NVIDIA GF100 & Friends speculation

[aaronspink]

Seems pretty cut and dry to me. Nvidia has three important businesses that would be affected - Tesla, Geforce, Quadro. AMD only has one. Is that distinction not a big deal in your eyes? Granted JHH is telling everyone that it's all about the software when it comes to Tesla and Quadro.

Not three, two. Quadro is the exact same as Geforce. And your assuming that Tesla is actually a business at this point...

 

[aaronspink]

Tegra is targeting a far larger market than AMD based PC's and has much higher margins than the low-end discrete segment. If it's successful I don't see why it can't completely replace that revenue and then some.

Its unlikely that Tegra realistically has as good margin as you believe. They are just get another arm+gpu chip and don't seem to be actually shipping in any real quantity. In fact if anything, given the NRE and the sales they have, its probably a lower margin part at this point in time.

If anything, the drama surrounding Cypress should give you a new perspective on how involved these decisions can be and in that scenario you didn't have multiple competing product lines to worry about (in terms of priorities).

Nvidia doesn't have multiple competing product lines either. They have Geforce. Everything else is a sucker fish sideline. If they are viewing it any other way, then no wonder they are having issues. You don't let the low volume markets that can't even cover NRE set the direction of a part.

 

[XMAN26]

Not three, two. Quadro is the exact same as Geforce. And your assuming that Tesla is actually a business at this point...

I'm sorry, but only fools think of Geforces and Quadros as being the same products. They may have the same chip, but unlike the versions before DX10 where you could move some resistors around and do a BIOS flash, it aint true today. Then there is also teh fact that no matter how much someone might wanna try and say perforamnce is the same either way, that just isn't true.

 

[willardjuice]

They may have the same chip

I think that was his point...

 

[aaronspink]

I'm sorry, but only fools think of Geforces and Quadros as being the same products. They may have the same chip, but unlike the versions before DX10 where you could move some resistors around and do a BIOS flash, it aint true today. Then there is also teh fact that no matter how much someone might wanna try and say perforamnce is the same either way, that just isn't true.

So what you are trying to say is: because they moved from on external resistor to an internal fuse its now totally different? Um, whatever.

The only difference in performance comes from them purposely crippling one card in wireframe performance in OpenGL via the drivers. The hardware is functionally and performance wise exactly the same.

 

[Ailuros]

How well have things been going as planned at nvidia? People are pinning way to much hope on tegra, in a market that has historically killed non-standard architectures, and pushed down to bargain basement pricing.

I don't see what Tegra has to do with GF100, but I could think of quite a few advantages AMD's Ontario will have against it. As for bargain basement pricing my memory is anything but weak for ATI/Imageon. It did so "well" with bargain basement pricing that the department got sold to Qualcolmm I guess.

 

[Squilliam]

Yep, that whole segment is essentially busted in a year or two. I don't think Fusion is gonna be anything near an enthusiast mobile part though. What's the current rumour - Redwood class? You have to compare it to the discrete mobile parts that will be out at that time.

Redwood is only 1/2 Juniper which the highest end GPU available in the laptop space. That puts it in the enthusiast territory. Though within a couple of years I suspect that most laptops won't have discrete graphics on the AMD side of things anyway.

 

[dkanter]

I'm sorry, but only fools think of Geforces and Quadros as being the same products. They may have the same chip, but unlike the versions before DX10 where you could move some resistors around and do a BIOS flash, it aint true today. Then there is also teh fact that no matter how much someone might wanna try and say perforamnce is the same either way, that just isn't true.

They may be different products, but it's the same IC. The difference is less than that between a notebook and server CPU from Intel.

The main differences area few fuses, the driver and support/certification.

More to the point, without the volume of the GeForce line, the Tesla/Quadra line would not make sense.

David

 

[trinibwoy]

How well have things been going as planned at nvidia? People are pinning way to much hope on tegra, in a market that has historically killed non-standard architectures, and pushed down to bargain basement pricing.

Its unlikely that Tegra realistically has as good margin as you believe. They are just get another arm+gpu chip and don't seem to be actually shipping in any real quantity. In fact if anything, given the NRE and the sales they have, its probably a lower margin part at this point in time.

Maybe you're right and they will fall flat on their faces. Or maybe they won't. At this point I see no reason to predict outright failure.

Nvidia doesn't have multiple competing product lines either. They have Geforce. Everything else is a sucker fish sideline. If they are viewing it any other way, then no wonder they are having issues. You don't let the low volume markets that can't even cover NRE set the direction of a part.

No you don't but at the same time Tesla is an important part of Nvidia's long-term strategy so I don't think it quite has the red-headed stepchild status that you're implying.

Anyway, didn't they ask us this already back in 2003? And we all know how that turned out. From Facebook:

 

[aaronspink]

No you don't but at the same time Tesla is an important part of Nvidia's long-term strategy so I don't think it quite has the red-headed stepchild status that you're implying.

Wise man says: "you can't have a long-term strategy without a short-term strategy that will succeed"

Remember, 3DFX also had a long term strategy. But then again, their short-term strategy painted them into a corner too.

