GPU vs CPU bottleneck

[thilanliyan]

Originally posted by Jaredpace in the CPU forums:
http://forums.anandtech.com/me...=2230807&enterthread=y

Tested at 1680x1050 though.

 

[AzN]

GTX280 and 4870x2 of course at 1680x1050. These things have enough fillrate even with AA to be affected by CPU speeds.

This is to be expected.

When you start becoming GPU limited the CPU stops being the factor. It's one or the other.

 

[apoppin]

i got stuck O/C'ing my CPU most of tonight and i nearly had a dead p35 MB - for a long time

- Resetting the CMOS and a BIOS flash ultimately rescued it .. but it will *not* hit 4.0Ghz ... i finally settled on 3.99Ghz =P
It's rock solid .. but +1 Mhz more and Vista BSODS .. not a FSB hole either .. probably a MB issue

finally,

19395 in 3DMark06 with X2 and e8600@3.99Ghz .. up from
16400 in 3DMark06 with X2 and e8600@3.33Ghz

so the bottleneck appears to finally diminish ~3.90Ghz
- i expect my Asus x48 MB will allow me to go further - temps are good!

i do most of benching at 16x10 and 19x12

 

[nosfe]

3dmark also includes "CPU benchmarks", that's why it'll always see improvements when you increase the frequency of the CPU, try benchmarking with a real game and see there

 

[apoppin]

Originally posted by: nosfe
3dmark also includes "CPU benchmarks", that's why it'll always see improvements when you increase the frequency of the CPU, try benchmarking with a real game and see there

in the same system 3DMark is a valuable tool

games have the approximately the same scaling from 3.33Ghz to 3.99Ghz; nothing else changing

look at the graphics score:

SM2.0 from 8090 to 8541;FPS 57.296/59.737 to 69.02/73.32

SM3.0 from 10492 to 10828: FPS 121.15/88.68 to 123.49/93.08

The CPU score increased from 3018 to 3648

so it is pretty clear that the CPU was also holding back the X2 @ 3.33Ghz

i am benchmarking ten games and they all show a similar FPS improvement by just increasing the CPU speed.

 

[nosfe]

i'm not saying that 3dmark is bad, only that you should also benchmark some of the games that you actually play to see if there's a real world difference also, a res of 16x is kinda very low for a 3 4870 so its most likely a CPU bottleneck at that resolution but again, it depends on the games, some use more CPU(half-life2 anyone?), some use less

 

[apoppin]

Originally posted by: nosfe
i'm not saying that 3dmark is bad, only that you should also benchmark some of the games that you actually play to see if there's a real world difference also, a res of 16x is kinda very low for a 3 4870 so its most likely a CPU bottleneck at that resolution but again, it depends on the games, some use more CPU(half-life2 anyone?), some use less

You are not paying attention are you?

i am benching with 10 games and they all show some increase just from OC'ing my e8600 with X2
- so it *formerly*, was a bottleneck at 3.33G

3DMark was just a convenient example and i DID include the FPS increase in the 4 game benches it uses

i am ONLY using my X2 at the moment .. and "diminishing returns" on OC'ing appears to occur around 3.9Ghz.

When i bench with X3, i expect to OC even further on my x48 MB

ediT:

FEAR at 19x12 .. everything maxed and with 16xAF including Soft Shadows ON
[min/av/max - FPS]
72/140/436 with x2/e8600@3.33Ghz
85/146/488 with x2/e8600@3.99Ghz

HL2's LC - Maxed at 19x12 - 4xAA/16xAF
214.48 av FPS @ 3.33Ghz
230.42 av FPS @ 3.99Ghz

 

[nosfe]

that's not that big of a difference, i mean, a 20% overclock on the processor gives +4% frames in fear and +7% frames in hl2; i'm taking average frames into account as they tend to be the most representative when working with percentages

 

[apoppin]

Originally posted by: nosfe
that's not that big of a difference, i mean, a 20% overclock on the processor gives +4% frames in fear and +7% frames in hl2; i'm taking average frames into account as they tend to be the most representative when working with percentages

i don't worry about average - neither HL2 nor FEAR even requires an X2

Minimum FPS are most important to me

When you OC the CPU, you eliminate the bottlenecking on the GPU
-it is a fact that the GPU makes the most difference

however in some games - like Crysis/Warhead and even CoJ - games where the min FPS is LOW [and largely determined by the CPU's sped] - even a 10% increase on the minimum makes a big difference to playability

 

[scheibler1]

I'de be interested in seeing minimum frame rates to b/c those are the most important. What about Company of Heroes, Company Of Heroes Opposing Fronts, World In Conflict, Crysis, Grid, etc @ 1920x1200

 

[apoppin]

Originally posted by: scheibler1
I'de be interested in seeing minimum frame rates to b/c those are the most important. What about Company of Heroes, Company Of Heroes Opposing Fronts, World In Conflict, Crysis, Grid, etc @ 1920x1200

i have Crysis

i am benching tonight and will update you; all i have now are my completed benches at 3.33Ghz
- Besides, Crysis, FEAR and HL2, i am using ET-QW, UT3, Stalker, CoJ, LP, CoD4, and maybe Clear Sky and Warhead

i guess i like different games than you do

 

[scheibler1]

yea. If your interested in running those benchmarks and don't want to play the games COH and WiC both have the benchmarking utilities in the demo's

 

[Tempered81]

Originally posted by: scheibler1
I'de be interested in seeing minimum frame rates to b/c those are the most important. What about Company of Heroes, Company Of Heroes Opposing Fronts, World In Conflict, Crysis, Grid, etc @ 1920x1200

crysis:
http://www.pcgameshardware.com...ticle_id=647744&page=1

WIC:
http://www.pcgameshardware.com...ticle_id=647744&page=1

 

[apoppin]

Originally posted by: scheibler1
yea. If your interested in running those benchmarks and don't want to play the games COH and WiC both have the benchmarking utilities in the demo's

i am still working on a few more including Crysis .. i might install WiC

4870x2 with e8600, first at stock 3.33Ghz then OC'd to 3.99Ghz


CoJ @ 19x12 Fully maxed DX10 bench inc 4xAA/16xAF
x2/e8600@3.33Ghz - 20.2/116.6/53.8
x2/e8600@3.99Ghz - 28.8/128.2/65.4


LP @ 19x12 Fully maxed DX10 bench inc 4xAA/16xAF
Snow
x2/e8600@3.33Ghz - 45.3/23/64
x2/e8600@3.99Ghz - 46.6/24/64

Cave
x2/e8600@3.33Ghz - 57.6/35/71
x2/e8600@3.99Ghz - 61.6/41/71

HL2/LC @ 19x12 - maxed DX9 in 4xAA/16xAF
x2/e8600@3.33Ghz - 217.69/105/301
x2/e8600@3.99Ghz - 236.50/109/301


FEAR: Persius Mandate @ 19x12 - maxed DX9 inc 16xAF
with SS on - no/AA
x2/e8600@3.33Ghz - 72/140/436
x2/e8600@3.99Ghz - 84/145/498

SS off/4xAA
x2/e8600@3.33Ghz - 36/183/659
x2/e8600@3.99Ghz - 41/192/667