[dan2097]

Originally Posted by digitalwanderer

Can 3dm2k1se utilize the dual aspect of the dual-opteron set-up it was run on? I didn't think 3dm2k1se supported multiple CPUs, but I ain't got any experience with multiple CPUs.

I dont think it can take advanatge of dual cpus, atleast based on the fact that the computers on the top of the orb are single cpu

[Joe DeFuria]

Originally Posted by epicstruggle

This thread over at nvnews.net should help you out joe.

http://www.nvnews.net/vbulletin/show...threadid=26167

later,
epic

Thanks, but it's not a big help.

I tried to run a few botmach benchmarks (even ones listed as either standard 2K3 or 2K4 Demo ones), and while the map launches and the correct quality settings seem to be applied, you as a player, actually spawn and just "sit there". (You could actually play if you want to, until the benchmark time limit, 30 secs (?) is reached.) IIRC, you're supposed to be in spectator mode, and it cycles through different bot views.

The standard "flyby" demos seem to work OK.

Anyone have a clue on how to properly benchmark bot matches in 2K4?

[JohnH]

Sorry if this has already been touched on but theres far to much rubbish to go through to find it, but the 7GPix/s single textured figure does bear more discussion.

Ignoring texture and depth BW (polys are flat so later should be irrelevent),
Single texture test requires 7Gx(32bit read+32bit write) = 53.4GB/s, memory config described seems to offer a peak BW of ~38GB/s.

This means one of,

1)The test was run in 16BPP mode
2)The device has a pile of embedded memory
3)The device is a tiler of some form
4)The device is using lossy compress (not really acceptable)
5)NV have come up with an effective lossless compression scheme applicable to random access.

Assuming 1 isnt the case, given NV's recent form 4 wouldn't surprise, but of course 2,3, or 5 are not impossible either....

John.

[pocketmoon_]

Originally Posted by pakotlar

can you call Tagrineth a ****. because if you do that both of us will be in complete accord.

http://www.channel4.com/ads/index1.html

[Rolf N]

Originally Posted by JohnH

3)The device is a tiler of some form

That should read "deferred renderer", not "tiler"

*runs back into the woods, giggling*

[Doomtrooper]

This article does alot of work for us already

http://www.xbitlabs.com/articles/vid...xtreme_13.html



vs.

UT2004 ( 4xAA, 4xAF ) 1600x1200 71.9fps [Botmatch]

[dizietsma]

3dmark2001 is not dual enabled

3dmark2001 is cpu limited in car and lobby but not nature or dragothic.

working out cpu capping on car and lobby but not drag and nature assuming 2.5x power of nv30/35 gives roughly 35000 points with AMD64 3400+ @2400Mhz

[Doomtrooper]

There is just far too much BS in this thread to take it seriousally...really...this does remind me of the Nv30 'leaks' of last year Bigtime.

[BetrayerX]

Originally Posted by MrGaribaldi

Just throwing out a wacky idea, but couldn't it be possible that the 9800Pro/XT could have more than one thing as it's limiting factor? And that this would (partially) explain why we're having trouble comparing the test vegetto talks about with the one x-bit has done?

There are so many unknown variables even when looking at the game alone, that saying the NV40 numbers are impossible since a 9800Pro/XT example can't explain the behaviour, is a bit strange to me. (Not to mention that that 9800pro and 9800XT has small differences between them which might in part explain some of it).

So why not try to understand what the limiting factors of the game is, before starting the fan-war that is currently building momentum?

Here is what the UT2k4 release notes says about the topic:

Also, since we don't know what settings the nv40 was run at, this might be relevant:

Anyways, my point is, even though I like a good nvidia bashing every now and again, that it might be more worthwhile first to figure out what has changed from 2k3 to 2k4 and then use that to figure out how accurate the numbers we've gotten are, instead of just attacking the guy/numbers based on the previous version of the game.

That wont make much of a difference since the botchmach main limiting factor are the AI/Physics running. Wont help much if you raise/lower settings or even if you cheat or not in drivers...to a certain extent of course, but it won't allow such huge jump in fps. Have to side with people believing this test is CPU bound.

[FUDie]

Originally Posted by dizietsma

3dmark2001 is not dual enabled

3dmark2001 is cpu limited in car and lobby but not nature or dragothic.

Dragothic is largely CPU limited on the fastest cards. Even Nature can be somewhat CPU limited.

[Joe DeFuria]

Please, someone tell me how to properly benchmarka UT2004 botmatch, and I can provide a few data points, and we can close this particular question. (Yes, I'm searching some other forums myself....)

[Miksu]

Originally Posted by Joe DeFuria

Please, someone tell me how to properly benchmarka UT2004 botmatch, and I can provide a few data points, and we can close this particular question. (Yes, I'm searching some other forums myself....)

Have you tried this one: http://www.unrealmark.com/

[pakotlar]

Originally Posted by Joe DeFuria

Please, someone tell me how to properly benchmarka UT2004 botmatch, and I can provide a few data points, and we can close this particular question. (Yes, I'm searching some other forums myself....)

pm me and I'll send you the files.

edit: applicable only to demo i believe...sorry.

[Joe DeFuria]

Originally Posted by Miksu

Have you tried this one: http://www.unrealmark.com/

Yup. No-go. It doesn't start botmatches in spectate mode, you you are just sitting there at a spawn point....

