[Gamegpu] GTA V PC Benchmark

[Silverforce11]

since we know AMD's DX11 overhead is a lot more critical, and GTA V should be bad case for it?

Highly game specific actually. Did not notice the problem in open world game as claimed by some, not in Watch Dogs, not in ACU, after these GW titles were patched up by their devs. Only prominent in Dying Light but that turns out to be devs never optimizing for AMD in the first place. Once they patched up with 1.5+, extending view range no longer crippled AMD performance.

GTA V also doesn't show any additional AMD driver overhead. Rockstar isn't awful dev like Ubifail.

 

[AtenRa]

The 570's cpu usage, is like 5% higher than the 5850, BUT the 570 system is spitting out 50fps while the 5850 is giving 30fps.

Isn't that a clear indication of AMD's driver cpu overhead?

HD5850 is slower than GTX570 which has the same performance of GTX480.

In any case I will test the 7950 on the Q9550 tomorrow, with the same settings as the 5850 and if it produces worse results than the 570, I will be greatly disappointed.

Yes that will be the correct way, but the game just released, new drivers and patches will change the performance soon. It will be helpful to bench the same hardware with new drivers when they arrive to see the difference in performance and if they will fix the cpu overhead.

 

[SPBHM]

HD5850 is slower than GTX570 which has the same performance of GTX480.

I think he knows that, but his point is showing the 570 with higher GPU load while delivering higher framerate, pointing to another bottleneck for the 5850, not just the expected GPU speed difference

(also the 5850 should be closer to the 570 than the stock results for both cards, considering the higher level of OC)

 

[Erenhardt]

My friend Erenhart, look closely at the 570/5850 graphs, at the part where the mouse pointer is. This is the cpu limited part.

The 570's cpu usage, is like 5% higher than the 5850, BUT the 570 system is spitting out 50fps while the 5850 is giving 30fps.

Isn't that a clear indication of AMD's driver cpu overhead?

In any case I will test the 7950 on the Q9550 tomorrow, with the same settings as the 5850 and if it produces worse results than the 570, I will be greatly disappointed.

Actually, there is something going on.
After a large section of 100% GPU load there is a spiky section where 5850 is lower utilized than 570.
Is it driver overhead bottleneck? 5850 have more free CPU resources left than 570, so there is that... I don't know, it is some kind of bottleneck.

 

[AtenRa]

I think he knows that, but his point is showing the 570 with higher GPU load while delivering higher framerate, pointing to another bottleneck for the 5850, not just the expected GPU speed difference

(also the 5850 should be closer to the 570 than the stock results for both cards, considering the higher level of OC)

Where are the clocks of the GPUs listed ??

Also, he said that because HD5850 produces lower fps its a sign of CPU overhead. That is not correct, the HD5850 is slower than the GTX570 in general. If the 7970 with the same CPU is slower/equal to GTX570 then yes we can say that is CPU overhead and the driver needs to be fixed.

 

[SPBHM]

Where are the clocks of the GPUs listed ??

Also, he said that because HD5850 produces lower fps its a sign of CPU overhead. That is not correct, the HD5850 is slower than the GTX570 in general. If the 7970 with the same CPU is slower/equal to GTX570 then yes we can say that is CPU overhead and the driver needs to be fixed.

the graphs he posted clearly shows a lower GPU usage by the 5850 while producing lower frame rate with the same CPU, he is not simply claiming that based on the frame rate.

as for the clocks
Grand Theft Auto V 1920X1080 v.high(-) GTX 570 @850Mhz Q9550 @4GHz - 56fps

Grand Theft Auto V 1920X1080 v.high(-) 5850 @950Mhz Q9550 @4GHz - 38fps

the default clock for the 5850 is lower than the 570 and it's running with 100Mhz more, certainly not enough to match the 570, but it should be a little closer.

