[Gamegpu] GTA V PC Benchmark

[psolord]

I just finished my custom pc gameplay benchmarks for anyone that is interested.The benchmark contains the first 15 minutes of the game, including most real time cutscenes, hence it is mainly focused on the GPU.

The end part is the Franklin and Lamar mission, which tends to shift the cpu/gpu load balance. I should have thought to extract a different benchmark number on that part, but I can still show you what I found out.

For starters, GTX 970 on 2500k,860 and Q9550, maxed settings with no MSAA (spicy wallpaper alert on all links)

Grand Theft Auto V 1920x1080 v.high GTX 970 @1.5Ghz Core i5 2500k @4.8GHz - 97fps

Grand Theft Auto V 1920X1080 v.high GTX 970 @1.5Ghz CORE i7-860 @4Ghz - 92fps

Grand Theft Auto V 1920X1080 V.High GTX 970 @1.5Ghz Q9550 @4GHz - 82fps

Secondly

GTX 970 vs 7950, both on the 2500k, again maxed settings with no MSAA (first below video is the same as the first above)

Grand Theft Auto V 1920x1080 v.high GTX 970 @1.5Ghz Core i5 2500k @4.8GHz - 97fps


Grand Theft Auto V 1920X1080 v.high 7950 @1.1Ghz CORE i5-2500K @4.8GHz - 62fps

and last but not least

GTX 570 vs 5850 on the Q9550, quite high settings (1480MB vram needed), no MSAA, 570 has one degree higher tesselation setting. This discrepancy occurred because I figured the 5850 would manage the same settings and while it did for the most part, the higher tesselation made it stutter a lot, so I had to turn it down a notch.

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

Now the interesting part comes in exactly this last test, because it has again to do with AMD's driver higher cpu needs.

Here are two captures from their respective videos, that show the framerate, along with the cpu usage.



You will see that the q9550+570 combination can produce ~50fps, while the 5850 is even more severly cpu limited and the system can only achieve ~30fps.

Here is a side by side comparison, with the two runs at the same location. You will need a fast internet connection and you may want to increase to 720p for both windows, to see the OSD better.

The same thing happens in the 970 vs 7950 on the 2500k.




For about the same cpu load, the 970 is producing 65+fps while the 7950 less than 50fps.

I think it will be best to do a q9550+7950 test with the 5850 settings, to see if this 30fps limit will also manifest in this configuration.


Be aware of what desktop background you post screenshots of. Nudity, even partial, isn't allowed.

-Elfear

Sorry about that, won't happen again.

-Psolord

Hello again. Sorry for the large quote. I just wanted readers to have easy access on my previous post. Spoiler tags work weird in this forum. :S

Soo..in succession to the above post and since I was fairly disappointed by the 5850's outdoors framerate of 30fps, which also showed cpu limits, I decided to do the same test, same settings, same system on the 7950 this time.

Grand Theft Auto V 1920X1080 V.High(-) 7950 @1Ghz Q9550 @4GHz


In all honesty, since the gpu load was not at max and cpu load was very high, I thought I was seeing an AMD driver overhead issue. However, I was mistaken and I recall my previous assumption. After doing benchmarks all these years, I should have known better.


So here are the q9550+7950 and q9550+5850 graphs from MSI AB, of course on the same system.



Also side by side comparison, of the 7950 vs the q9550. Both systems do not go to max gpu load, while having very high cpu load, but the end result is totally different.


What would be much more enlightening, is if I had a 290/290X and do the same run on the Q9550, to see if I would get any more performance, or the same.


Since I do not though, I did something different.


I run the same test on the 7950, using the same settings I did on the GTX 970, in 1280X720p resolution, thus shifting the bottleneck to the cpu, so I could see what the difference of the two drivers would be.


Here are the graphs. Remember that 970 is running the game in 1080p and the 7950 in 720p.





So things are overall not as bad as I had initially thought.


I will revisit the subject once the 390X comes out (with hopefully not crazy prices)

 

[psolord]

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,

I see what you mean.

To put this in perspective, my 570@850Mhz scored 56fps in my benchmark run and the 7950@1Ghz with the 5850 settings, (which means one less degree of tesselation setting), only 60fps. Both on the Q9550 and also note that the gpu usage was at max for at least 2/3 of the benchmark, which was indoors/low cpu.

