Fermi on the move: 480 surpasses HD3200! - Steam HW Survey

[Arkadrel]

Keep us informed Lurrker

Id like to know how much market the 4xx has now that 6xxx series are comeing out.
I believe in the start of this tread Nvidia was around 12% of the dx11 market, with 88% of it being AMDs. Since then what has happend with the market?

[lurrker]

Originally Posted by Arkadrel

Keep us informed Lurrker

Id like to know how much market the 4xx has now that 6xxx series are comeing out.
I believe in the start of this tread Nvidia was around 12% of the dx11 market, with 88% of it being AMDs. Since then what has happend with the market?

Just look for the reports from Mecury Research (http://www.mercuryresearch.com/schedule.shtml) and Jon Peddie (http://jonpeddie.com/) - there maybe others but they attempt to get as much information as possible.

[fusion]

Originally Posted by Chocolate Factory

So you're suggesting that the initial batch of 9000 wafers (is this right?) is enough to satisfy demand. And since we're all assuming that yield is awful, this implies that there are only a small number of GF100 based graphics cards out there. As a result, the Steam hardware survey must be inaccurate.

Did I get that right?

I definitely agree that something does not add up here. It's one of the following:

1) perhaps the initial batch of 9000 wafers is wildly wrong. Maybe it was 9,000,000 wafers! Hah!
2) or maybe Steam is very wrong. Maybe nvidia has a massive marketing campaign to increase apparent Steam marketshare?
3) or maybe gf100 yields are just fine

Given all the evidence at hand, if I had to pick one of the above to be suspicious about, I'd pick #3.

Steam supplies percents, nothing more. They don't supply the amount of people surveyed. You can only get general trends out of steam, that's it, and that's why they do it. Trying to get the amount of GPU's sold out of steam is like trying to drive a nail into wood with a saw.

[bh192012]

Of the 30 million "active" Steam users, currently about 3 million are on right now, in the middle of the work day on Monday.

Of those 3 million connected, statistically .1193 are DX 11 users. Times .0389 of those DX 11 users have GTX 480's. That's close to 14,000 GTX 480s probably currently connected to Steam and a total of 139 thousand GTX480 owners with steam accounts.

Therefore there were probably either higher yields, or more wafers than first predicted. Perhaps there was a second batch with signifigantly higher yields, then production stop (to help explain the lack of later die shots from other threads?) Also we're talking just the 480's here and not 470's or pro/tesla, which increase what I think most call "yield."


ps I believe Steam has the DX 11 market at 86% ATI and 14% Nvidia. 10/25/2010

pps Or maybe Steam is completely unreliable for graphics card statistics.

[douglar]

Originally Posted by bh192012

Of the 30 million "active" Steam users, currently about 3 million are on right now, in the middle of the work day on Monday.

...

pps Or maybe Steam is completely unreliable for graphics card statistics.

You are very fast and loose with your precision and the units behind your numbers.

1) Even if there were exactly 30 million "active" steam user accounts, that does not mean there are 30 million active hardware installations. Accounts are not the same thing as hardware.
2) I only see 2.7 Mil online users at 3pm EDT. That's a significant rounding amount there. And unlike "total user accounts", "online users" will correspond to 2.7 million pieces of hardware.
3) The unit for the steam hardware survey numbers is "surveys performed". The "survey's performed" may or may not correlate to "total accounts" or "current users". My guess is that it doesn't correlate close enough with either to give your calculations a statistically valid margin of error.

But I do agree that it seems likely that Nvida ran more GF100 wafers than the inital batch of risk wafers, but don't blame the statistics for not giving a clear result if the statistics are not used correctly.

[bh192012]

Thank you very much.

[krumme]

Let me understand this right

Let us asume we have 139k 480 and aprox the same numbers of 470, a total of 280k.
On 9000 wafers thats aprox 30 per wafer
If we have 104 candidates thats about 30% yield, for 480 including the salvaged part 470. Or about 15% for the 480.

Those numbers sounds like realy bad economic to me, even if Charlie says they are very wrong.

If a wafer is usd 5000, for 30 pcx. thats aprox. 170 usd for the die alone . It obviously doesnt work for at volume card, but at these low numbers its more of a Halo part anyway.

[lurrker]

Originally Posted by krumme

If a wafer is usd 5000, for 30 pcx. thats aprox. 170 usd for the die alone . It obviously doesnt work for at volume card, but at these low numbers its more of a Halo part anyway.

Fortunately there were quadros to absorb some of the pain - but how many 470's/480's were produced given those wafer counts and the financial pain of making them?