 

[AtenRa]

the graphs he posted clearly shows a lower GPU usage by the 5850 while producing lower frame rate with the same CPU, he is not simply claiming that based on the frame rate.

as for the clocks
Grand Theft Auto V 1920X1080 v.high(-) GTX 570 @850Mhz Q9550 @4GHz - 56fps

Grand Theft Auto V 1920X1080 v.high(-) 5850 @950Mhz Q9550 @4GHz - 38fps

the default clock for the 5850 is lower than the 570 and it's running with 100Mhz more, certainly not enough to match the 570, but it should be a little closer.

Well there is also the memory deficit in play so we are not sure if it is a CPU overhead in play here.

 

[Silverforce11]

RS's link to Sweclockers actually already clear up the issue, go have a look for more details.

Rockstar is a good dev, they optimized for all hardware.

 

[RussianSensation]

I think he knows that, but his point is showing the 570 with higher GPU load while delivering higher framerate, pointing to another bottleneck for the 5850, not just the expected GPU speed difference

It's not that simple. The same thing happens with a GTX980 vs. R9 290X in the graphs I posted, but when you look 'inside the second', R9 290X is smoother despite lower CPU usage and lower FPS count. Therefore, higher CPU and GPU usage alone are not indication that the game is running better on that setup to begin with.

This graph below also shows that you often need to look at how the FPS is delivered. The Titan X is only 14% faster than GTX980 SLI but it's far smoother in the actual frame time delivery. Alternatively, R9 295X2 pulls a big lead over the Titan X and crushes 980 SLI but its the VRAM bottleneck means widely fluctuating performance.

That's why all those sites that promoted FCAT testing for years when HD7000 CF suffered but now completely ignore FCAT testing are hypocrites.

Also, if 5850's 1GB is a severe bottleneck (i.e., draw distance), we should see lower GPU and CPU usage since the system constantly runs out of VRAM. Don't forget that HD5800 series had awful tessellation and GTA V uses tessellation. If the geometry engines in the 5850 are the primary bottleneck, overall GPU and thus automatically the CPU usage, would dive because the rest of the GPU is waiting.

When tessellation is used in a DX11 game, HD5870's performance is done.

When tessellation + 1GB of VRAM limits are reached, 5870 is toast.

We can no longer just compare HD5000 to GTX750Ti/GTX500 series and ignore the 1GB VRAM and tessellation bottleneck. Here is why:

"The first comparison at 1080p favored the HD 6970 over the HD 5870 by 36%, and at 2560x1600 this margin was blown out to 76%, which is a much larger gap than we recorded in our December 2010 review. Back then the 6970 was an average 24% faster than the 5870 based on our 1920x1200 results across 14 games." ~ TechSpot

While HD6970 was barely faster than HD5870 at launch, today, the combination of VRAM and tessellation bottleneck mean that the 5870 is behind the 6970 by 36-76%, something that cannot be explained by their shader, textures, pixel fill-rate or GPU clock speed performance alone. 5850 is approaching near 6 years old. Knowing these 2 major bottlenecks for Cypress, the fact that it can even run a 2015 game at 1080P is already shocking.

A comparison of Q9550 @ 4.0Ghz and 7950 vs. 570 would be more revealing since it removes the VRAM and tessellation bottlenecks of 5850 and will give us a better idea of CPU driver overhead.

 

[psolord]

Guys I am really sorry for throwing you off track. No, actually I threw myself off track.

I did test the 7950 on the Q9550 with the 15,4 Catalyst, using the 5850 settings.

The 7950 did a lot better. Actually it did as good as the 570 did.

So it was not the driver that was keeping the framerate low.

Also I did a test with the 7950 using the GTX 970 settings, but with 720p resolutuion, in order to avoid gpu limits and the results were not so much in favor of Nvidia.

I will post findings with pictures and videos soon.

For now, I was mistaken plain and simple.

 

[MBrown]

Does this game not do well with GPU overclocks?

 

[destrekor]

So, I seem to be discovered that this game does not like my Crossfire setup.

