LOL. Cut the crap. If u have something to share then get on to it. If not, then stop BS-ing. Being controversial is nothing wrong with, as long as you have some fundament. Your propositions are based on nothing.
NVIDIA GT200 Rumours & Speculation Thread
- Thread starter Arun
- Start date
- Status
LOL. Cut the crap. If u have something to share then get on to it. If not, then stop BS-ing. Being controversial is nothing wrong with, as long as you have some fundament. Your propositions are based on nothing.
Editor's day pushed back to May 22nd - http://www.theinquirer.net/gb/inquirer/news/2008/05/03/nvidia-reschedules-editors-day
Jawed
Legend
It pretty much does - except that you still need a working driver of some kind on the PC.I thought the general idea of CUDA was to bypass DirectX entirely... ?
But D3D itself is going to integrate general computation. And about time too. CUDA isn't cross-platform so before something truly open is adopted widely, D3D is going to take a stab. And in the meantime NVidia will make noises about offering CUDA as an open platform, AMD will make noises about Brook+ as an open platform and presumably Intel will do the same with Ct.
Jawed
Jawed
Legend
Hardware platform, not software platform. CUDA doesn't run on other devices - though NVidia would like people to believe it runs on CPUs it produces very different results (not just mathematically but in terms of control flow) and is exceptionally slow.
Jawed
Hey, that's not bad.
"GT200" = Gary Tarolli 200. Sounds like a terminator model to me...
forget it; i see my nasty reputation has preceded me here and there is nothing i can say for i am a "messenger" that simply needs to be 'shot'.
i cannot explain for everyone has alreay shut their eyes and ears. i give my opinion, just like you guys do.
But you are right .. i know absolutely nothing and i apologize for saying anything to you as you already know as least as much as i do; also, some of you are under NDA, i am not
at any rate, please point me in the direction of the "white paper" threads.
i promise never to post in any speculation thread EVER again at B3D
i am most interested in non-controversial benchmarking and IQ comparisons .. can you please direct me; i used to know your site a LOT better 2 years ago when i just lurked here; and i gained a lot of good info from you then
i would like to share in discussions .. not attack or be attacked; i do not fit in here - please .. the technical discussions please .. i will keep looking
thank you and please ignore me in this thread; i will not respond further and i am sorry again for posting in it
Peace and aloha; you are a great bunch of guys and i do not want to upset you
apoppin out
Remember this ?
Now for the follow-up:
(dubious) source: http://forums.vr-zone.com/showthread.php?t=271801
Now for the follow-up:
(dubious) source: http://forums.vr-zone.com/showthread.php?t=271801
Well, it's not like adding a few blocks to the core would make the basic design different in any significant way, right ?
Just look at R300 vs R420, for instance.
Besides, there's no firm indication yet that G92b (55nm) has the exact same number of functional blocks as G92 (65nm).
Jawed
Legend
Well, I'm still holding on to the hope it's really one big chip, but 2xG92b could be compelling.
If each G92b is 6 clusters, that's 96 SPs + 48 TMUs, which matches what I think is the peak of perf/transistor for G92 (given bandwidth limitation), i.e. half-way between G94's and G92's cluster count.
Of course this rumour could explain why it's "9900". Still, not convinced.
At the same time, maybe G92b is really 8 clusters and all that's happening in the twin die SKU is that better-yielding 6-cluster chips are being used. NVidia could always deploy an Ultra version later consisting of the full 16 clusters across two chips.
Jawed
If each G92b is 6 clusters, that's 96 SPs + 48 TMUs, which matches what I think is the peak of perf/transistor for G92 (given bandwidth limitation), i.e. half-way between G94's and G92's cluster count.
Of course this rumour could explain why it's "9900". Still, not convinced.
At the same time, maybe G92b is really 8 clusters and all that's happening in the twin die SKU is that better-yielding 6-cluster chips are being used. NVidia could always deploy an Ultra version later consisting of the full 16 clusters across two chips.
Jawed
Blazkowicz
Legend
2x crippled G92b, that's ridiculous, such a card would be a "9700GX2" rather than a 9900GTX
How so? 512-bit MC matches that figure.
Yes, I think 32 ROPs is pretty much a given with a 512-bit MC - not that that makes the specs real.
- Status
