I haven't done one of these in a while: 7800GT vs 7900GT benchmarkfest @ stock & OC

P

pxc

Extremely [H]
Joined
Oct 22, 2000
Messages
33,063
System config:
Athlon X2 3800+ @ 2.5GHz (10x250, 4x HT multiplier), 1GB dual channel PC3200 @ 414MHz, detonator 84.20 (no tweaks), WinXP Pro SP2

BFG 7800GT OC 425/1050 stock (red), under-clocked to 400/1000 (blue) and overclocked to 465/1180 (green)
XFX 7900GT Extreme 520/1500 stock (cyan), under-clocked to 450/1320 (purple) and overclocked to 445/1750 (orange)

I'm most thankful for automated benchmarking utilities. This is 384 benchmarks. :eek:

Part 1: Synthetic Benchmarks, all run at default settings:

3dm063qf.png
3dm050ue.png

Nothing unexpected except that the CPU score in 3DMark05 jumps around a lot.

3dm035tk.png
aqua5jv.png


(Don't hassle me over the aquamark scale. The graph program wouldn't let me change it.)

Give me a few minutes to upload all 4 parts.
 
Part 2: Game Benchmarks 1/3:

doom33eo.png
quake44kp.png

Both graphs are of timedemos with the listed demo in High Quality game setting.

cod28pb.png
ut20045on.png

Everything is maxed out in CoD2, except no AA enabled. I use the detected settings normally and get around 45fps at 1600x1200.

UT2004 fps is an average of 3 botmatches. Only 1600x1200 4xAA is a little interesting. It's probably completely playable at 8xAA.
 
Part 3: Game Benchmarks 2/3:

fc2ve.png
fchdr9bf.png

Ubisoft Regulator demo used for both. Nothing to add.

bf29kv.png
fear2pc.png

BF2 shows where my CPU is a bottleneck (1024x768) and where the ROPs (7900GT @ 1600x1200 w/AA) are the bottleneck. FEAR just keeps scaling.
 
Part 4: Game Benchmarks 3/3, Source games:

css4qe.png

The VST really scales with memory bandwidth

hl22am.png
hl2lc6sc.png

And these scale with GPU speed. The regular HL2 fps is an average of 2 demos: HOC_anticitizen and canals level. The HL2:LC fps is a full run through of the level.

Done.
---

I wish i had a 7800GTX to see if there are any clock for clock improvements between the G70 and G71.
 
As always, primo job pxc! :D

Thank you for putting in the time and effort to make this, I'm sure it will help many users.
 
MrSlacker said:
what software did you use to bench games?
Click to expand...
Home-made batch files (to swap game registry/ini/cfg settings and launch timedemos), some from HOC, Bench'emAll for some batch tests and unfortunately I did a few by hand since there are no automated programs for a few tests and/or some required manual intervention. I don't have any GPU/memory speed switching utilities so each set of data had be to restarted for each video card clock speeds.

All the game benchmarks are just timedemos. IMO, this method of testing is valid for a 7800GT vs 7900GT comparison as a way of showing potential improvements. I used to dislike the evaluation method of [H] but i've since come around. I still don't like the lack of apples to apples settings on every test. That would cause a lot more work in what's already a manual process so I understand. Real gameplay should always be measured instead of a simple timedemo or synthetic benchmarks.

----
There is one more mistake in the 7900GT 450/1320 stock test. I forgot that the geometric clock delta is different on stock and factory overclocked cards. It's a 20MHz difference in one function of the core. It makes a 3% difference on that functionality. I'm not going to re-run the tests since the difference is so tiny. Just keep in mind that the 450/1320 figures could be up to 3% higher, but most likely just up to 1% higher.
 
pxc said:
Home-made batch files (to swap game registry/ini/cfg settings and launch timedemos), some from HOC, Bench'emAll for some batch tests and unfortunately I did a few by hand since there are no automated programs for a few tests and/or some required manual intervention. I don't have any GPU/memory speed switching utilities so each set of data had be to restarted for each video card clock speeds.

All the game benchmarks are just timedemos. IMO, this method of testing is valid for a 7800GT vs 7900GT comparison as a way of showing potential improvements. I used to dislike the evaluation method of [H] but i've since come around. I still don't like the lack of apples to apples settings on every test. That would cause a lot more work in what's already a manual process so I understand. Real gameplay should always be measured instead of a simple timedemo or synthetic benchmarks.

----
There is one more mistake in the 7900GT 450/1320 stock test. I forgot that the geometric clock delta is different on stock and factory overclocked cards. It's a 20MHz difference in one function of the core. It makes a 3% difference on that functionality. I'm not going to re-run the tests since the difference is so tiny. Just keep in mind that the 450/1320 figures could be up to 3% higher, but most likely just up to 1% higher.
Click to expand...

are u sure ur not kyle with a different account? :eek: ;)
 
Mayhs said:
are u sure ur not kyle with a different account? :eek: ;)
Click to expand...
Kyle has 2 accounts. :p

timedemos are great for lazy analysis. When i read most other reviews, i lightly skim the text, read the test set up and look at the graphs. It gives an idea of relative performance.

real gameplay fps > timedemo fps >> synthetic benchmarks

Like I said, imma convert.
 
purrrtyyy graphs. and thank you too, I am going to love my 7900GT even more when I OC it I guess :D
 
eloj said:
Doesn't BF2 have a framerate limiter?
Click to expand...
Yep, but not for a timedemo-type benchmark. Same for Doom3 and i think Quake4.
 
You must log in or register to reply here.
Back
Top