Originally posted by: Anand Lal Shimpi
Something you should take note of, look at the configuration in which the test was run (16x12, 4X AA, 8X AF); that should tell you a bit about NV35 already.
You won't hear anything from NVIDIA until the chip is ready to go, they're keen on not making the same mistakes over again. I'm not too surprised this test info got leaked, I saw it a couple of months agoalthough I would expect tighter control from NVIDIA surrounding NV35
Take care,
Anand
Woot!!! I finally got my first Boss Comment!!
GTaudiophile
Lifer
- Oct 24, 2000
- 29,767
- 33
- 81
We heard within a few months... does anyone have a better guess as to the release date for NV35? Are we looking at NVidia returning to its 6 month product cycle or will they still be plagued by difficulties and have to push the release back? (I know it's mostly conjecture at this point, but the prospect of this being true and available by the time I build my new computer this summer is getign me excited).
Makes me wonder what it's conna' COST! The NV30 is *really* pricey right now... at this rate the NV35 will be through the roof! :Q
GTaudiophile
Lifer
- Oct 24, 2000
- 29,767
- 33
- 81
I thought both ATi and nVidia were having issues with TSMC's Low-K process with that being the stumbling block to both NV35 and R400.
All the companies except for intel have had lots of problems with getting decent yields out of .13micron processes.
between ati and nvidia, i think the winner will be the one who designs a product to take the most advantage of the process and therefore get better yields. nv35 should be nvidia's second major project on .13 so they should have more experience with dealing with the limitations of the process than other competitors...but then again i'm not sure b/c its all speculation.
between ati and nvidia, i think the winner will be the one who designs a product to take the most advantage of the process and therefore get better yields. nv35 should be nvidia's second major project on .13 so they should have more experience with dealing with the limitations of the process than other competitors...but then again i'm not sure b/c its all speculation.
Might be the case, but NV35 isn't relying on Low-K; NV40 might be a different story. I guess Orton was a bit bold in his claims that the 9800pro would retain the performance crown through the end of the calender year. Certainly doesn't look too likely that we'll see a CY'03 debut for R400 (or NV40 for that matter).
Chiz
I'm not sure about DDR I (there is no memory that can run at those speeds AFAIK). Maybe Hynix but not sure.
PrinceXizor
Platinum Member
- Oct 4, 2002
- 2,188
- 99
- 91
Here's an idea for all you modder's/tinkerer's out there....someone willling to downclock their 9700 Pro and/or 9800 Pro to 250 MHZ and see what IT gets for fps in quake3 at these settings...
Also notice that no mention of what the memory clocks was set at....it only mentions that the chip clocks were equalized.
An interesting tidbit to be sure, but since NV35 is still many months away...may I point out that comparing 9800 Pro to NV35 is comparing something that exists (i.e. production) vs. something that does not exist.
But then again....its fun to speculate!
P-X
Also notice that no mention of what the memory clocks was set at....it only mentions that the chip clocks were equalized.
An interesting tidbit to be sure, but since NV35 is still many months away...may I point out that comparing 9800 Pro to NV35 is comparing something that exists (i.e. production) vs. something that does not exist.
But then again....its fun to speculate!
P-X
Spicedaddy
Platinum Member
- Apr 18, 2002
- 2,305
- 77
- 91
GTaudiophile
Lifer
- Oct 24, 2000
- 29,767
- 33
- 81
A recent interview with Orton confirmed that ATi is sticking to a 6-month cycle and that we will see something else from them this year.
GTaudiophile
Lifer
- Oct 24, 2000
- 29,767
- 33
- 81
Exactly. We'll be comparing a brand-new product in NV35 to a several-month-old product in R350 by the time NV35 hits the street. NV35 has to be a faster product. Though, unless nVidia fixes their Ansio problems, ATi will still have better image quality.
Great read, but I must be missing a page (or its in German
). I read a 2-page interview that gives a great recent history of ATi (or should I say ArtX) for those that don't know/don't remember. The only mention of R&D and product cycles seems to confirm 6-month refreshes with 18-24 month core redesigns. If that's the case, that would again point to R400 next year and another R300 refresh in the fall (R9900pro?). I'm sure we'll see something, but I doubt it'll be R400 (based on Orton's comments on the 9800pro and various internet reports).Definitely a worthwhile read though; all you ATi users will find out why their driver support was so poor up until recently.
Chiz
While I can't say for absolute certainty, all current evidence points rather strongly towards the NV30 not quite having a traditional pixel pipe setup which could be the reason for some of the performance characteristics that we have seen to date and the reason for the massive jump with the NV30.
Based on all of the evidence, and extensive discussion on the topic over at Beyond 3D, the likely pipeline configuration for the NV30 is a 4x2+4x0 setup. They have four pixel pipes that have dual TMUs and four 'partial' pixel pipes that have no TMUs(nor can they output color). This type of setup is a bit lacking compared to a true 8x1 configuration, but it should allow solid performance in older applications while allowing titles like DooM3 to utilize the additional non TMU equipped pixel pipes for stencil ops. Based on extensive testing of fillrate using non traditional testing methods(compared to the run of the mill 3DMark style tests) the NV30 seems to be completely incapable of outputting more then four color filled pixels per clock, yet it can output closer to eight when running stencil(memory considerations were taken into account, tests were run in 16bit to eliminate the obvious issues trying to run an eight pixel pipe board with a 128bit bus).
