NVIDIA GF100 & Friends speculation

[GZ007]

They have the cards with final tesla design on the nvidia page too http://www.nvidia.co.uk/object/product_tesla_C2050_C2070_uk.html .
The frequency is 1.17-1.4 GHz (520GFlops - 630 GFlops) from the specs and 190 W max power consumption.
The american page is showing 190 W typical not max consumption.

 

[XMAN26]

They have the cards with final tesla design on the nvidia page too http://www.nvidia.co.uk/object/product_tesla_C2050_C2070_uk.html .
The frequency is 1.17-1.4 GHz (520GFlops - 630 GFlops) from the specs and 190 W max power consumption.
The american page is showing 190 W typical not max consumption.

Be nice if someone who visited here was able to look first hand at th cards. Cause even the pick in your link shows screws on the exhaust plate.

 

[GZ007]

Be nice if someone who visited here was able to look first hand at th cards. Cause even the pick in your link shows screws on the exhaust plate.

Under the picture "Product appearance may vary by manufacturer". The post was more about the 190W power which should be enough for single 8 pin or dual 6 pin.

 

[mczak]

Hmm C2050 available Q2 C2070 Q3. Now is that because they can't get enough good parts out in tiny quantities with even 448 cores and a quite modest 1.4Ghz clock or because the probably needed 2gb gddr5 chips aren't available?

 

[FrameBuffer]

Hmm C2050 available Q2 C2070 Q3. Now is that because they can't get enough good parts out in tiny quantities with even 448 cores and a quite modest 1.4Ghz clock or because the probably needed 2gb gddr5 chips aren't available?

the loyalists will place the blame of lack of memory
the naysayers will say it's lack of GPU availability

pick your poison

 

[Kaotik]

First 2 GF100 based GeForces will be called GeForce GTX470 and GTX480
http://www.facebook.com/NVIDIA?ref=nf

Fun Fact of the Week: GeForce GTX 480 and GeForce GTX 470 will be the names of the first two GPUs shipped based on our new GF100 chip!

So they finally learned from all the renaming confusions, and all GF3xx's will be DX10.1 chips?

 

[trinibwoy]

Yeah that makes sense. 4xx series for DX11. GTX 470 though? Does that indicate it's not going to be much slower than the big dog?

 

[mczak]

First 2 GF100 based GeForces will be called GeForce GTX470 and GTX480
http://www.facebook.com/NVIDIA?ref=nf


So they finally learned from all the renaming confusions, and all GF3xx's will be DX10.1 chips?

Sure of that? I'd be totally unsurprised if we'd see some GT4xx cards which would be just renamed GT3xx parts (which were just renamed GT2xx parts...).
FWIW there now seem to be some OEM-only (Medion for now) GT330 cards, and current rumors say it's probably renamed 9600GTO (mostly because of the 768MB memory hence implying 192bit bus plus ads saying DX10), so much for GT3xx parts being DX10.1...

 

[KimB]

Sure of that? I'd be totally unsurprised if we'd see some GT4xx cards which would be just renamed GT3xx parts (which were just renamed GT2xx parts...).

Perhaps, thought it rather depends upon how rapidly they fill out the GF100 family. Though I don't think they've crossed names between major generations since the GF4 MX.

 

[BRiT]

First 2 GF100 based GeForces will be called GeForce GTX470 and GTX480

So what will they name card numbers 3 and 4 or above? Surely they can produce more than 2 cards.

 

[thatdude90210]

I think this move, Fermi to be 4xx series, frees Nvidia to rename their entire current lineup to 3xx. They can introduce the 3xx series and the 4xx series together. Dx11 will be separate from Dx10/10.1 and AIB's will get their full (or fool) new lineup.

 

[seahawk]

Hmm C2050 available Q2 C2070 Q3. Now is that because they can't get enough good parts out in tiny quantities with even 448 cores and a quite modest 1.4Ghz clock or because the probably needed 2gb gddr5 chips aren't available?

Or because of make the most money with the GeForce line in the early days. So it makes sense to put the good chips into the 480 and delay the bigger Tesla a little.

 

[Ailuros]

Or because of make the most money with the GeForce line in the early days. So it makes sense to put the good chips into the 480 and delay the bigger Tesla a little.