The weird part comes in the fact the 7950@1.1Ghz (I had to use lower gpu clock on the Q9550 because I used my second 7950, which does not OC well), paired with my 2500k and using my GTX 970 settings which are everything maxed, no MSAA, scored 62fps.

Meaning better score for only 10% more clock, with significantly higher settings.


Here are all the graphs I am talking about in this post (repeated above), so you have a picture of what I am talking about.

2500k+970@1.5Ghz+7950@1.1Ghz=maxed settings, no MSAA
Q9550+7950@1Ghz+570@850Mhz=same settings with each other (I call them vhigh minus), with one degree higher tesselation setting on the 570



These settings the 570 cannot even dream of, while on the 7950 the game is playable.

Actually that's why I am making all these videos, because sterilized benchmark results can through people off track. Even so, I myself jumped to conclusions I shouldn't have.

Also note that a 570 with higher setting than it can handle, can really bug out.


GTA V this is what could happen if you go over your vram limits

 

[Silverforce11]

The take home message is that a strong GPU needs a strong CPU to drive it, particular for this game.

Computerbase.de did a thorough test with 2 cores, 2 cores + HT (4 threads), 4 cores, 4 cores + HT. They show as long as its a real 4 core, it will keep a 980/R290X bottlenecked. 2 cores got massively worse performance for both 980/R290X. 2 + HT (i3) improves it a lot but still below 4 cores and they noted, the game stutters with 2 or 2 +HT, so 4 cores is minimum.

@psolord
What you showed us with the Q9550 result at 4Ghz is that you are running into the CPU bottleneck, it appears to be at ~60 fps which is very good for such an old CPU. Take a look at the GameGPU.Ru charts for modern CPUs, you see that many of them struggle to be 60fps+ with only newer i5/7 Haswell capable.

 

[psolord]

Just to be clear, I specifically mentioned in my first big post (the quoted message on top of this page) that

I just finished my custom pc gameplay benchmarks for anyone that is interested.The benchmark contains the first 15 minutes of the game, including most real time cutscenes, hence it is mainly focused on the GPU.

The end part is the Franklin and Lamar mission, which tends to shift the cpu/gpu load balance. I should have thought to extract a different benchmark number on that part, but I can still show you what I found out.

When you are outdoors, in chases and stuff, in rush hour, things do tend to get quite a lot cpu limited. The Q9550 did what it could and especially if you reduce the advanced distance setting, it does even better. 40fps+ is possible, but that's it. Well at least for very high settings. This is the video I posted before, regarding this setting.

To illustrate cpu importance better, here is a side by link, of the 2500k vs the Q9550, in the outdoors Franklin and Lamar chase, on the GTX 970.

 

[Face2Face]

The take home message is that a strong GPU needs a strong CPU to drive it, particular for this game.

Computerbase.de did a thorough test with 2 cores, 2 cores + HT (4 threads), 4 cores, 4 cores + HT. They show as long as its a real 4 core, it will keep a 980/R290X bottlenecked. 2 cores got massively worse performance for both 980/R290X. 2 + HT (i3) improves it a lot but still below 4 cores and they noted, the game stutters with 2 or 2 +HT, so 4 cores is minimum.

@psolord
What you showed us with the Q9550 result at 4Ghz is that you are running into the CPU bottleneck, it appears to be at ~60 fps which is very good for such an old CPU. Take a look at the GameGPU.Ru charts for modern CPUs, you see that many of them struggle to be 60fps+ with only newer i5/7 Haswell capable.

Even my 3570K is bottlenecking my GTX 780 slightly @ 1080p. At 1440p there's a little breathing room.

https://www.youtube.com/watch?v=I61Q926Aq24

 

[TheELF]

Also note that a 570 with higher setting than it can handle, can really bug out.


GTA V this is what could happen if you go over your vram limits

You can,and have to,limit your FPS to counteract the glitches.
https://www.youtube.com/watch?v=UK0_wOKv-rU

And despite what anyone says the game is quite smooth on just two cores, you just have to stay within your systems capabilities.

