GPU Wars: Attack of the $200-300 GPUs

It’s a good thing NVIDIA (NVDA) does more than just desktop GPU discrete graphics these days. If NVIDIA had never expanded beyond the desktop, the company would really be in a world of hurt. NVIDIA’s revenue share, for example, in high-end GPUs went from 80% to 30% from the second to third quarter (NVIDIA Corporation at Credit Suisse Group Technology Conference Presentation, December 2, 2008)!

High-end desktop graphics has traditionally been a high-margin business, with cards commanding $400-600 in prices. Not anymore. AMD recently released chips intended for the $200 and $300 markets, the Radeon HD 4870 and 4850. However, the $300 chip was good enough to take on NVIDIA’s $400 offering, and ATI’s $200 GPU NVIDIA’s $300 GPU. “ATI now had a $300 card that was competitive with NVIDIA’s brand new $400 GTX 260“.

A price war ensued. There went the margins.

The pricing of the card above the $200-400 cards in the GPU hierarchy was even affected.

Today cards that originally cost around $300 and $400 can be had for little more than $200. Amazing.

While price wars are good for consumers, they can be bad for margins. Make no mistake. The smaller AMD chip costs less to produce than the larger NVIDIA chip. Nevertheless, the fact that AMD’s ATI graphics division has yet to return a net profit argues that ATI’s GPUs are not making that much money.

Likewise, NVIDIA’s desktop margins are under pressure. High-end GPUs have traditionally been a high-margin business for NVIDIA and chipsets a lower-margin business. Yet the CEO was recently asked if “chipset gross margins right now are greater than your desktop gross margins”, and the CEO replied, “Yes, they are close“.

AMD’s current company-wide strategy is simply to return to profitability. The company’s gross margins are such that all they have to do, to achieve a net profit, is get their sales up. It should happen eventually, though it doesn’t look like it’s going to happen this quarter. Consequently AMD is presently trying not so much to push the envelope of technology, or to expand the overall CPU or GPU markets, as it is trying to take market share from what’s already out there.

One area where AMD would love to take market share is the incredibly lucrative area of professional workstation graphics. This space is practically owned by NVIDIA, which in the third quarter had “90% of overall units” and enjoys gross margins of around 70% (Credit Suisse Conference, 12/2/08).

I don’t see how a chip designed for the $300 price point, such as ATI’s, is going to scale far into the high end of professional workstation graphics. It seems that it would be more difficult to scale up a GPU than to scale down: to add to the complexity, than to simplify a design.

That being said, the reviews I have seen of ATI’s professional workstation cards have been good. We’ll have to see how this plays out. I always like a good fight.

Disclosure: Long NVIDIA.

Comments

[Jon Smirl:]

Long run Nvidia is toast. The new integrated GPU/CPU combos under design at Intel/AMD are going to destroy the low and middle cost GPU market. Nvidia could still dominate the high end but it will be a much smaller company. Over time the GPU/CPU combos will move towards the high end.

There have been rumors of Nvidia buying an x86 CPU design in order to compete with the integrated GPU/CPU but there is no confirmation that they have bought one.

2008 Dec 15 02:39 PM

[Kris Tuttle:]

Collapsing margins are certainly a well-known part of the story at this point. The question now is how well can NVDA adjust to the new environment. Some have pointed out that the company culture may not lend itself well to the lean times ahead but I don't have firsthand experience with them.

That the standard UI is moving into the high end visual domain is clear. NVDA and ATI are also improving the ability of the GPU to handle some high-end processing to improve performance on certain applications.

Personally I think that the demands for visual computing will far outstrip the power of integrated solutions for something like a decade. The human field of active vision is fairly large (2-4x current screen size) and appreciates responsive photo-realistic imaging. Said another way the expectation is that the "graphics" will be nearly indistinguishable from the actual object.

The processing required to handle the computational complexity between physical objects and their visual rendering is pretty major stuff.

(We have a positive research opinion on NVDA for some of the reasons noted above combined with their historically high ROIC and huge cash/asset position.)

2008 Dec 16 08:33 AM

[mr_mike:]

"....I don’t see how a chip designed for the $300 price point, such as ATI’s, is going to scale far into the high end of professional workstation graphics. It seems that it would be more difficult to scale up a GPU than to scale down: to add to the complexity, than to simplify a design....."

Professional cards workloads are very different than gamer cards
and considering that the single chip 4870 can to 1.2TFLOPs vs a single 280 doing only 0.933 TFLOPS. So for pro cards they don't need to scale anything up, the drivers are different to reflect the different workload.

Jan 02 09:34 PM

[Phil Coffman:]

On Jan 02 09:34 PM mr_mike wrote:

> "....I don’t see how a chip designed for the $300 price point, such
> as ATI’s, is going to scale far into the high end of professional
> workstation graphics. It seems that it would be more difficult to
> scale up a GPU than to scale down: to add to the complexity, than
> to simplify a design....."
>
> Professional cards workloads are very different than gamer cards

>
> and considering that the single chip 4870 can to 1.2TFLOPs vs a single
> 280 doing only 0.933 TFLOPS. So for pro cards they don't need to
> scale anything up, the drivers are different to reflect the different
> workload.
>
Thanks for the clarification.
>

Jan 22 01:44 AM