Mid-high end cards on low end CPUs - an issue that needs some light shed

[happy medium]

Think about the i5 2500K, its quite old

How about for guys like me with my cpu? I was gonna buy a used 7970 for 150$ but it might be better for me to buy the Nvidia equivalent.

 

[LTC8K6]

I agree , but the i3 has 2 cores and 4 threads and that seems to help a lot.

It seems to help a little in most games, a little more in a few.

The haswell pentium dual cores don't seem far behind equally clocked i3's in most games.

 

[Silverforce11]

How about for guys like me with my cpu? I was gonna buy a used 7970 for 150$ but it might be better for me to buy the Nvidia equivalent.

Depends on the games you play. If you play NV sponsored titles, go for NV GPUs. But if you play GTA V or ACU, don't bother getting anything above a 750ti performance because you will be CPU bottlenecked regardless of GPU brand. Also, Witcher 3 is very kind on CPUs, so in your case, you should spend the most $ you have budgeted on the most powerful GPU instead of worrying about "driver overhead" that's heavily present only in a few GameWorks titles.

I did have a 7850 OC by about 50% (1.2ghz) in a Q9400 playing BF3 just fine >60 fps.

 

[happy medium]

bother getting anything above a 750ti performance because you will be CPU bottlenecked

What? a gtx580 is faster than a gtx750ti. My 6870's are faster than a gtx580.
My cpu can push a gtx680/gtx770/gtx960/r9 280x/7970 as per the research I've done.

Not saying that's the route i'm going.

 

[Silverforce11]

My cpu can push a gtx680/gtx770/gtx960/r9 280x/7970 as per the research I've done.

Not saying that's the route i'm going.

Depends on the game.

i3 is bottlenecked in GTA V at ~30 min fps and ~42fps average. Not sure where an older Q series lands on there.

Also, if you think it can drive 7970 class, don't forget if you throw in a R290, you can run VSR for superior IQ and put the bottleneck on the GPU.

 

[happy medium]

i3 is bottlenecked in GTA V at ~30 min fps and ~42fps average. Not sure where an older Q series lands on there.
__________________

How about 55 fps average 26 fps min.. See 18:21 of first video.
http://youtubedoubler.com/?video1=4vnvt6KpODw&video2=IZ0EqIZmYkI

My guess is the gtx570 was holding the old q9550 @ 4ghz back a bit.

 

[TheELF]

i3 just suck for modern gaming on any GPU that is stronger than it's CPU bottleneck.

There's no magic, check out NV's "driver overhead" with Titan X with an i3.

http://www.eurogamer.net/articles/digitalfoundry-2015-the-best-pc-hardware-for-grand-theft-auto-5

GTA V, the most CPU limited game of late:
https://www.youtube.com/watch?v=SLs-sMteggg

I wouldn't say its a driver overhead issue, its simply a case of major CPU bottlenecks, unable to deliver fps potential that the stronger GPU can handle.

The 750Ti or R260X is maxed out on FPS potential even with the lowly i3, but the R270X & R280X has GPU grunt left on the table, similar to the Titan X being held back by the i3.

Pre scripted gpu benchmarks...
they have nothing to do with actual gameplay it's like showing cinebench scores and claiming that you will get the same difference in games that you will see in cinebench.

It just doesn't work like that.
Even a celeron gets 25fps on gta 5 ,with no drops, while recording so it's definetly no cpu problem,it's just not a real life benchmark.
See a celeron run gta 5


For the second part, about the i3 beeing close to the pentium,again people look at benches like cinebench and draw their conclusions from that.
Look at some real life examples, youtube is full of them,with games that actually use 4 or more threads hyperthreading works wonders giving double the fps of a same Ghz pentium.
https://www.youtube.com/watch?v=nIcVetS92ic

 

[Silverforce11]

How about 55 fps average 26 fps min.. See 18:21 of first video.
http://youtubedoubler.com/?video1=4vnvt6KpODw&video2=IZ0EqIZmYkI

My guess is the gtx570 was holding the old q9550 @ 4ghz back a bit.

Not bad, Q series OC kicking i3 butt in latest blockbuster & highly CPU limited game.

Also, epic scale open-world game, done right:

http://www.computerbase.de/2015-05/...en-vergleich/2/#diagramm-frametimes-2560-1440

Any issues with AMD + 2 cores or 2 + 2 HT is entirely game-specific and so far seems only to happen in a few NV sponsored games.

 

[SteveGrabowski]

I hate having to recommend the GTX 960 or even worse the GTX 750 Ti over the R9 280 to budget builders, but god, look at how the R9 280 tanks against the GTX 750 Ti in the GTA V benchmark in the driving section at the end when both are paired with an i3-4130.

https://www.youtube.com/watch?v=9pxeF08qmtg

 

[Silverforce11]

I hate having to recommend the GTX 960 or even worse the GTX 750 Ti over the R9 280 to budget builders, but god, look at how the R9 280 tanks against the GTX 750 Ti in the GTA V benchmark in the driving section at the end when both are paired with an i3-4130.

https://www.youtube.com/watch?v=9pxeF08qmtg

Watch the Titan X tank the same.