Tesla is fine as a pie in the sky we can sell dirt for 30K per ounce, but the problem is getting there and even if there is a there there. I think Tesla like most niche HPC products is going to get run over by both its own drawbacks and by the steady march of the micros.

Anyway, didn't they ask us this already back in 2003? And we all know how that turned out. From Facebook:

I wonder how long until we get "Dust Buster 2010: The year we make contact... with the ground!"

 

[KimB]

Yeah, sounds to me like you're talking out your ass here, Aaron. Sure, there are reasons why nVidia might not do all that great on this product cycle. But there is as yet no reason whatsoever to believe that this product will be a flop in any respect. It might be not quite as good as the competition, but the chance that it's going to be another NV30 is slim to none.

 

[rpg.314]

Tesla is fine as a pie in the sky we can sell dirt for 30K per ounce, but the problem is getting there and even if there is a there there. I think Tesla like most niche HPC products is going to get run over by both its own drawbacks and by the steady march of the micros.

Tesla is unlike most niche HPC products, which were specialised architectures used in the HPC market only. It is a special version of the Geforce line, that's all. In fact, it is no more and no less a niche HPC product than Xeon is a niche server product.

Tesla is important part of nv strategy simply because 80% of the volumes in discrete market are going to vanish in the next 5 years, so they need something to get their ASPs high (and lots of CUDA only code written) before that happens.

 

[Jawed]

You're starting off on the assumption that they could get out a smaller chip in a materially shorter timeline.

Cypress was adjusted both for cost (die-size) and timeliness.

But yes, you're right, NVidia's general inability to execute does imply the chances of GF100 benefitting would be lower than seen with Cypress.

Jawed

 

[XMAN26]

So what you are trying to say is: because they moved from on external resistor to an internal fuse its now totally different? Um, whatever.

The only difference in performance comes from them purposely crippling one card in wireframe performance in OpenGL via the drivers. The hardware is functionally and performance wise exactly the same.

You'll have to provide some links to this "same performance" as you claim. As in the past and today, the Quadros are anywhere 2x to 5x the performance in Pro apps compared to Geforce cards and this is because they tune the drivers for those apps not just "wireframe". It is also the reason why Quadros make for fairly bad gaming cards.

 

[Sontin]

Nvidia doesn't have multiple competing product lines either. They have Geforce. Everything else is a sucker fish sideline. If they are viewing it any other way, then no wonder they are having issues. You don't let the low volume markets that can't even cover NRE set the direction of a part.

Without the "low volume market" there would be no high-end gpu. Why should they worry with huge dies when they can build small chips, put two or more on a pcb and call it "high-end"?
But the quadro business has higher margin and you don't need to sell a lot of cards. So thx to the quadro that the gamer get real high-end and not this stupid AFR mGPU thing.

 

[GZ007]

You'll have to provide some links to this "same performance" as you claim. As in the past and today, the Quadros are anywhere 2x to 5x the performance in Pro apps compared to Geforce cards and this is because they tune the drivers for those apps not just "wireframe". It is also the reason why Quadros make for fairly bad gaming cards.

Yep but just one driver costing u 1000-2000 dolars more is not very "fair" and is the reason of the incomes from quadro cards.
Iam still convinced that the actual tuning should be made by software companies not nvidia if u pay 10K dolars for their license. And also u pay for all the software testing from nvidia while u use just 1-2 programs from all.

 

[rpg.314]

You'll have to provide some links to this "same performance" as you claim. As in the past and today, the Quadros are anywhere 2x to 5x the performance in Pro apps compared to Geforce cards and this is because they tune the drivers for those apps not just "wireframe". It is also the reason why Quadros make for fairly bad gaming cards.

Since, the IC's behind the Quadro and Geforce are the same, what exactly prevents them from shipping the tuned drivers for geforce drivers for the same apps?

Here's an example of this market segmentation. Can't find it in nv forums, had to dig up google cache.

http://74.125.153.132/search?q=cach...+forum+sriramkashyap&cd=2&hl=en&ct=clnk&gl=in

 

[digitalwanderer]

But there is as yet no reason whatsoever to believe that this product will be a flop in any respect. It might be not quite as good as the competition, but the chance that it's going to be another NV30 is slim to none.

I don't agree with your odds of "slim to none" there Chal, the delays and the continued lowering-of-the-expectation-bar feels a hella lot like the days before the FX30 came.

I'm not saying it will be that bad, but I don't see any indications yet to say it's out of the question either.

 

[XMAN26]

http://www.tomshardware.com/reviews/quadro-fx-4800,2258-10.html

The results from the above link should put a damn end to this Geforce is the same performance as a Quadro in Pro apps for once and for all.

 

[Sontin]

I'm not saying it will be that bad, but I don't see any indications yet to say it's out of the question either.

I do. And the name is cypress.
r300 was monster. A lot more processing power, a better architecture, 256bit interface, new aa...
Cypress is a small refresh of the same architecture. Nothing really new, not really fast - a lot like the NV30. It's funny that somebody could thinks, that the card would be slower than cypress. Only < 30% faster than a GTX285?