THE BATTLE FOR GPU SUPREMACY this coming winter solstice holiday season is looking like Nvidia (Nasdaq: NVDA) is bringing a dull butter knife and a blindfold to a howitzer fight. At least that is the impression its latest round of roadmaps is giving.
The roadmaps recently shown to SemiAccurate confirm the bleak outlook we have been talking about recently. There are two of them each dated mid-August, or about two weeks ago. One covers Q3/2009 through Q2/2010 and the other has Holiday 2009 and Spring 2010 selling season coverage.
One years worth of code names
The information covered is pretty easy to interpret. The D10 generation is a 55nm part, but Nvidia is up to its usual sleazy naming tricks, even to its NDA’d partners. It is claiming that renamed G9x parts are the D10 generation even though they are not. If it can’t be honest with itself, what chance does it have to be honest to customers? Don’t answer that. There is literally nothing new here.
D11 is the code name for the 40nm shrinks of, well, we aren’t really sure anymore. They could be G9x derivatives, or they could be G200 based, but since they are legitimate shrinks, we’ll give them the new generational name.
Take note that the D11 parts are only in the bottom three categories, Mainstream A and B, along with Performance A. As we said a few days ago, this is nothing more than a cynical renaming borne from a gross inability to make new parts. It also is quite emblematic of Nvidia’s inability to deliver even a G92 class part on 40nm. To bang that drum once again, if Nvidia has huge problems yielding ~100mm^2 and smaller dies on TSMC 40nm, what chance does it have of producing a >500mm^2 part? For the math averse, the answer is less than 1/25th the chance.
The most interesting item is D12U or GT300. The U stands for Ultra, the highest end variant. Nvidia is claiming that will be out in Q1, but that forecast is really dependent on what it gets back from the fab later this week or early next. If its hot lots need a respin, this Q1/10 date is suddenly going to seem hopelessly optimistic.
On the original roadmaps shown to SemiAccurate, all of the lower 2 classes had the D10 to D11 transition happening on the Q4 to Q1 boundary, that is, January 1. The Enthusiast segment has the D10U to D12U transition happening after that – the bar of the graph clearly goes well into Q1. This is either fudging or a CES paper launch, followed by availability sometime later.
Product names by season
More interesting bits can be found on the second page, the seasonal roadmap. ATI will have DX11 parts on sale in Q3/2009, which would be this month. For Q3 and Q4, all that Nvidia has to offer is the GTX280 with the RAM doubled to 1792MB. It will be pricing a 480mm^2 part with a hugely complex PCB against a 181mm^2 part with far fewer layers. GTX280 won’t be able to compete with Cypress at all, and Juniper can obliterate it on price/performance. Dull butter knife indeed.
The rest of the stack is known, with the only changes to note being the GT230 in Performance A getting double the memory. Until the new year, it is simply more of the same.
Remember when Jen-Hsun said that Nvidia was producing more 40nm wafers than anyone else? Remember when he said that it would be mostly 40nm by the end of the year? Anyone believe him? Where is the SEC when we need it? Where are the non-gullible analysts?
Pain really starts for Nvidia in 2010. Q1 brings D12U/GT300, if it can make them, likely in very short supply unless it does something radical to fix the yield toilet it is swimming in. Assuming it gets that out, and it performs better than we hear, then Nvidia is going to be shipping a 530mm^2+ part in very short supply against a low 3xxmm^2 part in plentiful supply.
To give you an idea of how confident it is of this plan working out, the roadmaps for the Spring 2010 season still list the GTX280 with 1792MB as the lead card. If GT300 doesn’t work perfectly and it needs to do a second spin, Nvidia is literally out of the high end graphics game. If that does work out, it gets to promote a card that it can’t make at the expense of the products it can produce. Great choices there.
The meat of the market, Performance B, is still served by the venerable G92. Ouch. No, really. Nvidia’s inability to make a part is getting painful to watch at this point. Juniper will kill this part at a lower cost.
The bottom three, the GT330/320/310 chips, are simply shrunk versions of their 230/220/210 predecessors. Nvidia is listing the performance gains from the 230 to the 330 as +20%, 220 to 320 as +60%, and 210 to 310 as 30%. For Performance B and Enthusiast, there are no gains listed because the parts don’t change at all.
Some of those might look like decent numbers, but ATI has a new generation top to bottom, and should be at about +100%. Game over, top to bottom.
In the end, the Q4 and Q1 roadmaps are what we have been saying all along, devoid of any hope. What we were not expecting was the Q2/2010 roadmaps to be as bleak as they are now. If Nvidia isn’t sandbagging on Q2, it is in deep deep trouble.
At this point, it is going to have to compete on price, and only in the low end at that. A single Cypress should be about on par with a GTX295, and Cypress will cost less than a GTX285 to produce. How long can Nvidia sell at a loss to make up volume? We will know by the end of Q1/2010. Until then, all it seems to have is spin, bluster, and the usual demos full of promise but lacking purchasable silicon.S|A
Latest posts by Charlie Demerjian (see all)
- More on Intel’s 10nm process problems - Sep 17, 2018
- Intel puts out another 14nm 2020 server platform - Sep 11, 2018
- Why Can’t Intel Supply Enough 14nm Xeons? - Sep 10, 2018
- Intel can’t supply 14nm Xeons, HPE directly recommends AMD Epyc - Sep 7, 2018
- AMD reintroduces the Athlon name with two CPUs - Sep 6, 2018