WIN2k(rc 2128) Quake3 and SMP

[Asylum 0]

Yes it would have been better to do a but load of benchs with and with out smp. But time did not allow, the dual machine we built was more or less three computers slapped into one. We built it on our bar hoping night so we didn't leave it together for long, can't miss out on penny picthers. As for the relevance of the other machines. They both out preformed the dual machine with earlier generation hardware. I have talked to a few people about running duals and have come to the conclusion the something must have been a miss in oour setup. Others have had better marks with the same setup. My guess is alot of our problems or slowdown came from a bad cab14 for win2k, we had to modify a few things to get it to work, as well as force the driver install.

 

Later..

Asy.

[Dan 0]

So you people have got it working under NT4 with a Voodoo3? I can't get mine working...

[Greggy 0]

Ok people... more Q3A SMP results!

 

System: Dual 366 Celerons @ 550MHz, 256Mb RAM, Creative TNT2 Ultra @ 170MHz core/ 200MHz memory, Creative PCI128 sound, NT4 Workstation SP6a.

 

Quake 3 ver 1.09 with v-synch enabled, sound on, 32bit colour & textures, max texture detail, billinear filtering, timedemo = 1.

 

Results: R_SMP =

0 1

demo001.dm3

640x480 55.7fps 65.2fps

 

1024x768 32.2fps 31.6fps

 

demo002.dm3

640x480 53.3fps 62.8fps

 

1024x768 33.0fps 32.2fps

 

Summary

Obviously fill-rate limited at 1024x768, but at 640x480 that second CPU is making a difference. I am getting hold of a GeForce card next week and will post more results then. Of course all scores will be higher, but I'm expecting to see the 1024x768 scores pick up with SMP enabled.

 

What have others found/think about these results?

[JonRico 0]

I agree with your results entirely.

 

Its what I would expect. Still not seen any results with a GeForce in SMP so I will be very interested in your next post. Is the card a DDR?

 

I dont think you will see much SMP improvement in 1024 with an SDR card.

[Greggy 0]

Now for GeForce Q3A SMP results!

System: Dual 366 Celerons @ 550MHz, 256Mb RAM, Creative GeForce @ 120MHz core/ 190MHz memory, Creative PCI128 sound, NT4 Workstation SP6a.

 

Quake 3 ver 1.09 with v-synch enabled, sound on, 32bit colour & textures, max texture detail, billinear filtering, timedemo = 1.

 

Results: R_SMP 0 / 1

 

demo001.dm3

640x480 71.7fps / 84.8fps

 

1024x768 39.1fps / 38.8fps

 

demo002.dm3

640x480 69.0fps / 84.3fps

 

1024x768 40.1fps / 39.5fps

 

Summary

Good performance increase over the overclocked TNT2 Ultra. Still fill-rate limited at 1024x768, but at 640x480 that second CPU is making a BIG difference.

I wasn't very lucky with the Creative GeForce card I received, as it is not stable above its default core speed of 120MHz. Even 125MHz causes Q3A to spit-the-dummy BIG TIME. Memory (5ns) at 200MHz seemed to have the odd problem as well, but seeing the core is such a DOG its probably not al the memory's fault!

[JimmyK 0]

good lord I cant belive this discussion has gone on this long and nobody with some actauly knowledge has come up. good lord there is so much to say I dont think if I have enough pateince or time to say it all...

 

You or NOT fillrate limited with that geforce or the tnt2 at 1024x768...... not with ur system at least

 

with a tnt2u at 1024x768you can get around 50 fps with a geforce u can get about 80. You have to have a hi end system to meet a graphics card fillrate especially when its at 480.

 

Smp in q3 can give u a 15-80 percent performance boost using smp depending on the scene that is being rendered. Just because u got a dual 550 dont think enabling smp in q3 will give u a 1100 mhz worth of power being directed to q3a. It would be nice if it was coded like that but it jsut isnt. Check ur cpu usage in taksmanager when q3a is running in smp ull notice they are nowhere close to being maxed out.

 

The usage will get much higher in intense fire fights with lotsa models on the screen and particles.

 

3dfx cards CANNOT!! do smp because they do not have an actual opengl ICD. they got that gheto minigl stuff going on.

 

Ivv been very brief and concise, when I have more time Ill write more, I must go now.

[JimmyK 0]

http://media.agn.com/newsposts/101199/nvidia.gif

[Asylum 0]

Note to The Man, JimmyK

 

Thanks for insulting all of us on your first comment.

 

I have been playing and tweaking quake since the first quake1 test was released. I have also owned every generation graphics card other then the gforce which I plan on buying as soon as my wife says its okay. Hehehe. So lets knock off a few comments with some answers.

 

Fill rate and limits.

 

vertical res * horizontal res / fillrate gives us the rough amount of fps that can be achived per second.

 

The geforce I agree is not fillrate limited or is less fillrate limited then other cards, due to the fact that everything is handled by the gforces cpu. As for the ultra tnt2, it hits the main cpu pretty hard and is therefore limited by the cpu. Which could mean fillrate, or could mean triangle calculation, this is highly dependent on what cpu you are useing. So, dependent upon what res you are running a utnt2 at will make the difference of being fillrate limited or not.

 

Your bench marks or bench guesses, don't make much sence with out adding in a cpu speed. A utnt2 at 1024x768 can get around 50 fps on a ???, my guess is a p3 500. The geforce can probably pull the 80 on any machine, I would agree with that. As for saying that you have to have a hi end system to meet a graphics cards fillrate limit. This is true for the last generation of graphics cards, but the gforce doesn't have this problem. Due to the fact that the card is handling all the graphics. The fillrate of gforces available now is 240Mps, unless someone has released the ddr version of the geforce and I happened not to notice.

 

I don't think anyone that has posted here has said anything about two processors doubling the amount of total processing power. But as for everything else you said about smp, I would have to agree.

 

I hope you keep posting JimmyK.

[JimmyK 0]

Bro, your right I should apologize for my first comment, my mistake.

 

 

But man again there are many misinformed coomments in ur last post as well. I dont mean to be condescneding or rude at all I apologize if I am but its the truth

 

1. "The fillrate of gforces available now is 240Mps, unless someone has released the ddr version of the geforce and I happened not to notice."

http://www.digital-clips.com/Hardware/Reviews/CLGeForce256/rev02.htm

 

"Peak fill rate of 480 million bilinear filtered, multi-textured pixels per second "

 

and that is the SDR version NOT the DDR. I mean for gods sake the TNT2's fill rate is around 250-300 million bilinear filtered, multi-textured pixels per second.

 

2."The geforce I agree is not fillrate limited or is less fillrate limited then other cards, due to the fact that everything is handled by the gforces cpu. "

 

If you are referring to the onboard GPU you are wrong again. That is mainly for T&L, while T&L is implemented into all opengl coding, not to the extent that would deliver the proper usage of the geforces onboard GPU.

 

If a game was to properly use the GPU processor it would yield 5!!! times the performance of a tnt2ultra. For current games the processor is hit just as hard by the geforce as the tnt2u.

 

If you had taken the least amount of time to look at the link I gave you. YOu have noticed when the onboard GPU was being used there was a 480%!!!! increase in performance over the tnt2u. now thats phenomenol. The large increase is only there because that benchmark takes advantage of the GPU properly.

 

With current games where no "special" advantage is taken there is about a 20-80 percent increase in performance.

 

But still because of the geforces monster fillrate/quad pipeline the faster ur processor the betteer ur performance is going to be.

 

3. They arent guesses, if its being limited at 1024x768 at 50 fps (the tnt2u) that obviously means that it doest matter what cpu u have. I mean if you have ap3 600 or a athlon 750 your getting the same perofrmance because the card is limited.

 

Now my geforce number is only applicable to the athlon 700 because that is the fastest processor that has been used in geforce benchmarks. But it is still not fill rate limited with a athhy 700....

 

4. "The geforce can probably pull the 80 on any machine, I would agree with that. As for saying that you have to have a hi end system to meet a graphics cards fillrate limit. This is true for the last generation of graphics cards, but the gforce doesn't have this problem."

 

Sorry bro but thats wrong again, I partially explained that above taking about the GPU. NO GAME up to date has taken any special use of the GPU. ANd if q3a did you would be getting around 180+ fps on ur rig witha geforce.

 

I hope u understood my correction of comment #4 if not Ill repost and be more specific.

[JimmyK 0]

Once again I dont mean to be an ass and I apologize if I sounded like one. I just have 3 things in life, college, computers, and the gym, so Im a lil harsh when I think something concerning my 3 things is a little misrepresented .

 

here read some reviews.

http://www.agn.com/html/agn3d-search_display.cfm?ArticleID=18652

[Asylum 0]

Thanks for the apology.

 

I think our communication is a little off here. I don't know where you got the idea a sdr card could run 480Mps, probably from your links, which I admit, I am ignoring and shouldn't be.

 

 

In reference to the sdr card this little line here screams for some correction.

 

"Peak fill rate of 480 million bilinear filtered, multi-textured pixels per second "

 

taken from www.nvidia.com

"Four independent pixel-rendering pipelines deliver up to 480 million 8-sample fully filtered pixels per second. Guarantees highest color quality and texturing special effects at maximum frame rate"

 

The SDR card is the one that hits 240Mps, the DDR card runs twice as fast, up to 480Mps.

 

I think the real question we need to ask is how much processor power does the geforce free up off the processor. Because the freed power can then be used to push a higher fill rate. and therefore a higher resolution. Correct?

 

The other question should be how much processor does it take to increase fill rate.

 

For the most part I agree with everything you said, but I don't think your right in saying I am wrong. Even if the gpu is not being used to its full, your still freeing up processor for use with fill rate.

 

This conversation is getting interesting. I don't claim to know it all, but I know enough to be dangerous.. Hehehe.

[Asylum 0]

I would just like to add, this is the longest topic on this message board by a long shot. I am glad to see so many people interested.

 

Okay, enough of that or I am going to gag.