With that in mind, if they were to apply TMUs to the four pixel pipes that currently are reserved for stencil ops then they could significantly increase their performance in games that are not heavily reliant on stencil ops, in the range of 100% is reasonable under the proper circumstances with a 256bit bus.
This particular set of benches also has another caveat with it, both boards were downclocked to run @250MHZ, this wasn't the NV30@500 v the NV35@250, both boards were running 250. Also, while the NV35 will obliterate the NV30 in games like Quake3 they should perform much closer in Doom3 and games that use a like rendering system(although the NV35 should still hold an edge due to its far greater real world bandwith even if we assume no other optimizations).
Based on all of the evidence, and extensive discussion on the topic over at Beyond 3D, the likely pipeline configuration for the NV30 is a 4x2+4x0 setup. They have four pixel pipes that have dual TMUs and four 'partial' pixel pipes that have no TMUs(nor can they output color). This type of setup is a bit lacking compared to a true 8x1 configuration, but it should allow solid performance in older applications while allowing titles like DooM3 to utilize the additional non TMU equipped pixel pipes for stencil ops. Based on extensive testing of fillrate using non traditional testing methods(compared to the run of the mill 3DMark style tests) the NV30 seems to be completely incapable of outputting more then four color filled pixels per clock, yet it can output closer to eight when running stencil(memory considerations were taken into account, tests were run in 16bit to eliminate the obvious issues trying to run an eight pixel pipe board with a 128bit bus).
With that in mind, if they were to apply TMUs to the four pixel pipes that currently are reserved for stencil ops then they could significantly increase their performance in games that are not heavily reliant on stencil ops, in the range of 100% is reasonable under the proper circumstances with a 256bit bus.
This particular set of benches also has another caveat with it, both boards were downclocked to run @250MHZ, this wasn't the NV30@500 v the NV35@250, both boards were running 250. Also, while the NV35 will obliterate the NV30 in games like Quake3 they should perform much closer in Doom3 and games that use a like rendering system(although the NV35 should still hold an edge due to its far greater real world bandwith even if we assume no other optimizations).
Dr Tom posted a screenie of the 4 dawns demo claiming it was running on the next gen of the GFX.
Retail availability of the 9800 should hit next month. After the NV30 fiasco I hope that Nvidia can regain some trust in their name as the performance leader with the NV35.
IMO ATI did not go far enough with clock speeds on the 9800 Pro.
IMO ATI did not go far enough with clock speeds on the 9800 Pro.
Hi, I'm the guy that hinted nvnews to chip.de and the nv35 news and just saw the link onto this forum on nvnews.net
PrinceXizor said that nothing was said about the memory clock. In fact it is on chip.de:
"Die NV35-Karte benötigte trotz vollem Speichertakt von 500 MHz keine Extra-Kühlung für das Video-RAM"
Translation: The NV35-card didn't require a special cooling for the video-ram (=memory) while clocked at full memory-speed of 500 mhz (so 1000 effecitve 1000mhz).
about the ddr I memory at 500mhz: there is already a small amount of ddr I running at that speed, no wonder nvidia used that memory at cebit 2003.
there will be a fine performance boost. however, the first benchmark at that resolution and those settings is still irritating, because the geforce fx 5800ultra didn't perform so good at those and simliar settings anyway (compared to the radeon 9800pro), so we have to wait for final benchmarks again.
i hope the card is much faster anyway :-D
there are some rumours around, that ati will add to their radeon 9800pro ddr ram II with a 256bit bus when they release graphic cards with 256mb, but that are only roumors and I don't believe that, to difficult at the moment.
PrinceXizor said that nothing was said about the memory clock. In fact it is on chip.de:
"Die NV35-Karte benötigte trotz vollem Speichertakt von 500 MHz keine Extra-Kühlung für das Video-RAM"
Translation: The NV35-card didn't require a special cooling for the video-ram (=memory) while clocked at full memory-speed of 500 mhz (so 1000 effecitve 1000mhz).
about the ddr I memory at 500mhz: there is already a small amount of ddr I running at that speed, no wonder nvidia used that memory at cebit 2003.
there will be a fine performance boost. however, the first benchmark at that resolution and those settings is still irritating, because the geforce fx 5800ultra didn't perform so good at those and simliar settings anyway (compared to the radeon 9800pro), so we have to wait for final benchmarks again.
i hope the card is much faster anyway :-D
there are some rumours around, that ati will add to their radeon 9800pro ddr ram II with a 256bit bus when they release graphic cards with 256mb, but that are only roumors and I don't believe that, to difficult at the moment.
I hope this isnt the core that produces 120w like Nvidia's CEO said a while back...
On the plus side if it isn't 60db that is nice.
On the plus side if it isn't 60db that is nice.
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads
- Started by Tigerick
- Replies: 21K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter