Big anomalies found in 3DMark03

[FallOutBoyTonto]

<A HREF="http://www.aceshardware.com/read.jsp?id=60000242" target="_new">Read more @ Ace's Hardware</A>

A <b>PII 350 MHz</b> equipped with a Radeon 9700 Pro <b>BEAT</b> a <b>P4 2.8 GHz</b> equipped with Radeon 9600 in <b>three out of four</b> tests. Overall final score difference between the two is <b>less than 100 3DMarks</b>; PII 350 got 3373 3DMarks, P4 2.8 GHz got 3443 3DMarks. Interesting results, any thoughts?

<A HREF="http://www.anandtech.com/mysystemrig.html?id=24106" target="_new">My System Rig</A>
<A HREF="http://service.futuremark.com/compare?2k3=535386" target="_new">3DMark03</A>

 

[TheGreatGrapeApe]

There are no Anomalies the R9600Pro IS better! No question about it!
All you people trying to malign my Card!

Ba$tardZ!

Actually I don't see the point of the article? We ALREADY knew it was VPU/GPU centric so what's the news/story?
This isn't PC-mark, SYSMARK, SETI benchmark it's a GRAPHICS CARD benchmark. I think the ANOMALY is 3Dmk01 which relies TOO much on the CPU/MEMORY/HD/ETC.

If you don't know how to use the tools don't complain about the fact that you are ignorant to their results (directed at ACE, not F.O.B.T.).

And the tests they used for comparison don't stress the graphics cards either.

No surprises here!


- You need a licence to buy a gun, but they'll sell anyone a stamp <i>(or internet account)</i> ! <A HREF="http://www.redgreen.com" target="_new"><font color=green>RED</font color=green> <font color=red>GREEN</font color=red></A> GA to SK

 

[FallOutBoyTonto]

We ALREADY knew it was VPU/GPU centric so what's the news/story? . . . No surprises here!

For me, the news is that Ace is still learning. Although I hardly ever visit his site, I thought that something like this would be common sense for a site dedicated to hardware. But apparently, it looks like he just realized it. That is what surprised me.

<A HREF="http://www.anandtech.com/mysystemrig.html?id=24106" target="_new">My System Rig</A>
<A HREF="http://service.futuremark.com/compare?2k3=535386" target="_new">3DMark03</A>

 

[phsstpok]

Futuremark has been saying all along that their 3DMark programs have been intended to benchmark video cards not systems.

However, games do need a good system AND a good video card (at least modern games do). Take the games Comanche 4, JK2, UT2003. These games are known to be system intensive. Of course, a PII 350 is not going to fair well against a P4 2.8.

Don't know about Gun Metal but I suspect this game is problably more dependent on video card and less on system performance than the other games.

Ace's could have easily verified their results. All they had to do was swap video cards. If the trends remained the same for each game/benchmark then game/benchmark was more dependent on CPU/System performance. If the trends reversed then the game was more dependent on video card.

<b>56K, slow and steady does not win the race on internet!</b>

 

[TheGreatGrapeApe]

I guess the thing for me is I'm hoping that there might be some 'mythical' benchmark that does nothing but give you a GPU/VPU + Drivers score.
It would work something like this (I use my card as the primary example but you can subsitute whatever you want, and don't FREAK-out on the figures, I'm making them UP! [just a note to Fanbois!]).

Ok VGPU-Marks will be the figure.
All sub tests have smaller components like fill rate etc. Which 'useually stays rather constant but not a big issue.


R9600Pro using CAT 3.2 running @ 400/300 1000 VGPUs
R9600Pro---"-----"-----running @ 456/300 1111 VGPUs
"---"------"-----"-------"-----@ 400/333 1090 VGPUs
"---"------"-----"-------"-----@ 456/333 1234 VGPUs
R9600Pro using CAT 3.4 running @ 400/300 999 VGPUs
"---"------"-----" 3.6 running @ 400/300 1001 VGPUS
FX5600U-R1 w/ DET44.44 running @ 350/350 987 VGPUs
FX5600U-R2 w/ DET44.44 running @ 400/400 1010 VGPUs
FX5600U-R2------"---------"----@ 450/444 1277 VGPUs
R9700Pro using CAT 3.2 running @ 325/310 1357 VGPUs
R9700Pro---"-----"--"----"-----@ 350/350 1468
R9700Pro---"-----" 3.6---"-----@ 325/310 1441 VGPUs
FX5900U using DET44.44 running @ 450/450 1470 VGPUs
R9800PRo using CAT 3.2 running @ 380/340 1482 VGPUs

I know you probably already get my point, but it wouldn't matter if it was matched with a P4 or a P2, with a 100mhz FSB or 200mhz, it would always get the same score based on card and drivers. That to me would be a GOOD graphics card test. We have far to many blunt instruments, give me a precise one. Now I know this is very unlikely (if not impossible) but at least I felt that 3Dmk03 was closer to that than any previous tool. It's not necesarrily a determinant of what will play D]|[ well or HL2 (does ANY game's benchmark REALLY do that either?), but it would give us a standard for head to head comparison of just the card and drivers themselves. I know people are looking for something else from 3Dmk bu that's what I'm looking for and I think it's not TOO bad at that if it could cut out the cheats.

Anywhooo, that's my take on it.


- You need a licence to buy a gun, but they'll sell anyone a stamp <i>(or internet account)</i> ! <A HREF="http://www.redgreen.com" target="_new"><font color=green>RED</font color=green> <font color=red>GREEN</font color=red></A> GA to SK

 

[phsstpok]

LOL! Do a large number of made up numbers give more weight to your assertion.

Actually, I feel 3DMark03 does come closer to your ideal than just about any other benchmark I've seen. I mean if a PII 350 with a 9700 Pro produces 33xx 3Dmarks (or whatever it was) while a P4 3.0 with the same card does something like 6000 (stock vid speeds) well that's not even a 2x increase in video performance compared to who knows what in system performance.

<b>56K, slow and steady does not win the race on internet!</b>

 

[TheGreatGrapeApe]

Do a large number of made up numbers give more weight to your assertion.

Yes! :tongue:

As does the Exclamation mark.

!!!!! and !


- You need a licence to buy a gun, but they'll sell anyone a stamp <i>(or internet account)</i> ! <A HREF="http://www.redgreen.com" target="_new"><font color=green>RED</font color=green> <font color=red>GREEN</font color=red></A> GA to SK