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

Again, moral of the story: powerful GPUs need a powerful CPU to drive it, especially in games that utilize lots of cores.

 

[bystander36]

Pre scripted gpu benchmarks...
they have nothing to do with actual gameplay it's like showing cinebench scores and claiming that you will get the same difference in games that you will see in cinebench.

It just doesn't work like that.
Even a celeron gets 25fps on gta 5 ,with no drops, while recording so it's definetly no cpu problem,it's just not a real life benchmark.
See a celeron run gta 5


For the second part, about the i3 beeing close to the pentium,again people look at benches like cinebench and draw their conclusions from that.
Look at some real life examples, youtube is full of them,with games that actually use 4 or more threads hyperthreading works wonders giving double the fps of a same Ghz pentium.
https://www.youtube.com/watch?v=nIcVetS92ic

Pre-scripted benchmarks are game play benchmarks at a repeatable location of the game. They are game play benchmarks, so I don't know how you can say they are not accurate. They may not tell the whole story, if there are other areas of the game which are easier and/or harder on your system, but that is no better than you remembering an easy spot of the game which made things look good.

 

[happy medium]

Again, moral of the story: powerful AMD GPUs need a more powerful CPU to drive it than its Nvidia counterpart (except if you pair a 1000$ Titan x with a 140$ cpu)t, especially in games that utilize lots of cores.

I fixed that for you.

 

[Silverforce11]

I fixed that for you.

You don't understand the results. Min/Avg fps of Titan X is very similar to 750ti, R260X, R270X and R280 with an i3. Think about what that means for a bit before you claim there's a CPU overhead penalty for using AMD GPUs.

If you want to prove your point, then show me a neutral (non NV sponsored) game where an i3 + AMD GPU tanks hard compared to an i3 + NV GPU (of similar performance class), because I haven't seen it. Enlighten me.

It doesn't happen in GTA V, one of the most recent modern title to really depend on CPU cores & IPC. It doesn't happen in Witcher 3, an open world game where it should stress CPUs more.

http://www.computerbase.de/2015-05/...en-vergleich/2/#diagramm-frametimes-2560-1440

I mean look at that. Dual core + R290X drop the same % as + GTX980.

Seriously how can you claim the benches by Eurogamer with GTA V, first with i3 + 750ti vs R280, and it shows the stronger R280 GPU tanking down... fair enough you jump to the conclusion, the stronger AMD GPU needs a stronger CPU to drive it. What if you put in a stronger NV GPU, would it NOT require a stronger CPU to drive it? Well, they did that, they put in the strongest NV GPU (Titan X) and it tanks just as hard.

Not only that, when they expanded their test to use an AMD GPU of similar performance to the 750ti (R260X), they get comparable performance to the 750ti...

From that, you somehow conclude AMD has more driver overhead? No logic.

 

[sm625]

This has always been a problem. You always have to scratch and claw your way through search engines to find reasonable benchmarks, like those computerbase.de ones. (Its funny how they are usually German too.) IMO every video card review should always include tests with cores disabled, HT disabled, and a couple different underclock settings. It's not like you even need to install different cpus to test this stuff, although two different cache sizes should be tested.

 

[Silverforce11]

Here's a recent article from computerbase.de where they analyzed CPU scaling on R290X & 980 in a bunch of games:

http://www.computerbase.de/2014-12/early-access-spiele-benchmarks-test/2/#abschnitt_dayz

No difference. A few games lost performance going to dual cores, but the % loss is the same for AMD/NV.

The only benches I've seen where AMD GPUs was negatively affected by CPU cores/speed are under 2 scenarios:

1. NV sponsored games.
2. AMD FX CPUs. This one is easy, the CPU arch really needs multi-threaded drivers to take advantage of the architecture, but AMD doesn't have it. Whereas on an Intel SMT CPU, even a lowly i3, AMD performs just fine compared to similar NV GPU.

So its kinda funny that AMD GPUs work poorly with AMD CPUs, but pair it with a cheap Intel, it does just fine (outside of NV games).

 

[Enigmoid]

Here's a recent article from computerbase.de where they analyzed CPU scaling on R290X & 980 in a bunch of games:

http://www.computerbase.de/2014-12/early-access-spiele-benchmarks-test/2/#abschnitt_dayz

No difference. A few games lost performance going to dual cores, but the % loss is the same for AMD/NV.

The only benches I've seen where AMD GPUs was negatively affected by CPU cores/speed are under 2 scenarios:

1. NV sponsored games.
2. AMD FX CPUs. This one is easy, the CPU arch really needs multi-threaded drivers to take advantage of the architecture, but AMD doesn't have it. Whereas on an Intel SMT CPU, even a lowly i3, AMD performs just fine compared to similar NV GPU.

So its kinda funny that AMD GPUs work poorly with AMD CPUs, but pair it with a cheap Intel, it does just fine (outside of NV games).