 

Later..

 

@sylum..

 

Thanks to JimmyK for making this topic interesting again.

[JimmyK 0]

hehe okay back to the biggest point

 

the sdr version do, do the 480 mps fill rate thats the whole thing quad pipeline, thats part of the geforce technology. I mean for crying out loud the tnt2 does more then 240 check nvidias site. Im telling you it has a fill rate of 480 mps... I dont think you really know the difference betweed sdr and ddr do ya? let me explain it just incase ur a lil misinformed.

 

the ddr is better/faster ram. At higher resolutions with SDR ram ur limited by the memory bandwidth not FILLRATE. Now the DDr ram has pretty much no actual limitations get it bud? below is just a summary of what I said above.

 

sdr = a memory bandwidth limitation

 

ddr = no real limitation

 

the problem with sdr is if u wanna play 1024 X 768 X 32 bit you are limited by memory bandwidth

 

ddr uses ddr ram thus no limitation

 

Okay yes the gpu does free up a lil processor usage but ONLY for opengl games, NO direct 3d games as of now have any kind of support for T&l. Unless the game is written natively for dx7 which has support for hardware t&l, but the game requires extra coding to use T&l with dx7.

 

Meaning only opengl games are taking a limited usage of it, the d3d games arent.

 

Oh yeah, if the onboard GPU is properly used oh man its going ot be all over. We will be able to focus mad amounts of processor cycles to AI and other good stuff like that. The fillrate is also dependant on the GPU not only the main processor. Better use of the GPU yileds much higher fillrate usage. So if you had a game that took total pure 100% gpu power all ud need is about 20-30 percent of your normal cpu to power the game and the rest would be allocated to other goodies such as AI, sound, etc.

 

hehe yeah this is by far the longest thread here

[Asylum 0]

Okay, I must have been mistaken, but I could have sworn that I had read an artcle or two the clearly stated that the sdr card ran at 240Mps. But today when I went to look for them I found this which proves your point.

 

Just recently, the topic of using Double Data Rate SDRAM or SGRAM on a GeForce 256 board was brought up with the announcement of DDR GeForce 256 boards from Creative Labs, Guillemot, and Leadtek. Double Data Rate (DDR) SDRAM/SGRAM has been around for quite some time and is theoretically a very intelligent solution that offers a low latency and a greater memory bandwidth. The way DDR SDRAM/SGRAM works is that it transfers data on both the rising and falling edges of the clock like the AGP bus, therefore doubling the amount of available memory bandwidth. This way, NVIDIA can boast that their 128-bit memory bus offers just as much bandwidth as Matrox's Dual 128-bit bus or as a 256-bit memory bus.

 

At the current memory clock speed, using DDR SDRAM on a GeForce 256 would result in an amazing 5.2GB/s of memory bandwidth, and at 200MHz (5ns chips) it would offer 6.4GB/s. This would come in handy particularly in 32-bit color rendering modes and at high resolutions when a considerable amount of memory is required.

 

Unfortunately, there is a downside to all of this. Currently, DDR SDRAM/SGRAM is very rare and it will continue to be rare for quite some time. The chances of picking up a GeForce 256 this year with DDR memory on it is next to nothing and the first signs of publicly available DDR models will probably be sometime next year. Samsung recently announced their DDR SDRAM chips but for now they are out of the reach of the mass market, making our conventional SDRAM/SGRAM (also referred to as Single Data Rate - SDR) the choice for most GeForce 256 boards.

 

Now I have a question for you. Why wouldn't more memory bandwidth give you higher fillrate. What memory are we talking about. Probably not frame buffer, and probably not system memory.

 

Later.

[Greggy 0]

Please correct me if I'm wrong (I'm sure someone will)...

 

Put it this way, my Dual Celeron 550's running Q3A are just as busy at 320x240 resolution as they are at 1024x768 and at 1280x1024. Now if 320x240 requires less CPU, THEN WHAT THE HELL ELSE ARE THEY DOING?

 

I agree that the theoretical peak fill-rate for the 120MHz GeForce (both SDR & DDR) is 480Mpixels/sec. But it will never be that high in a graphically complex game like Q3A as the GeForce simply cannot get the data out of the memory fast enough! The DDR version will be much better, but I'd say the memory speed of the GeForce DDR may still be the limiting factor for its performance.

[Asylum 0]

I would say your right your cpu is just as busy at low resolutions as it is at high resolutions. Triangle calulation should use the same amount of cpu at all resolutions, since the resolution doesn't effect how many triangles are on the screen. I don't know the direct effect, but higher resolutions require a higher fill rate and there is a link between fill rate and the processor. This is one of the topics I have been discussing with JimmyK. Hopefully he can help shead some light on the subject. I believe that in smp mode, the second processor helps with triangle calulations, or takes other calulations off of the main cpu. Therefore the main cpu has more resources for triangle calulations, better known as float point calulations.

[JimmyK 0]

yeah I got a few things to say, about hi vs low res fill rate and cpu usage but I'm in college hell week (exams) so gimme a few days fellas Ill post on the 17th, excuse me