I know, I know, we’ve been posting a lot of news and information about the recently released GeForce GTX 480 – but what else are we going to talk about over the weekend where my Alma Mater is left a tattered version of itself thanks to West Virginia?  Anyway…

I came across this interview at TechReaction with NVIDIA’s Bryan Del Rizzo that I at first overlooked but after giving it a read found it had some interesting quotes worth visiting.  Take a look at some highlights:

TechREACTION.net: Given the presentation today, it’s pretty clear that you guys are putting a strong focus on gaming.  But originally, before the product was officially launched, people were talking about how the GPU was really geared towards Tesla environments.

Bryan Del Rizzo: … So people came to GTC or they looked at the presentations for GTC, and they said “Well, what the crap?  Nvidia’s leaving the gaming market?”  Haha, no!  It wasn’t a gaming conference, it was an HPC conference.  So people didn’t grasp that, they need to understand that we’re doing all of this great stuff, and a chip can have many forms and functions, it can have many flavors.  For some reason a lot of people didn’t seem to get that.

This definitely WAS the reaction from both the fans and media after the conference last year.  I think NVIDIA deserves some of the blame for this fact though for not realizing that many of the media on-hand EXPECTED to see information about the GeForce variant during the event.  That didn’t happen, but rather than say “hey, we’ll have gaming details for you in January” NVIDIA decided to say nothing at all, leaving gamers everywhere to speculate on the reasons why.

TechREACTION.net: So they figured it just had one thing it was geared toward and that was it?

Bryan Del Rizzo: Yea, and the chip we’re designing for GeForce is dramatically different than the chip we’re designing for Tesla.

TechREACTION.net: So there are hardware differences between GeForce and Tesla then?

Bryan Del Rizzo: Of course, the easiest one is double precision.  There aren’t any games, there aren’t any user applications today that use double precision.  But in the medical markets, in the HPC field, that’s dramatically important to them.  So we’re not going waste the time and waste silicon space, and all that kind of stuff, providing a feature to gamers and incurring a cost for them, that they don’t need or possibly want.

NVIDIA Interview: GF100 variants, dual-GPU GTX 480s and a bit more - Graphics Cards 3
3 billion transistors.  Probably the same for Tesla as they are for GeForce.

I take some issue with this because everything we have learned about Fermi and the GPUs coming out of NVIDIA based on it is that they are essentially the same part used in the Quadro, GeForce and Tesla lines of products.  Bryan’s point about gaming and gamers not really needing double precision calculation support is true but to say that GF100 is someone how a smaller, re-spun chip that with “less wasted silicon space” would be incorrect I think.  We have never seen NVIDIA do this before – GT200-based Tesla, Quadro and GeForce products are known to be based on the same GPU core with memory differences and validation changes; but not much else.  If somehow NVIDIA did actually completely respin the GPU to remove double precision from the GF100 line then this would be a noticeable corporate shift meant to keep hardware vendors from replacing expensive Tesla-branded cards with lower priced GeForce options.

TechREACTION.net: Some news sources have suggested that there are more Fermi-based GPU’s due sometime in June / early Summer. Is there any credibility to that?

Bryan Del Rizzo: Well, actually there probably is because I think one of our executives in the financial offices said that you can expect other variants of the GeForce family sometime in the second half of this year.  So, based on what they said that probably is true.

So you can figure that we’re going to have some more mainstream products with that, and you can figure that we can probably fit some cool stuff on top of that as well.  So, expect the unexpected.

Many have speculated that in GF100’s current form, taking the traditional route of standard lower-priced and lower-performing GPUs might not work due to the high power demands they would have compared to the competition from AMD’s 5600/5500/5400 line of cards and even NVIDIAs own G9x parts.  If this is true then Bryan is probably referring to a re-spun Fermi GPU that would hopefully bring yields up and power down and thus enable the normal “huge monolith to smaller GPU” transition.  Does that mean that the GTX 480s and GTX 470s that will be sold next month are going to be replaced by potential GTX 485s and GTX 475s?  It seems pretty likely.

TechREACTION.net: Previously you’ve had dual GPU cards such as the GTX 295.  Do you have any plans to do the same thing with the GTX 480 GPU?

Bryan Del Rizzo: So we can’t comment on that today, we haven’t announced any product.  But I have to kind of laugh because when we launched the GeForce GTX 285, there were a lot of rumors that said we couldn’t do a dual GPU product with that chip.  They said it was architecturally impossible, it was too power hungry, it would just be impossible to do.  And of course we brought the GTX 295 to market; it did really well!  (laughs)  So, if you look at the history of Nvidia, you can probably pretty much figure out what we’re going to do, but I can’t give you a confirmation on that today.

NVIDIA Interview: GF100 variants, dual-GPU GTX 480s and a bit more - Graphics Cards 4
GTX 295 above.  Would a dual-GPU variant of GTX 480 require a 1.21 gigawatts power supply?

What Bryan fails to mention here is that the GTX 295 did not use the same GPU as the GTX 280; the GPU of which the media claimed NVIDIA would be insane to offer a dual-GPU variant of.  Instead, the GTX 295 was the first graphics card to use the 55nm version of GT200, later found in the GTX 285, and thus was able to run on much lower power than the original 65nm GPU.  Personally, I think we can all agree that making dual GPU version of the GTX 480 as it stands today would be a bad idea and if/when we see that Fermi dual-GPU card it will be based on a much more refined version of the architecture. 

The interview with Bryan Del Rizzo over at TechReaction.net has a lot more goodness to read if you are interested so I suggest you all give it a look.