 

[bepo]

Worth upgrading from R9 290 to 2x970 SLI or waiting for 980Ti? I've had to tone setting way down (1080p, med/high) to get decent frame rates online and the AMD experience leaves a lot to be desired. I want to be able to play 3440x1440 @ 60 fps average, med/high+ settings when possible.

My setup:
LG 34UM95 (3440x1440, 60Hz)
i5 3570k @ 4.2 GHz
8 GB RAM
R9 290 @ 1100 MHz

 

[xthetenth]

I'd probably wait for the single card solutions and see if the prices drop from competition. 970 SLI seems like it has way too much GPU for the memory situation, and if there's something other than performance that you're hating about AMD the real decent multi GPU solution for this generation isn't on the table.

You might want to see if that 290 unlocks into an X, that might help tide you over.

 

[Face2Face]

Worth upgrading from R9 290 to 2x970 SLI or waiting for 980Ti? I've had to tone setting way down (1080p, med/high) to get decent frame rates online and the AMD experience leaves a lot to be desired. I want to be able to play 3440x1440 @ 60 fps average, med/high+ settings when possible.

My setup:
LG 34UM95 (3440x1440, 60Hz)
i5 3570k @ 4.2 GHz
8 GB RAM
R9 290 @ 1100 MHz

How's your CPU usage?

 

[escrow4]

This reminds me why I got a 5930K. Haven't experienced any issues and am running at 3.7GHz all cores. After these patches if anything, those cores are all frequently loaded up fully.

 

[guskline]

No problems here.

 

[Silverforce11]

Worth upgrading from R9 290 to 2x970 SLI or waiting for 980Ti? I've had to tone setting way down (1080p, med/high) to get decent frame rates online and the AMD experience leaves a lot to be desired. I want to be able to play 3440x1440 @ 60 fps average, med/high+ settings when possible.

My setup:
LG 34UM95 (3440x1440, 60Hz)
i5 3570k @ 4.2 GHz
8 GB RAM
R9 290 @ 1100 MHz

You cannot be serious in expecting a single R290 to drive that resolution. Even a Titan X will struggle. You need Titan X SLI for that kind of resolution in GTA V, unless you want to turn everything down.

So a good idea is to wait and see whats coming from NV & AMD then go for a dual card upgrade for the grunt you need.

 

[destrekor]

You cannot be serious in expecting a single R290 to drive that resolution. Even a Titan X will struggle. You need Titan X SLI for that kind of resolution in GTA V, unless you want to turn everything down.

So a good idea is to wait and see whats coming from NV & AMD then go for a dual card upgrade for the grunt you need.

I've got the game running at playable settings (not constant 60fps but very close on average) with a single 290X (OC'd, 1150/1500) at 5760x1080 or thereabouts (bezel corrected).

Crossfire makes it worse - FPS numbers are higher but the minimums are also far lower, and it's a stutter-fest. I've got mostly High settings, IIRC. And a 2600K @ 4.4GHz, w/ 16GB DDR3-1600.

 

[Pandora's Box]

You cannot be serious in expecting a single R290 to drive that resolution. Even a Titan X will struggle. You need Titan X SLI for that kind of resolution in GTA V, unless you want to turn everything down.

So a good idea is to wait and see whats coming from NV & AMD then go for a dual card upgrade for the grunt you need.

Titan X does just fine at 3440x1440 without SLI. 40fps minimum at max in game settings with grass on very high and FXAA.

 

[escrow4]

Some CPU load screenshots. Heart of Los Santos needs more, RAGE engine does and can scale across more than 6 threads:









Click twice for full res.

 

[Silverforce11]

Duo cores are just dead for this game. Titan X on an i3 setup is so lolworthy.

https://www.youtube.com/watch?v=SLs-sMteggg

 

[skipsneeky2]

What are the best settings outside of AA to lower or disable without much detail or quality loss for a 970 at 1080?Just replaced my 720p t.v with a 1080 monitor.Should be here Monday but looking into making adjustments with DSR today.

So far the most obvious things are the advanced setting sliders,both are all to the left.MSAA is off and Grass is Very high.Everything else is pretty much maxed atm.

