Phil Coffman's Comments

NVIDIA and the Big Chip, the Top Line and Gross Margins

Wed, 27 May 2009 12:33:39 -0400

nVidia appears to have gained share in Q1. See Jon Peddie Research: "net result was Intel and Nvidia were the big winners" (jonpeddie.com/press-re.../).

Anything Charlie writes about nVidia I would take with a big grain of salt.

On May 27 11:10 AM amdman wrote:

> NVidias chips are outdated and failure prone, they are losing market
> share.
> Nvidia is out at Dell
> www.theinquirer.net/in...

Nvidia's Fourth Quarter: As the Semiconductor Bellwether Turns

Sun, 01 Mar 2009 11:22:26 -0500

A very good article. Thanks for pointing it out.

On Feb 28 06:34 PM Misha wrote:

> I found that the following recent IEEE Spectrum article provides
> a concise and accessible description of why Intel’s (seekingalpha.com/symbo...)
> Larrabee may be such an important product with consequences with
> for NVDA:
>
> www.spectrum.ieee.org/...

GPU Wars: Attack of the $200-300 GPUs

Thu, 22 Jan 2009 01:44:07 -0500

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.
>

Intel's Q4 Earnings Call: Other Than That, Mrs. Lincoln, How Was the Play?

Tue, 20 Jan 2009 23:24:33 -0500

On Jan 20 08:34 PM Crocodilian wrote:

> Just a guess, but I'd rather compete with Intel's balance sheet than
> Abu Dhabi's at this point.

AMD + Abu Dhabi is something any investor in Intel, NVIDIA, or TSMC should pay close attention to.

Intel's Q4 Earnings Call: Other Than That, Mrs. Lincoln, How Was the Play?

Tue, 20 Jan 2009 17:49:47 -0500

On Jan 20 04:16 PM User 341043 wrote:

> You failed to mention that 32nm is the architectural sweet spot for
> Intel. It is the process node at which Intel can fit their chips
> into every form factor. You will see an explosion of "intel inside"
> when that happens.
>

Good point.

AMD's Sweet-Spot Strategy: Desperation or Stroke of Genius?

Sun, 30 Nov 2008 01:21:06 -0500

On Nov 28 06:17 PM Ames Tiedeman wrote:

> If AMD survives

AMD practically just received over $8 billion from Abu Dhabi.

If the deal goes through, AMD will survive.

AMD's Sweet-Spot Strategy: Desperation or Stroke of Genius?

Sun, 30 Nov 2008 01:14:21 -0500

On Nov 28 05:26 PM Shellshock wrote:

> Another factor to consider is that AMD/ATI may be selling their GPUs
> for lower price/performance because they are actually cheaper to
> produce. If their GPUs are actually cheaper to produce, then they
> can maintain profits at lower prices, forcing NVIDIA to take in less
> profit on a more expensive Silicon process.

No doubt the die of AMD GPUs cost less to produce. However, the fact that the ATI subdivision did not score a net profit, rather only an operating profit, strongly suggests that their margins are under pressure. In other words, they are selling their graphics really cheap.

Similarly NVIDIA did not have that good gross margins on its desktop GPUs last quarter.

There's a price war.

AMD's Sweet-Spot Strategy: Desperation or Stroke of Genius?

Sun, 30 Nov 2008 01:00:37 -0500

Intel and AMD put multiple cores in CPUs because the chipmakers can no longer drive performance by increasing the frequency at the same rate as in the past.

I believe AMD's dual-GPU strategy is the result of AMD being unable to effectively compete with NVIDIA on the high end since at least 2005.

I know AMD says otherwise, that it was because of power consumption that they went with their approach (See venturebeat.com/2008/1.../).

However, the power consumption of the GTX 260, as near as I can tell, is actually better than that of the HD 4870. This strongly suggests that power consumption ought not to be a show stopper for a large die.

On Nov 28 11:59 AM schkube wrote:

> While it certainly is not a perfect analogy, but seems like Intel
> did a very nice job packaging two C2D CPU's on a single die, in the
> same fashion ATI is doubling up on GPU's on a PCB with the X2.

>
> You are likely correct that the X2 model will not last (the market
> should decide that) but to say that approach makes ATI a poseur may
> be a bit of a stretch.

AMD's Sweet-Spot Strategy: Desperation or Stroke of Genius?

Sat, 29 Nov 2008 22:06:32 -0500

Actually the thermal characteristics (and power draw) of the GTX 280 and 260 are quite good. See techreport.com/article.... The GTX 260 drew less power than the HD 4870 at idle and full load. The temperature of the 260 under load was less than that of the 4870.

A more probable reason for not issuing a dual-GPU card at the present time is the size and possibly cost of the die. NVIDIA is supposed to do another GX2 when they shrink the process to 55nm.

On Nov 28 10:23 AM jcb wrote:

> I think a couple of points have been left out:
> Cost: Is a card with 2 ATI chips less expensive than 1 NVIDIA? Now
> compare the performance.
> Heat: If one puts 2 NVIDIA 280 chips on one card, I think it will
> spontaneously combust.
>
> Right now AMD cannot afford to be Bleeding Edge. They have to hope
> to survive by offering exceptional pricepoints.

AMD's Sweet-Spot Strategy: Desperation or Stroke of Genius?

Fri, 28 Nov 2008 22:47:50 -0500

Point taken about efficiency per silicon mm2 and the 4870 being much smaller than the 280. However, one of the reasons the die of the 280 is so large is that it is manufactured on a larger process technology: 65nm as opposed to 55nm.

I would argue that to get a fair comparison of performance or efficiency per silicon mm2, you would need to use a 280 or 260 manufactured at 55nm. It's coming.

Two 4870s may trump one 280, but two 280s trump two 4870s. So it all depends on your perspective: whether you want to compare GPUs or cards.

The other reason the die of the 280 is so large is that it has more transistors. One of the reasons it has more transistors is to extend the GPU beyond graphics.

On Nov 28 06:35 AM jkj wrote:

> I think you are way off here - whats important is the efficiency
> per silicon mm2. And if you took the time to look at the die sizes
> of the NVIDIA 280 and the ATI 4870 chip you would find that the 4870
> is less than half the size of the NVIDIA 280. So add two 4870s which
> by the mile floors the NVIDIA 280 in performance despite using less
> silicon die size.