First, Benchmark results with Crossfire:

Frames Per Second (Higher is better) Min, Max, Avg
Pass 0, 17.945349, 128.546494, 59.244549
Pass 1, 25.656855, 113.578568, 72.068260
Pass 2, 41.223614, 126.239677, 67.279251
Pass 3, 42.192184, 121.906082, 76.448212
Pass 4, 5.207225, 167.677811, 79.364113

Time in milliseconds(ms). (Lower is better). Min, Max, Avg
Pass 0, 7.779286, 55.724747, 16.879190
Pass 1, 8.804478, 38.975941, 13.875734
Pass 2, 7.921440, 24.257940, 14.863423
Pass 3, 8.203036, 23.701073, 13.080750
Pass 4, 5.963819, 192.040848, 12.600153



Below, results with Crossfire disabled:

Frames Per Second (Higher is better) Min, Max, Avg
Pass 0, 19.048677, 126.570992, 56.625408
Pass 1, 20.567129, 85.178345, 48.534981
Pass 2, 36.848415, 131.394409, 45.783569
Pass 3, 43.002304, 127.793747, 57.753010
Pass 4, 26.428862, 104.725975, 58.350861

Time in milliseconds(ms). (Lower is better). Min, Max, Avg
Pass 0, 7.900704, 52.497082, 17.659916
Pass 1, 11.740073, 48.621273, 20.603697
Pass 2, 7.610674, 27.138208, 21.841896
Pass 3, 7.825109, 23.254568, 17.315115
Pass 4, 9.548730, 37.837421, 17.137709


With CF disabled, the actual experience was quite simply, a world apart, it was immensely superior in feel. So smooth - a framerate cap or even vsync would have made it perfect. With CF enabled, a cap or vsync does nothing. It's a stuttering mess.

I've done a clean install of the Catalyst 15.4 driver, and I've even begun playing around with customizing the application profile in CCC. I tried forcing the Crossfire mode to "Optimize 1x1" and then "AFR Friendly" - no improvement. 1x1 was actually far worse in FPS, though equal in relative lack of smoothness [to the standard CF profile].


Hardware:
2x290X Lightnings in CF (at 1150/1500)
2600K @ 4.4GHz
16GB DDR3-1600
game on SSD (1TB 850 Evo)



edit:

Just to make sure I wasn't losing my mind and ended up on the wrong settings, I tested HPET on in BIOS and set it on in Windows as well. Recently, I determined HPET was the cause of issues since making the move to AMD cards (it's been forever since I last played with settings, but I swear on my 560 Ti's I settled on HPET On/On for smoothness). Games I have tested with HPET Off/Off have shown good smoothness. Unfortunately, this did not fix the stuttering mess that this is with Crossfire enabled.

 

[sandorski]

Looks like I can put my 5870 to use for a bit. Was thinking I'd need to upgrade to a 280 first, but will get the game and upgrade later. Maybe the 290 will get down to $200CDN($150-60US) on a clearance sale. I won't hold my breath, but maybe just maybe...

 

[SPBHM]

Guys I am really sorry for throwing you off track. No, actually I threw myself off track.

I did test the 7950 on the Q9550 with the 15,4 Catalyst, using the 5850 settings.

The 7950 did a lot better. Actually it did as good as the 570 did.

So it was not the driver that was keeping the framerate low.

Also I did a test with the 7950 using the GTX 970 settings, but with 720p resolutuion, in order to avoid gpu limits and the results were not so much in favor of Nvidia.

I will post findings with pictures and videos soon.

For now, I was mistaken plain and simple.

7950 as good as a 570 is also very disappointing,
but I guess the 5850 is just suffering more, less vram, and more "abandoned" architecture; Fermi is even getting DX12 support,

 

[escrow4]

My 780 Ti @ 1215MHz @ 1200p is enough to max out all the standard settings (reflections and post are Ultra), exception being Grass at Very High and all AA except FXAA is disabled. Advanced graphics sliders are pushed up to 60% and High detail flying is enabled, rest is disabled (out of VRAM for High Resolution Shadows). 70+ in Los Santos, 55FPS or so in some areas outside the city, grass areas dips to 30FPS (not all of them). Perfectly happy with that as 80% of the settings are maxed. This is well optimized (except underwater which tanks to 20FPS).