[Chalnoth]

Originally Posted by JohnH

3)The device is a tiler of some form

This wouldn't a single-texturing test with no z-buffer usage.

Edit:
By the way, you don't need to resort to those things you explained to get 7Gpix/sec out of 38GB/sec of memory bandwidth.

Here's the scenario:

1. Scene has z-buffer disabled. Each pixel requires one color write.
2. Scene uses a low-resolution texture that requires very little texture memory bandwidth.
3. V-sync is disabled, such that the contents of the framebuffer are output to the RAMDAC once every screen refresh, but the framerate is much higher.

Or, another possibility:

1. Scene has z-buffer disabled, but overdraw is large. Each pixel requires one color write.
2. Since scene has large overdraw, there are many pixels written for each one output to the monitor.

Anyway, if one takes either case to the extreme limit, it may be possible to have single-textured fillrate approach 4 bytes per pixel required bandwidth, which would be only 28GB/sec of bandwidth required for 7GPix/sec.

[Sazar]

aaargh..



what about the GPU and the information on it? this whole "cpu bound blah and athlon unlocked blah and more blah" is getting repetitive init

I am doing a write-up on athlon64 processor's... if it turns out to be any good I'll hook some of you guys with a URL and you can thrash over it...

in the interim... how bout the architectural specs on the card? anyone care to discuss

-edit-

when I buy ut2k4 next week I'll toss up some numbers... no supah-dupah gfx card but heck its an r3xx core.. gotta be worth something in an argument plus... its an a64 cpu...

[dizietsma]

Originally Posted by FUDie

Dragothic is largely CPU limited on the fastest cards. Even Nature can be somewhat CPU limited.

No it's not

FX at 3071 but faster video card

http://service.futuremark.com/compare?2k1=7382155

FX at 3140 but slower video card

http://service.futuremark.com/compare?2k1=7487438


The slower cpu is a lot quicker due to the video card.

[dizietsma]

Originally Posted by Joe DeFuria

Please, someone tell me how to properly benchmarka UT2004 botmatch, and I can provide a few data points, and we can close this particular question. (Yes, I'm searching some other forums myself....)

You could do the following

1) run FRAPs for a set period and run around shooting.
2) Change to 1024 and then 1600 and although the game will vary the average should be pretty constant.
3) Repeat 3 times to get an average of averages

If 2 gives 50 for 1024 and 50 or so for 1600 it sis probably cpu limted, if it gives 50- and then 30 it is not.

It's pretty rough test, but I am sure you could get a feel for it Joe to be able to pass judegment, might be a quicker and dirtier, but more fruitful test than scouring the Internet

[Joe DeFuria]

Originally Posted by dizietsma

You could do the following

I may try that later tonight, if no one knows.

I just don't have the time right now to sit down and actually "play" to do the FRAPS thing. (I'm watching / playing with the kids). Whereas I could run a batch file (or 10) and just let them go.

[FUDie]

Originally Posted by dizietsma

No it's not

FX at 3071 but faster video card

http://service.futuremark.com/compare?2k1=7382155

FX at 3140 but slower video card

http://service.futuremark.com/compare?2k1=7487438

The slower cpu is a lot quicker due to the video card.

What does that prove? First, Dragothic is at least somewhat CPU limited, as I said. Second, I know for a fact that Nature can be partially CPU limited as well as I got higher framerates on my 3.0 Ghz P4 vs. Athlon XP1900+. It wasn't a big difference, but it was measurabe.

-FUDie

[Sazar]

Originally Posted by dan2097

I dont think it can take advanatge of dual cpus, atleast based on the fact that the computers on the top of the orb are single cpu

-edit-

doom answered Q already

[Doomtrooper]

3Dmark 2001 is single threaded

[dizietsma]

Originally Posted by FUDie

What does that prove? First, Dragothic is at least somewhat CPU limited, as I said. Second, I know for a fact that Nature can be partially CPU limited as well as I got higher framerates on my 3.0 Ghz P4 vs. Athlon XP1900+. It wasn't a big difference, but it was measurabe.

-FUDie

Well it proves that the faster card on a slower cpu has greater fps than a slower card on a faster cpu

hence nothing is capped, the video card takes the strain.



Joe, I know what you mean, that would take time. To be honest I think it is fruitless because xbit test is not a botmatch and does show variation with video card, the unknown testers result says botmatch but all that means is the nv40 might have been pegged and therefore might have been even greater than the 70+fps which was quoted and called into question.

It might have been 100fps given a 3Ghz AMD 64 .. LOL !

I am not arguing this saying that the tester is being truthful, I am saying that the arrows thrown at him for his quotes are not justified.

Time will tell I guess.

[FUDie]

Originally Posted by dizietsma

Well it proves that the faster card on a slower cpu has greater fps than a slower card on a faster cpu

hence nothing is capped, the video card takes the strain.

Except that it looks like the slower system had faster memory. 3DMark 2001 results are heavily affected by system bandwidth.

-FUDie

[dizietsma]

You are actually quoting video card / cpu card / system scaling, which I agree with, the latest video cards get more points with the faster cpu's, as your expample with the P4 shows.

However, this is not the point under consideration, and that is cpu capping. A cpu capped systyem will not let a video card get better results no matter what video card you throw at it, ie massive overclocking of the video card or a brand new generation X video card.

I'd still like to see the nv40 card 3dmark score per game to get me a better feeling for the claims.