[aaronspink]

Originally Posted by trinibwoy

Gah! So his "thnk scientist" clue actually makes sense? WTF does that mean?

supposed to be a snarky reference to tesla. But its useless. it s like some big secret, oh nvidia is going to do exactly what they have already been doing but now its in the chip name. See now everyone should be scared! and I mean SCARED! cause they are putting it in the name and my oh so prophetic vision has seen that everyone else is going straight to hell!!!!!!!!

Let me give everyone a clue, ignore is a function for a reason. Its because posts by apoppin exist.

Aaron spink
speaking for myself inc.

[trinibwoy]

Haha you're kidding right? Wow that's a pretty weak basis for proclamations of Jen-Hsun's impending domination of the GPU world. I don't believe in ignore functions but those posts are a bit on the looney side for real.

[Pete]

Silly question time. Everyone's aware of this slide ("Era of Visual Computing," that marks Tesla 2 for 2008 and "Next Gen" for 2009, both under the "Programmable Graphics: CUDA-DX11" umbrella) from this Bit-Tech interview with Kirk, right? I'm guessing Tesla 2 is "GT200?"

[Razor1]

Seems like they were highlighting new generation chips but forgot about one of thier lines....heh fx series didn't make into that chart?

[nAo]

DX11's compute shader is likely to be losely based on CUDA

[aca]

Originally Posted by apoppin

NP, if that is what you really believe, then we are OK and will agree to leave my "noise" alone. I am not so sure about what you mean by my "aim"

i aim to share what i learned and also to learn from you guys. Especially about benchmarking and IQ comparisons. Would you kindly point me in the right direction? i will get the heck out of this type of discussion as i am evidently too controversial in my off-the-wall views and summaries of my analysis for here.

. . . and you will be shocked about the 'T' as new architecture i think .. wait and see
=D

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.

[_xxx_]

T stands for (3dfx) T-Buffer...

[trinibwoy]

Editor's day pushed back to May 22nd - http://www.theinquirer.net/gb/inquir...es-editors-day

[INKster]

Originally Posted by nAo

DX11's compute shader is likely to be losely based on CUDA

I thought the general idea of CUDA was to bypass DirectX entirely... ?

[Jawed]

Originally Posted by INKster

I thought the general idea of CUDA was to bypass DirectX entirely... ?

It pretty much does - except that you still need a working driver of some kind on the PC.

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

[INKster]

Originally Posted by Jawed

It pretty much does - except that you still need a working driver of some kind on the PC.

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

CUDA isn't cross-platform ? Then why is it available for Linux and MacOS X, besides Windows ?

[Jawed]

Originally Posted by INKster

CUDA isn't cross-platform ? Then why is it available for Linux and MacOS X, besides Windows ?

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

[SirPauly]

Originally Posted by _xxx_

T stands for (3dfx) T-Buffer...

Didn't the T in T-buffer stand for Tarolli?

[INKster]

Originally Posted by SirPauly

Didn't the T in T-buffer stand for Tarolli?

Hey, that's not bad.
"GT200" = Gary Tarolli 200. Sounds like a terminator model to me...

[SirPauly]

Hehe, I was hoping you would get it.........kind of neat the Gary Tarolli-200! Gary Tarrolli offering in his terminator voice, "I'll be back!"

[apoppin]

Originally Posted by aca

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.

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

[INKster]

Remember this ?
Now for the follow-up:

GeForce 9900 GTX (GT200) specs leaked

55nm TSMC process
Single chip with "dual G92b like" cores (!?)
330-350mm2 die size
900M+ transistors
512-bit memory interface
32 ROP
192SP (24X8)
6+8 Pin
550~600W PSU min
9900GTX SLI runs Crysis 2560x1600 VH 4XAA smoothly

(dubious) source: http://forums.vr-zone.com/showthread.php?t=271801

[Lukfi]

Dual G92b-like cores, but 24 SP's per block? Don't they contradict themselves?

[INKster]

Originally Posted by Lukfi

Dual G92b-like cores, but 24 SP's per block? Don't they contradict themselves?

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]

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

[Skinner]

I hope it's more then just two G92's if that's what it supposed to be. Or the B version should be a lot stronger then the regular G92

[Blazkowicz]

2x crippled G92b, that's ridiculous, such a card would be a "9700GX2" rather than a 9900GTX

[Kaotik]

"32 ROPs" should tell one enough, it's BS

[ShaidarHaran]

Originally Posted by Kaotik

"32 ROPs" should tell one enough, it's BS

How so? 512-bit MC matches that figure.

[Tim]

Originally Posted by ShaidarHaran

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.