 

[Silverforce11]

7950 as good as a 570 is also very disappointing,
but I guess the 5850 is just suffering more, less vram, and more "abandoned" architecture; Fermi is even getting DX12 support,

It's too bad he didn't have a 460 to compare it to his 5850, so you can use his data to conclude such things that you just did.

 

[SPBHM]

It's too bad he didn't have a 460 to compare it to his 5850, so you can use his data to conclude such things that you just did.

I didn't make any conclusions I'm just open to possibilities, as I said, the best way to compare with his hardware and this game would be the 970 underclocked vs 7950, with both GTX 570 vs 5850 or GTX 570 vs 7950 there is a significant difference in vram for this game,

as for the DX11 overhead on GTA V, I'm still seeing a problematic situation here
https://youtu.be/9pxeF08qmtg?t=1m50s

once the in game driving starts the 280 is being outperformed by the 750 TI when paired with an i3.

 

[Silverforce11]

I didn't make any conclusions I'm just open to possibilities, as I said, the best way to compare with his hardware and this game would be the 970 underclocked vs 7950, with both GTX 570 vs 5850 or GTX 570 vs 7950 there is a significant difference in vram for this game,

as for the DX11 overhead on GTA V, I'm still seeing a problematic situation here
https://youtu.be/9pxeF08qmtg?t=1m50s

once the in game driving starts the 280 is being outperformed by the 750 TI when paired with an i3.

I think its well known that AMD GPUs doesn't work well with dual cores (or fake quads like i3s 2c/4t). Benchmarks from other games have shown a major drop in performance on core-disabled tests.

The question here is given quad-core, is there more CPU bottlenecks on AMD? The answer from professional reviewers are no.

Also if I can quote @psolord:

"Also I did a test with the 7950 using the GTX 970 settings, but with 720p resolutuion, in order to avoid gpu limits and the results were not so much in favor of Nvidia."

So far in in every reviewer example of worse AMD driver overhead, it's been in GW titles that were never optimized for AMD during development. ACU, Watch Dogs and especially Dying Light, all of them ran poorly and showed major CPU bottlenecks UNTIL their developers finally release an AMD optimization patch and performance improves massively.

The only game that still to date show AMD's DX11 driver is lacking is Civ 5, due to specific usage of MT rendering which is disabled on AMD.

 

[RussianSensation]

once the in game driving starts the 280 is being outperformed by the 750 TI when paired with an i3.

This isn't that surprising; what's surprising is how in 2015 people still recommend and worse keep buying i3s for gaming, while trying to save $50 over the course of 4-5 years of CPU platform ownership and not just man up and get an i5! In today's day and age when Intel's CPUs last for half a decade, it absolutely does not make sense to get an i3 for gaming but people still buy them. Like seriously if you can't afford to spend $50-60 extra over 5 years of ownership, how in the world these people gaming on modern GPUs and buying $50-60 AAA games on day 1? This in-congruency is impossible to explain. The i3 has always been made for media consumption and office work, that's it! And it shows because it's crap for gaming and has been for decades.

1920x1080, 2x MSAA, Very High, 100% distance scaling
i3 4130
R9 290 = 30 fps minimums :thumbsdown:
GTX780Ti = 40 fps minimums :thumbsdown:

i7 4790K
R9 290 = 54 fps minimums :thumbsup:
GTX780Ti = 57 fps minimums :thumbsup:

With R9 280
1920x1080, FXAA, High, 50% distance scaling
i3 4130 = 34 fps minimums :thumbsdown:
i7 4790K = 78 fps minimums :wub:
http://www.eurogamer.net/articles/digitalfoundry-2015-grand-theft-auto-5-pc-performance

I bet if you take a 2008 Core i7 920 @ 4.2Ghz, it will beat the fastest i3 in modern games. That's how bad it is.

