Nvidia GT300 core: Speculation

[Domell]

I write that GT21x is not dead. And yes, GT212 will be launched around the summer.

Following notice to GT300: G80 was running in Q1/2006 by Nvidia and the launch of the card was in Q4/2006.

Well, i didn`t know that G80 was running in Q1/06 but if you are sure it really did, it`s OK.

GT300: i really hope that this chip will be the next NV40 or G80 when we take a performance jump over GT212 and some fresh architectural changes. (as we know GT2xx is still a slightly improved good, old G80).

 

[Lukfi]

Just a thougt here: Some time ago I read about nVidia using a cluster computer to simulate their GPUs without having to manufacture actual samples. So what for would they need a physical G80 sample in Q1/2006 and a GT300 sample now?

 

[Jawed]

if NVIDIA already has a running GT300 then why there are GT212 in it`s plans??

All of GT2xx is running at least 6 months late, if the baseline for GT200 should have been a 2007Q4 release.

Also, there are rumours that G80 in one form or another was up-and-running at NVidia for roughly one year. Sure, they're vague, but...

Against this are rumours that D3D11 has only just been finalised or is just about to be finalised. I don't have a good idea what the final few months of finalisation for D3D actually amount to, i.e. I don't know what effect there might be on chip implementation.

For all we know the final few months of finalisation are "AMD: we've built this, this should be the spec"; "NVidia: we've built this, this should be the spec"; "MS: now now children, this is the spec ... oh wait ... erm ... ok? ... alright let's do it ... oh that didn't work ... ah ... sorry guys ... yeah we'll do it this way ..."

Jawed

 

[Arun]

Following notice to GT300: G80 was running in Q1/2006

Uhm... No. Your timeframes are quite out of whack. Here are two examples: the real A11 G80 chip was running in the labs in late Q2 2006. GT200 taped-out in late Q4 2007. The kind of delays you are thinking of are completely unusual, and unless many respins are required (ala R600 or, to only a slightly lesser extend, G98 and GT200b) they do not make any sense.

It is one thing to test one part of the chip in a TSMC CyberShuttle. It is quite another to have the actual chip working in your labs. I know the difference is subtle enough that you can't really expect sources or leaks to properly account for it, but you can't blame me for trying!

 

[Lukfi]

=>Arun: If I may ask a stupid and OT question, what is the TSMC CyberShuttle, how does it work for companies like nV/ATI?

 

[INKster]

=>Arun: If I may ask a stupid and OT question, what is the TSMC CyberShuttle, how does it work for companies like nV/ATI?

Here, hope this helps: http://www.tsmc.com/download/english/a05_literature/CyberShuttle.pdf

 

[KonKort]

the real A11 G80 chip was running in the labs in late Q2 2006.

Perhaps he had his tape out at this timeframe. But before he was running by Nvidia, too.

I did not say that GT300 were tape out!

 

[trinibwoy]

It's running but not taped out? Huh?

 

[silent_guy]

Perhaps he had his tape out at this timeframe. But before he was running by Nvidia, too.

I did not say that GT300 were tape out!

You really have no clue, do you?

 

[BRiT]

Perhaps he had his tape out at this timeframe. But before he was running by Nvidia, too.

I did not say that GT300 were tape out!

 

[homerdog]

Is the Idiot Bit List about to get longer?

 

[CarstenS]

For what it's worth some of the D3D11-Compute Shader presentations seem to imply available D3D11 hardware:

http://s08.idav.ucdavis.edu/boyd-dx11-compute-shader.pdf

slide 24. Though I admit it's possible to interpret this differently. Specifically, HD4870 is way way faster at FFT than anything else due to changes in the ALUs and this graph might merely be referring to the program running on HD4870, even though it's shown after the D3D11 data-point as the graph "progresses into the future". It may be that HD4870, in this case, is functioning identically to a D3D11 GPU?

I tend to interpret in the sense you line out in the second half of your posting. Being able to run a certain subset of D3D11 doesn't necessarily mean you have access to fully blown D3D11-hardware. Especially in the realms of compute shaders i could imagine current solutions being fit to at least emulate this functionality and even run certain sub-subsets, i.e. FFT, at full speed.

 

[KonKort]

It's running but not taped out? Huh?

Before a tape out there are internal tests. These tests are necessary to check the functions and to optimize. I would call them as pre-revisions.

I hope this is understandable. English is not my favourite language.

 

[Lukfi]

Then maybe these internal tests are the ones done on a computer. Simulations. I wrote it on the last page, but my post went unnoticed...

 

[trinibwoy]

Then maybe these internal tests are the ones done on a computer. Simulations. I wrote it on the last page, but my post went unnoticed...

To answer your question, the computer simulations can only flesh out the logic. It won't help them debug any issues they run into with manufacturing. So having an actual physical part to test is very important.

Before a tape out there are internal tests. These tests are necessary to check the functions and to optimize. I would call them as pre-revisions.

Internal tests of what? What is your definition of tape-out anyway? Maybe that's where the confusion lies.

 

[CarstenS]

Then maybe these internal tests are the ones done on a computer. Simulations. I wrote it on the last page, but my post went unnoticed...

Simulations are not as good for frontpage-news.

 

[KonKort]

Internal tests of what?

Internal tests of functions. Look at my post.
(The layout has not been completed yet!)

What is your definition of tape-out anyway? Maybe that's where the confusion lies.

In my definition tape out means that the layout is completed and reviewed and will be in production at a semiconductor manufacturer like TSMC.

 

[Arun]

KonKort: The problem I have with that is you made a specific claim, which is that "G80 was running in Q1/2006 by Nvidia". You also claimed GT300 was running right now. The problem with this is if 'running' means 'being simulated', then this is a very continuous process. There is no time when suddenly, after 3 years of work, you suddenly light up a switch and you go from not knowing whether anything works to suddenly having lots of information to debug potential problems.

Simulation is gradual. You simulate various subsystems at various levels of precision throughout the development process. I am not aware of any way you could say G80 was running in Q1 2006 but not in Q4 2005. Maybe there is a noteworthy part of the development process outside of possible CyberShuttles that I cannot think of right now, but no matter that it still feels like a very dubious claim to me.

I am sorry if this may not be easy to understand for someone whose native language is not english. If there's any part you want me to repeat with more but simpler words or something like that, just let me know...

 

[KonKort]

I think I can understand what do you mean with your posting. Do you know what a dress rehearsal (German: Generalprobe) is? I think this describes the current status of GT300 in best.
I know of a few GT300 developers that they are now involved on other projects and that the GT300 already very far advanced. As mentioned above but still no tape out.

PS: You have a private message.

 

[Lukfi]

Somehow I can't imagine how do you do a dress rehearsal with a graphics chip
If I understand correctly, a tape-out is when you get first full wafer of your chips, while from the CyberShuttle you just get a few chips because there are more chips from several manufacturers on one wafer. So either you have a CyberShuttle prototype, or a taped out prototype, or you have nothing. What does nVidia have now?