4870 Xfire to 5870 work log - now with intel to AMD!

upgrading to a 5870 soon, but getting the option to test 5870 crossfire while i'm at it and see how much awesomer it is


Basically just running a small group of benchies on both sets of cards, including measuring power for my own reference (it'll help me decide if crossfire is worth it on this system later on). Because its for my own reference (and bystander curiosity) i am NOT bothering to post screenshots of test results - just numbers.

Tests are being ran on system in specs - but with a few changes to clocks.

CPU is at 4.2GHz 1.35v
Ram is at 886Mhz 5-5-5-18

4870 crossfire results:

Company Of Heroes:

DX9 8xAA

avg 160.4
max 250.7
min 76.8

DX10 8xAA

avg 117.3
max 217.5
min 65.6


3dM06

19210 total

SM 2.0
8378

3.0
10565

CPU 3714



vantage (just GPU tests ran)

1280x1024
test 1 47.34
test 2 50.44


1920x1080
test 1 34.17
test 2 35.72


power consumption:

Boot max: 353W
Desktop Idle: 290W
furmark 450W
linpack x64 388W
Furmark + linpack 510W
Bad company 2 Vsync on: 380-420W

Click to expand...

Single 5870 results:

CoH DX9:

Avg 173 fps
Max 261 fps
Min 79.6 fps

CoH DX10

Avg fps 249.8
Max fps 349.6
Min fps 106.8

^ i'm getting the impression this card is designed for DX10/11

3Dm06:

Vantage:

Power:

Boot max: 251W
Desktop Idle no video card driver: 240W
Desktop Idle: 141W
furmark: 344W
linpack x64: 238W
Furmark + linpack: 321W - odd how this was lower here, than furmark alone

Click to expand...

I'm expecting similar performace from the 4870's to the 5870 - perhaps slightly slower due to OC, perhaps the same due to CPU bottlenecks. Crossfire may use more CPU power than a single GPU, so i'm expecting it to be quite even - that said, i'm expecting massively lower idle power usage (~120W less).

As for 5870 crossfire: i expect shit results due to my CPU holding everything back. Yeah i may be able to run more AA in the few games i cant already max out, but my min FPS should stay the same due to the CPU. Playing around with MSI afterburner has shown me that my CPU tends to max out before my GPU's do.

edit: crossfire testing has been nuked. i no longer have access to the second card.

heky said:

I would say his Corsair TX750 can handle 2x HD5870

Click to expand...

with ease.

updated some results

for a triple post update: After talking to myself for a bit, i have a theory about why the 5870 is faster, despite reviews saying otherwise.


Theory goes:

CPU power is used to feed the video cards. I was CPU limited in most games, not GPU limited.

Being two GPU's, the 4870 crossfire setup must have taken more CPU power to feed it, than a single 5870, despite the overall similarities in GPU power.

Thus, by reducing that extra CPU overhead i've reduced the CPU bottlenecking for overall faster performance.

randomflip said:

So basically, 1 GPU will always have less bottleneck than 2 GPU's even if 1 is reasonably faster than 2 older GPU's combined? If so that is awesome. I'm yet to upgrade to a 5870, since by the looks of my CPU and PSU there maybe a significant bottleneck? I'm not too sure, but how do you find the improvement overall?

Are you getting an urge to upgrade to 1156/1336 to optimize the card?

In any case, congrats glad you replaced those 4870's.

Click to expand...

its just a theory, but its backed up by a trend i've seen in a few threads... a user with a mid range CPU goes and slaps in some high end card, and reports that its 'slower' with the extra GPU(s) enabled.

The latest example of this was a guy with an E8600 and TWO 5970's... the more GPU's he had, the slower it got. This would back up my theory of more GPU's eating more CPU power.

oh and i'm going AM3/Thuban by christmas... no rush yet, i'm waiting for a board with more USB 3.0


edit: pics of the old vs new

upgrading to a thuban 1090T 6 core sex god. using this thread to compare since its already got some benchies


Intel system benchies (focused on CPU/ram)

CPU is at 4.2GHz 1.35v
Ram is at 886Mhz 5-5-5-18

Superpi



Wprime:




the thuban is due 8am tomorrow morning

AMD is up and running




Wprime has detected my idle clocks, but it was at load clocks while the test was running.

I should note that atm, my AMD system is stock CPU with ram at 1066Mhz 7-7-7-21 - SPD timings since i dont have time to tighten them before work.


superpi:

Intel: 11.468
AMD: 19.757

58% of the speed of the intel at this test. This test DOES favour intel and my ram timings are shite, so this is expected.

Wprime:

32M:
Intel: 22.2
AMD: 8.55

259% the speed. considering i have 300% more cores, the ram timings could boost this to a perfect speed 300% (making me think that at stock, my new CPU is as fast as the old one OC'd)

128M:
Intel: 712.2
AMD: 266.5

267% faster - longer test is more reliable, backs up what i said above.

tested with ram at 5-5-5-18
max temp reached (reported by AMD overdrive) was 51.5C




it was slightly SLOWER this time (i guess due to AOD in the background)