Analysis: Q3 2007 GPU shipments reach record highs

Tuesday 30th October 2007, 11:11:00 AM, written by Arun

After Q2 2007 being stronger than historical seasonality, few were expecting Q3 to be up sequentially by more than single-digits. Well, it looks like the skeptics were wrong, as the GPU market is up no less than 20.2% overall, according to JPR. This represents a larger jump than even the 18.6% one experienced in Q3 2002, and that year didn't have to account for a remarkably strong Q2 either!

It is noteworthy to point out that the increase in shipments is a desktop phenomenom; notebook shipments, on the other hand, have barely moved compared to Q2. This shouldn't be a major surprise, however, as Q2 corresponded to the release of the Santa Rosa notebook platform, and you'd expect mobile shipments to be especially strong then for that reason.

Compared to Q207, NVIDIA's shipments grew 25.2% sequentially to 33.14M units, while Intel's grew by 21.6% to 37.2M and AMD's by 17.7% to 18.66M. The overall market grew by 20.2% compared to the previous quarter, achieving record unit shipments of 97.85M units. But if you've been keeping track, you'll notice the numbers only add up to about 91% of the market.

The rest is attributable to Matrox's discrete solutions (0.12M units), SiS's IGPs (2.10M units) and VIA's IGPs and their S3 discrete GPUs (6.63M units). As can easily be seen, VIA and SiS remain non-negligible (although shrinking) parts of the market; it remains to be seen what'll happen in April 2008 when VIA will no longer be able to sell Intel IGPs, as their license will have expired completely.

In the notebook market, NVIDIA lost a slight amount of share to AMD in Q3. This seems strange on first glance, as NVIDIA was supposed to have gained share with their discrete GPUs in the Santa Rosa cycle. But upon further examination, it makes sense: AMD likely gained share in the notebook IGP market thanks to the 690G, which benefits both from superior performance compared to the MCP68 and bundling opportunities with AMD's CPUs.

The desktop numbers are even more mysterious, however. According to other market reports, desktop shipments only increased by 7.7% (although this tends to lag behind chip sales, which seem to have been about 15% higher or so). So why are GPU shipments 20% higher? We believe there could be two explanations to this. First of all, while Q2 results were excellent, inventories seemed to have gone down not only at GPU manufacturers, but also at OEMs and in the channel. In Q3, consumer demand remained strong, and inventory concerns were mostly solved by the end of the quarter. This might have generated a spike in chip sales to refill inventory.

The other factor to consider is that if a computer has an integrated graphics chipset, but you add a discrete GPU, you've actually got two GPU units sold for one PC! As such, it is possible that this boost represents an increase in the penetration of discrete GPUs, while manufacturers have not yet had the time to switch to non-IGP chipsets. That is, if they even want to, since the cheapest motherboards are often based on IGPs!

Another explanation for higher shipments than computer sales numbers would be that more gamers have been upgrading their GPUs (and not their CPUs at the same time). This would seem to make sense, given the release of a number of AAA games in recent and coming months. Furthermore, JPR claims G80 shipments specifically (not G8x) have exceeded all expectations, which would go along well with this theory.

AMD's discrete GPU revenues for Q3 were about 25% higher than in Q2, which makes a relative amount of sense given JPR's numbers. NVIDIA's results for the period (although with a slight offset; NVIDIA's financial quarters are not calendar-based) will come out on November 8th.

Discuss on the forums

Tagging

b3d ± nvidia, amd, intel, via, sis


Latest Thread Comments (7 total)
Posted by Arwin on Tuesday, 30-Oct-07 21:53:30 UTC
What historical seasonality? I'd be interested in knowing if people have done any analysis on the effects of pc games related sales in the context of console generation transitions. I doubt there is a lot of data, but I do think there's bound to be a relation ...

Posted by Arun on Tuesday, 30-Oct-07 22:16:28 UTC
Ah, I obviously meant quarterly variations - Q2 tends to be down compared to Q1, this has been the case in the GPU industry (and PC industry in general really) pretty much systematically for many, many years. Q3 and Q4 tend to be up, and Q1 very slightly down or flat compared to Q4.

Posted by Geo on Tuesday, 30-Oct-07 22:17:09 UTC
What I found very interesting in these numbers is there's basically no evidence here that ATI picked up any desktop market share after HD 2600/2400 shipped. Picked up significant *unit* sales on the desktop, as did everyone, but no increase in market share.

I'd love to see the DX10 numbers tho for 3Q.

Posted by Rangers on Wednesday, 31-Oct-07 05:00:31 UTC
Quote
NVIDIA's results for the period (although with a slight offset; NVIDIA's financial quarters are not calendar-based) will come out on November 8th.
Lets hope it's good news. Nvidia stock is killing my portfolio!

Posted by Geeforcer on Wednesday, 31-Oct-07 05:19:48 UTC
Quoting Rangers
Lets hope it's good news. Nvidia stock is killing my portfolio!
@~$35??? Did you buy it at all-time high or something?

Posted by Voltron on Thursday, 01-Nov-07 19:04:20 UTC
Quoting Geo
What I found very interesting in these numbers is there's basically no evidence here that ATI picked up any desktop market share after HD 2600/2400 shipped. Picked up significant *unit* sales on the desktop, as did everyone, but no increase in market share.

I'd love to see the DX10 numbers tho for 3Q.
I have seen some superficial Mercury numbers from a Goldman report that diverges considerably with the JPR numbers. I think a 4% increase in notebook share and 3% decrase in desktop share for Nvidia. But these might be discrete only. I'm not sure - the info was very limited.

Or it could be due to methodology. I am curious how close Mercury and JPR are and what the differences are.

A tangent, but to me JPR seems a little unprofessional in the way it reports. Colorful writing about "thuds" seems a little sensational. And what is the point of having a JPR section on Beyond3D when the same info is available elsewhere? Or are they providing something more to you guys?

Posted by Arun on Thursday, 01-Nov-07 21:39:53 UTC
Quoting Voltron
I have seen some superficial Mercury numbers from a Goldman report that diverges considerably with the JPR numbers. I think a 4% increase in notebook share and 3% decrase in desktop share for Nvidia. But these might be discrete only. I'm not sure - the info was very limited.
Those numbers would make a lot of sense from a discrete-only perspective. Although JPR's numbers would still be strange for desktops if true - I doubt NVIDIA shipped a gazillion MCP73s behind the scenes.

Quote
Or it could be due to methodology. I am curious how close Mercury and JPR are and what the differences are.
Both get their numbers from the IHVs, but JPR does extra channel checks to try to make sure it's accurate and get some extra data to analyse, AFAIK.

Quote
Or are they providing something more to you guys?
There have been a few JPR pieces on B3D that weren't available elsewhere, but there hasn't been any for at least a couple of months now IIRC.


Add your comment in the forums

Related b3d News

Book reviews: The Magic of Computer Graphics and 3D Engine Design for Virtual Globes
A new approach to graphics performance analysis
RWT Analyzes Bulldozer Benchmarks
Really Techy Worldly Hot Par 2010 coverage
Carmack Rage: either console could look superior
Sub $100 graphics at Tech Report
Broadcom purchase AMD's DTV business
Lucid Hydra 100 multi-GPU scaling demonstrated at IDF
The Khronos Group announce Heterogeneous Computing Initiative
3DMark, the Game