Moral of the story is even if you throw a GTX780Ti on an i3, it's going to give you sub-standard performance in GTA V where you are not getting anywhere close to the potential graphics card's potential. That's why I will not recommend an i3 for games as I've seen it bomb way too many times and an i3 means you can't just add future GPUs to the rig since you are almost always CPU bottlenecked. Maybe if in the future Intel makes an i3 a quad-core CPU, I'll change my opinion about it.

 

[SPBHM]

I think its well known that AMD GPUs doesn't work well with dual cores. Benchmarks from other games have shown a major drop in performance on core-disabled tests.

The question here is given quad-core, is there more CPU bottlenecks on AMD? The answer from professional reviewers are no.

Also if I can quote @psolord:

"Also I did a test with the 7950 using the GTX 970 settings, but with 720p resolutuion, in order to avoid gpu limits and the results were not so much in favor of Nvidia."

the i3 is a 4 threads CPU, not just a dual core like the G3258, and this changes things drastically for the worst cases at least (minimum)
http://pclab.pl/zdjecia/artykuly/chaostheory/2015/04/gtav/charts5/gtav_vhigh_cpu.png

what professional reviewers? I would like to see this test being repeated with slow quad cores like the 860K and both these cards, I suspect the result would be similar or worse, also the psolord tests and comments are very inconclusive, he tested a 7950, a 570 and 5850 with OCs and a 4GHZ Q9550 with different settings.

the Eurogamer test is for the moment the best reference I can find for this,

This isn't that surprising; what's surprising is how in 2015 people still recommend and worse keep buying i3s for gaming, while trying to save $50 over the course of 4-5 years of CPU platform ownership and not just man up and get an i5!

the video I posted shows a relevant combination, i3 + 750 TI, compared to a combination which is a common recommendation (keep the cheap CPU invest some extra on the GPU) i3 + 280

with the second, most expensive, delivering worse results during gameplay because of the efficiency of the nvidia drivers/optimization

 

[TheELF]

This isn't that surprising; what's surprising is how in 2015 people still recommend and worse keep buying i3s for gaming, while trying to save $50 over the course of 4-5 years of CPU platform ownership and not just man up and get an i5!

Seriously you can't say its $50 more and then go and compare it to a cpu that is $200 more...

 

[Silverforce11]

the i3 is a 4 threads CPU

Pretty sure every single poster here knows what an i3 is, a dual core with HT enabled. Not a real quad like i5 onwards.

I've seen AMD GPUs tank on dual cores and i3 in the past from multiple reviews in quite a few games, its a common behavior with their GPUs, it needs a real quad core to fully drive it.

That could be describe as a driver overhead, related to core count.

In your 750ti vs 280 example, the 280 is a much stronger GPU so it shines better when driven by a stronger CPU, one that isn't a 2 core.

You can look at computerbase.de tests, even in ACU before it was patched with AMD optimizations, as long as its a quad core driving AMD GPUs its fine even if downclocked to 2ghz.

 

[Silverforce11]

the video I posted shows a relevant combination, i3 + 750 TI, compared to a combination which is a common recommendation (keep the cheap CPU invest some extra on the GPU) i3 + 280

with the second, most expensive, delivering worse results during gameplay because of the efficiency of the nvidia drivers/optimization

You can recommend gamers to buy i3 + 750ti for their rig.

I'll recommend gamers to go for at least an i5 + whatever.

Amazing SLI 980 performance on the i3. That's awesome driver optimizations right there.

 

[SPBHM]

You can recommend gamers to buy i3 + 750ti for their rig.

I'll recommend gamers to go for at least an i5 + whatever.

because they cost exactly the same, right?
I will recommend a 750 Ti for a slow CPU, as I will recommend an i5 for a 280 and faster VGAs.

Amazing SLI 980 performance on the i3. That's awesome driver optimizations right there.

where is the 290X OC CF result with the i3 for us to compare? if you don't have them this looks 110% irrelevant for the discussion.

 

[casiofx]

because they cost exactly the same, right?
where is the 290X OC CF result with the i3 for us to compare? if you don't have them this looks 110% irrelevant for the discussion.

I would be very interested to know if anyone coupled i3 and CF 290 in their system