My memory is quite weak when it comes to the 1st generation of Tesla variants, but I think the C10x0 weren't based on GT200 but on GT200b@55nm (stands open for correction though). If it should be true though NV itself rates the GTX280 with a 236W maximum power consumption while the GTX285 with a 204W power consumption, despite the latter having slightly higher frequencies than the 280.

If all the above should align, then NV's primary concern for Tesla products is power consumption; especially since the GTX280 had 6pin+8pin connectors (like the coming 480) and not like the 285 2*6pin.

 

[seahawk]

Very possible indeed. But that would also mean they expect to have a revision of GF100 in Q3 that will need much less juice.
Currently I think they just do not have that many chips that would be good enough for a top of the line Tesla, so they do the clever thing and sell those as GeForce 480, while the margins in the product line are still the highest. Later when production matures Tesal is added. The time window in which they can demand a premium is much smaller for the GeForce line, than for the Tesla line.

 

[KimB]

Very possible indeed. But that would also mean they expect to have a revision of GF100 in Q3 that will need much less juice.
Currently I think they just do not have that many chips that would be good enough for a top of the line Tesla, so they do the clever thing and sell those as GeForce 480, while the margins in the product line are still the highest. Later when production matures Tesal is added. The time window in which they can demand a premium is much smaller for the GeForce line, than for the Tesla line.

Huh? I'm pretty sure Tesla will be a lower-volume market.

 

[Blazkowicz]

For Tesla the margin is much higher and perf/watt critical, so I believe they'd use some of the best chips.

 

[GZ007]

My memory is quite weak when it comes to the 1st generation of Tesla variants, but I think the C10x0 weren't based on GT200 but on GT200b@55nm (stands open for correction though). If it should be true though NV itself rates the GTX280 with a 236W maximum power consumption while the GTX285 with a 204W power consumption, despite the latter having slightly higher frequencies than the 280.

If all the above should align, then NV's primary concern for Tesla products is power consumption; especially since the GTX280 had 6pin+8pin connectors (like the coming 480) and not like the 285 2*6pin.

Maybe not power consumption but heat. With 40nm process the chips will have different heat output on same clocks than the 55nm (and with higher leakage it will be more). In the november tesla pdf http://www.nvidia.com/docs/IO/43395/BD-04983-001_v01.pdf the whole thermal specification was missing but power specifications not
Heat could be equaly important in hpc market than power consumption.

 

[Psycho]

So what will they name card numbers 3 and 4 or above? Surely they can produce more than 2 cards.

Depends on when they find the next working board - could be #481 if really, really lucky, or maybe the next will be #712

 

[GZ007]

The 8 pin and 6 pin could be also there for multi tesla cards. If your MB cant provide enough power to the second PCIe u could have full 225W from cables in that case.

 

[Ailuros]

Very possible indeed. But that would also mean they expect to have a revision of GF100 in Q3 that will need much less juice.
Currently I think they just do not have that many chips that would be good enough for a top of the line Tesla, so they do the clever thing and sell those as GeForce 480, while the margins in the product line are still the highest. Later when production matures Tesal is added. The time window in which they can demand a premium is much smaller for the GeForce line, than for the Tesla line.

Let me see the first Tesla C20 variant to ship is the C2050 which is a 448SP chip clocked at 1.2GHz in early Q2 2010, while the C2070@1.4GHz is to ship in Q3 2010: http://www.nvidia.com/object/product_tesla_C2050_C2070_us.html

Now include how critical in the rack mount markets power consumption (and yes GZ007 is right heat is an important factor too since those usually operate on a 24/7 basis) and think that the GeForces will be clocked higher.

The 480 is rumored to have a 280W TDP which is in line with the GTX280 (GT200@65nm, 6pin+8pin, 236W@NV's homesite) vs. the later introduced GTX285 (GT200b@55nm, 2*6pin, 204W@NV's homesite). Yes it's very well possible that they're planning a re spin which is projected for release somewhere in Q3 2010. For the C2050/2070 they state a typical power consumption of 190W on their relevant page and it would have to be for both more or less within those limits.

Just for the record NV's site rates the 285 with a maximum power consumption of 204W vs. 219W for the 275, despite the latter having slightly lower frequencies and both being GT200b chips.

Blazkowicz is also right IMO that they'll select the chips with the highest possible tolerance for the Tesla market. I'm not so sure how you're drawing your conclusions but no IHV could take any risks and carefully virtually hand select chips for markets like that.