Thanks to anyone who could give some input

 

[sandorski]

Duo cores are just dead for this game. Titan X on an i3 setup is so lolworthy.

https://www.youtube.com/watch?v=SLs-sMteggg

Still Playable, but would seem to be more prudent to buy something like a 290 and better CPU instead.

 

[SPBHM]

I think it's worth posting the link to the entire article
http://www.eurogamer.net/articles/digitalfoundry-2015-the-best-pc-hardware-for-grand-theft-auto-5

the i5 is a pretty good match for the Titan X

while the i3 is good match for the 750 TI/260x

 

[RussianSensation]

Duo cores are just dead for this game. Titan X on an i3 setup is so lolworthy.

https://www.youtube.com/watch?v=SLs-sMteggg

Ya, and a lot of gamers still buy i3s and recommend them. The CPU is a dog. Interesting how i5 keeps up until the bench goes on the ground. There the i7s start to pull away and especially the minimum frames on the 5960X skyrocket over the i5 4690K. Makes me want to get 6-core Skylake-E.

 

[SPBHM]

Ya, and a lot of gamers still buy i3s and recommend them. The CPU is a dog. Interesting how i5 keeps up until the bench goes on the ground. There the i7s start to pull away and especially the minimum frames on the 5960X skyrocket over the i5 4690K. Makes me want to get 6-core Skylake-E.

people normally recommend it as a cheap CPU, to be combined with cheap VGAs like the 750 Ti, and their test shows that the i3 is adequate for this VGA, obviously a TitanX is normally only bought by i7 users.

 

[RussianSensation]

people normally recommend it as a cheap CPU, to be combined with cheap VGAs like the 750 Ti, and their test shows that the i3 is adequate for this VGA, obviously a TitanX is normally only bought by i7 users.

But what happens when in 2 years you go and upgrade that 750Ti when it's no longer good enough? Oh-oh. There goes that plan because the i3 will become a bottleneck. Once we move on to DX12, more and more games will target quad-cores as the minimum and i3 will be toast. The $60-70 saved now on an i3 will cost more long-term because you'll have to do a full system overhaul, or end up getting an i5 anyway.

 

[destrekor]

But what happens when in 2 years you go and upgrade that 750Ti when it's no longer good enough? Oh-oh. There goes that plan because the i3 will become a bottleneck. Once we move on to DX12, more and more games will target quad-cores as the minimum and i3 will be toast. The $60-70 saved now on an i3 will cost more long-term because you'll have to do a full system overhaul, or end up getting an i5 anyway.

A good thing with buying a system for an i3 - you can probably upgrade that CPU to one that is leagues better.

We get more limited in CPU upgrades when we have near or at the top end for any given chipset and/or socket, but when choosing near the bottom end, there is a whole world of options to upgrade into.

I do understand what you were intending to argue, but at the very least, it's not boxing yourself in with no room to go anywhere.

I intended to have room to grow on my Z68 board when paired with the 2600K I have (overclocked from week one). Well, as it turns out, there was zero incentive to upgrade to any newer chips because the performance differential was basically nil.

 

[escrow4]

A good thing with buying a system for an i3 - you can probably upgrade that CPU to one that is leagues better.

We get more limited in CPU upgrades when we have near or at the top end for any given chipset and/or socket, but when choosing near the bottom end, there is a whole world of options to upgrade into.

I do understand what you were intending to argue, but at the very least, it's not boxing yourself in with no room to go anywhere.

I intended to have room to grow on my Z68 board when paired with the 2600K I have (overclocked from week one). Well, as it turns out, there was zero incentive to upgrade to any newer chips because the performance differential was basically nil.

Why spend $130 now and $200 later when you can just buy a top shelf i5 now for $230? Better yet, why not just go a 4790 and forget upgrading period? You can't just slap in a new CPU like a new GPU. Plus the CPU and mobo are more or less tied. You yank one, you likely have to yank the other.

 

[TheELF]

Because first of all benchmarks, even ingame ones ,work completely different from the games itself,compare actual gameplay to make up your mind.









Secondly top tier VGA's will always(well for a long time to come) be very very expensive,if you have the money to get a I7 AND a huge VGA,sure go ahead,but don't tell people that they are forced to play at 1080p@60FPS all Ultra at the least,most people are extremely happy with 40-50 fps with medium/high settings.