Assetto Corsa - Completely GPU limited.
Dayz - Nvidia does better except for 2C @ 3.5
Wreckfest - Nvidia solidly domintes
Project Cars - Nvidia dominates
Star Citizen - Slight Nvidia edge
Talos Principle - GPU limited.


Interestingly it looks like AMD does better in 2C setups, add HT or more cores and Nvidia tends to scale better (alternatively you could think of it as nvidia tanking at 2C CPUs). Nvidia also seems to have problems with HT on 4C CPUs.

You are looking at it as a function of scaling which is only part of the problem. You must also look at the absolute performance (if one driver is 30% more efficient than another and gets similar core scaling then the scaling as you change the CPU grunt remains the same but one GPU can get 30% better frame rates).

Looking at wreckfest for instance AMD exhibits better scaling than Nvidia however, the framerates are across the board higher for Nvidia. At 4C @ 2.5 ghz AMD gets 22 fps and nvidia 39.8. Moving up to 4C @ 3.5 ghz Nvidia goes to 44.6 fps while AMD improved to 29.8 fps. AMD scaled much better however it is clear that at 4C 2.5 ghz both are CPU limited and Nvidia is getting massively better FPS.

I'm not trying to get into any AMD vs. Nvidia debate, my point is that relying solely on scaling as a measure of driver efficiency is wrong.

Also note that these are all early access games when tested and may not be indicative of patched performance. Thus data from these tests is questionable.

 

[dazelord]

Interestingly it looks like AMD does better in 2C setups, add HT or more cores and Nvidia tends to scale better (alternatively you could think of it as nvidia tanking at 2C CPUs). Nvidia also seems to have problems with HT on 4C CPUs.

Yes this is confirmed by pclab.pl
http://pclab.pl/art55238-3.html

 

[Flapdrol1337]

It'd be interesting to see the results if it's properly investigated.

Also what the best setting is for these things on different cpu's.
"threaded optimisation"
"pre rendered frames"

 

[Azix]

DX 12 is worth looking forward to. I think for real driver improvements they'll need something close to a fresh start (both amd and nvidia). the current drivers are likely too big and complicated from years of fixing games for dx9 to 11.

 

[Despoiler]

Yes this is confirmed by pclab.pl
http://pclab.pl/art55238-3.html

You cannot trust any results from that site. Their numbers never look anything like anyone else's with the same exact hardware. It's well known that their benches have favored Nvidia ever since AMD denied giving them free cards to review.

 

[Innokentij]

This video sheds light on the situation I'm concerned with

https://www.youtube.com/watch?v=lQzLU4HWw2U

Some gamers on a budget will choose to spend more on a graphics card than a CPU, that makes perfect sense since even i3s don't bottleneck most games below 60fps yet. The problem is, those graphics cards are often tested on high end CPUs, which does not tell the whole story.

The current batch of AMD cards lose a big chunk of their performance if paired with a low end CPU, more than Nvidia does, even if the AMD cards are more powerful when paired with high end CPUs.

This is something reviews have not addressed. It's quite important, as if you were looking for a build under say 700, you'd probably go with an i3 and try to put 200 into the graphics, just for example. If you look at AMD card reviews, they may do better than an Nvidia one, but the problem is they'd lose more performance with the low end CPU than the Nvidia does.

This was great info, the money u save on going AMD u loose in performance if u gimp the cpu to.

 

[Headfoot]

This was great info, the money u save on going AMD u loose in performance if u gimp the cpu to.

uhhhh have you read any of the other posts in this thread dude?

 

[Innokentij]

uhhhh have you read any of the other posts in this thread dude?

Yes and i also watched how that 280 dipped down everytime the cpu struggled unlike the nvidia cards. All i needed to know, rest of u can continue bickering about whatever u want for all i care.

 

[Stuka87]

Pre-scripted benchmarks are game play benchmarks at a repeatable location of the game. They are game play benchmarks, so I don't know how you can say they are not accurate. They may not tell the whole story, if there are other areas of the game which are easier and/or harder on your system, but that is no better than you remembering an easy spot of the game which made things look good.

Pre-scripted benchmarks can be accurate for the GPU in many cases, but VERY inaccurate for the CPU. Biggest place this is an issue is online games.

Take BF4 for instance, the "scripted benchmark" shows low end CPUs are just fine! But jump on a 64 man server and suddenly your FPS tank and the CPU will be pegged. Even my 4690K @4.5GHz will run at ~85% utilization. My old Phenom II 95BE @4GHz would run at 95-98% utilization and cap my FPS in the mid 30's in DirectX mode, but allow me to get 50-60 in Mantle.

 

[Headfoot]

Yes and i also watched how that 280 dipped down everytime the cpu struggled unlike the nvidia cards. All i needed to know, rest of u can continue bickering about whatever u want for all i care.

Obviously you haven't read Silverforce11's post with the computerbase.de link which disproves your hypothesis. But please, continue ignoring evidence you don't like to hear because it contradicts your preconceived narrative. The drive by monologues are very helpful as well.