOT here runs the same setup and he's using 200% steam VR supersampling in AC with the same specs PC and Odyssey+. Andrew, I wonder if you noticed the same pattern of CPU becoming a bottleneck with higher SS settings

No, GPU only, after my update to i9-9900k only GPU bottlenecks, easily observed in fpsVR.
In AC I am using 212%, in PCars2 need to dial down to 106%. Keep in mind that new 100% for Odyssey is 147% now (distortion correction multiplier), so it's more like 312% and 155% of native resolution.
Native: 1440x1600=2,304,000
AC SS: 2404x2992=7,192,768

May be Oculus process things differently in their SDK.
But it's common knowledge that resolution only impacts GPU, CPU processing of the scene should stay the same.

 

[demetri]

No, GPU only, after my update to i9-9900k only GPU bottlenecks, easily observed in fpsVR.
In AC I am using 212%, in PCars2 need to dial down to 106%. Keep in mind that new 100% for Odyssey is 147% now (distortion correction multiplier), so it's more like 312% and 155% of native resolution.
Native: 1440x1600=2,304,000
AC SS: 2404x2992=7,192,768
View attachment 314631
May be Oculus process things differently in their SDK.
But it's common knowledge that resolution only impacts GPU, CPU processing of the scene should stay the same.

Thank you, that's what I thought as well. Kind of strange for that to put more load on the CPU

Do you mean that old 147% setting equals 100% for the current Steam VR version or the other way around? Is it for beta only? I use the non-beta version 1.5.16 and I'm getting a smaller resolution if I choose 212% (2072x2592)

 

[Deleted member 197115]

Thank you, that's what I thought as well. Kind of strange for that to put more load on the CPU

Do you mean that old 147% setting equals 100% for the current Steam VR version or the other way around? Is it for beta only? I use the non-beta version 1.5.16 and I'm getting a smaller resolution if I choose 212% (2072x2592)

Beta
The old one with 210%

Now you can go up to 256% with default renderTargetScale 2.0 without losing sharpness, they increased internal buffer to accommodate Reverb.
But you can also bump it up as high as 5.0.
We can continue here to keep this thread on topic.

 

[RasmusP]

Their 9700K usage never went above 49% in that PC2 test, so I guess it was not the extra HT cores that made the i9 faster there. Could be higher boost clock or 33% larger L3 cache that made the difference

Like the others already said, that doesn't mean a thing.
As I've written, each splitsecond one core is peaking to 100%, then the next one. Windows shuffles every cpu task across all cores so while one cores is processing its task, the next one can prepare.
You gain performance by this although in theory, a single thread of any application can only use one single core to 100% (which is this 12.5% border in process explorer for an 8 thread cpu).

The more cores a cpu has, the lower the cpu load will be when it reaches the single thread performance limit.
So the 9700k might be limited at 49% in pcars 2 and the 9900k might be limited at maybe 35%.
It still pushes out a tiny bit higher fps.

The more cores (or threads in case of HT), the more windows can spread the load.
You'll gain a tiny bit of performance for each core or with HT but it's nowhere near the gain with a higher IPC!

So the 9700k gains a little bit over the 9600k and the i9 gains a very very tiny bit over the 9700k but it drains more power, needs better cooling, is more expensive etc.

 

[F_B]

I totally believe you that you're totally fine with your 2600x, not going against it.
But "most sims just fine" and "good enough" aren't very accurate statements.
Do you have some statistics of how much reprojection % you experiencing with some example-situations?
Like maybe playing a full grid replay in Assetto Corsa that you could share for us for comparison?
(Since a replay has a slightly lower load, maybe run a mod track with 30+ cars?)
And of course what apps you're running in Assetto Corsa.

My 2600k is dropping down to sub 60 fps during 24 car multiplayer race starts but I do have quite some apps active!

Well, that wasn't quite exact, I know. »Just fine« means I can run most sims on medium/high and don't feel any severe stuttering if at all. I have to keep shadows on low/medium though because they're often fps killer. In the Oculus Tray Tool I set ASW to Auto. I can run car fields +20 and have a good fps rate. Keep in mind that everyone experiences lack of frames differently, some notice when ASW kicks in and others - myself included - don't have a problem with that.

 

[RCHeliguy]

It is possible that I am seeing something different with ASW than I will when using the SteamVR drivers when I switch to the Valve Index in a few weeks, but as it stands now with the Rift, it does have an impact. I'll try to get some CPU load images before and after for comparison.

 

[Frank]

:')

 

[Sim Strata]

Looking at the reviews, the Ryzen 3rd gen looks better for gaming. Even though the max frames are lower, the min frames are higher. So they say.

 

[demetri]

Looking at the reviews, the Ryzen 3rd gen looks better for gaming. Even though the max frames are lower, the min frames are higher. So they say.

Not in the GamersNexus test that I posted here. His .1% lows were higher on 9700K

 

[RasmusP]

Indeed.
Overall I really like what amd is doing, I like their CPUs and I think overall the new ryzen are very good, really hitting the market well with their high amount of cores, allround capabilities, efficiency etc etc.

But if you are desperately needing the very last fps to just stay above the VR refresh rate, then it's, sadly, still the Intel's one should buy.

I mean not many do simracing in VR and if only them would buy Intel I guess it's fine and the competition will stay alive and not go back to what it was the last decade

 

[Sim Strata]

Not in the GamersNexus test that I posted here. His .1% lows were higher on 9700K

Yeah, I guess. In my head I would never pay so much for gaming so I was only thinking of r3600 vs i5-9600k, for which the Ryzen does seem to be marginally better across all reviews. But yes, the 3700x/3900x vs 9700k, Intel looks clearly better.

 

[Terry Rock]

In real-world testing of my R5-3600X, the variation between minimum and maximum frame-rate remains extremely consistent.
To me anyway...that represents a better gaming experience all around.
I have not experienced any slowdown or micro-stutter in any racing sim I currently own. No hitching is remotely present.
I have no plans to manually overclock, so I went directly with the X variant...much as I did with the R5 2600X.
If gaming is what you do most, the 3600 or 3600X is a much better option than the 3700...given the lesser cost versus performance.
As to the I7-9700 series, I wouldn't even consider it at this point...not at $329 non-K to $379 k current Newegg price...unless... you already had an Intel motherboard and was stepping up CPU only.
Sure!...you'd still get higher frame-rates but there does come a point of diminishing returns.
Higher framerate for the sake of saying higher framerate, when you can't visibly see the difference on an actual screen, is just that....higher framerate.
I can load up every car ACC has on tap, maximize almost every option and am at or above broadcast quality.
I can grab my WMR headset and run sampling up and the experience is just as good...and very smooth.
That's the important thing...the gaming experience.

 

[Deleted member 197115]

Sounds too good to be true, sorry, as nothing else on the market can run ACC in VR as you describe.
The title is just not VR ready.

 

[Terry Rock]

Sounds too good to be true, sorry, as nothing else on the market can run ACC in VR as you describe.
The title is just not VR ready.

Really?...
I'm not sure what others are doing or how they're configured, but this was me last week doing a bit of WMR testing ....with my now removed R5-2600X
Wish you could see the original video. It is smooth as glass.
Youtube seems to introduce a bit of what looks like stutter and pauses.
The original is flawless.

 

[RCHeliguy]

aaaahhhhh, no cars on the grid. Got it.

Yeah I had a similar experience with my i9-9900K and 1080Ti. All settings set to max, solid 90fps, perfect rendering, no stutter or pauses.