Possible NVIDIA GM200 Specs Surface (Not)

[Kaapstad]

Somebody sent our GPU-Z validation database a curious looking entry. Labeled "NVIDIA Quadro M6000" (not to be confused with AMD FirePro M6000), with a device ID of 10DE - 17F0, this card is running on existing Forceware 347.09 drivers, and features a BIOS string that's unlike anything we've seen. Could this be the fabled GM200/GM210 silicon?

The specs certainly look plausible - 3,072 CUDA cores, 50 percent more than those on the GM204; a staggering 96 ROPs, and a 384-bit wide GDDR5 memory interface, holding 12 GB of memory. The memory is clocked at 6.60 GHz (GDDR5-effective), belting out 317 GB/s of bandwidth. The usable bandwidth is higher than that, due to NVIDIA's new lossless texture compression algorithms. The core is running at gigahertz-scraping 988 MHz. The process node and die-size are values we manually program GPU-Z to show, since they're not things the drivers report (to GPU-Z). NVIDIA is planning to hold a presser on the 8th of January, along the sidelines of the 2015 International CES. We're expecting a big announcement (pun intended).

http://www.techpowerup.com/208501/possible-nvidia-gm200-specs-surface.html

Can anyone spot why this is a fake.

Spoiler

When the original Titan came out it ran PCI-E 2.0 unless you used the hack.

If you notice in the link the spec table also has this err new card running PCI-E 2.0

I think all this faker has done is get hold of the original Titan specs and scale things up a bit forgetting about the PCI-E 2.0 lol.

Or I could be wrong.

 

[Rroff]

Getting PCI-E 2/3 working properly has always been a little odd at times.

 

[Boomstick777]

If the spec is legit for the workstation card, having a relatively high clock bodes well for the Geforce Variant. Guessing 6GB Vram 1000+ Turbo on the core, 96 ROP gaming monster !!

Bring it on !!

 

[Kaapstad]

I think if the GTX 980 comes with PCI-E 3.0 as standard they are not going to allow anything less on the Maxwell Titan.

Specification:-
- GeForce GTX 980
- GPU: GeForce GTX 980 (GM204)
- Core Base Clock: 1127MHz
- Core Boost Clock: 1216MHz
- Memory Clock: 7010MHz
- Memory Size: 4096MB GDDR5
- Bus Type: PCI Express 3.0
- Memory Bus: 256-bit
- CUDA Cores: 2048
- DirectX 12: Yes
- DVI Port: 1x Dual-Link DVI, 3x DisplayPort & 1x HDMI
- DisplayPort: Yes
- HDCP: Yes
- HDMI: Yes
- Dimensions: 268.6 x 98.4 x 40 mm
- Power: 2x 6-Pin
- 500W PSU Required
- 178W TDP
- Warranty: 3yr

 

[Gregster]

It could be the Quadro specs in truth but I doubt it.

 

[pgi947]

Wasn't the 'hack' for nvidia PCIE 3.0 purely for SB-E? Don't recall Z77 having issues running 3.0 at the time.

As for the PCIE validation, its just reporting what the users system PCIE lanes are running at, if hes not on IB/IB-E or above then they will likely be running 2.0

Not saying its legit or anything, as it has the memory down as running 6600mhz, would have been expecting 7000mhz tbh.

 

[Kaapstad]

Wasn't the 'hack' for nvidia PCIE 3.0 purely for SB-E? Don't recall Z77 having issues running 3.0 at the time.

As for the PCIE validation, its just reporting what the users system PCIE lanes are running at, if hes not on IB/IB-E or above then they will likely be running 2.0

Not saying its legit or anything, as it has the memory down as running 6600mhz, would have been expecting 7000mhz tbh.

Would anyone in their right mind test it on a rig not capable of PCI-E 3.0

Having said that I have seen some of the tech sites do much worse.

 

[andybird123]

It would make the die size something like 600-625mm
That probablty puts it at around $100 to produce given that 28nm must have dropped in cost by now, depends what yields are like for such a monster

 

[Orangey]

Is that a normal profit margin? They'll be doing well when it comes out, people are thirsty for 4K and new shiny, so probably sell as well or better than Titan.

 

[Rroff]

It would make the die size something like 600-625mm
That probablty puts it at around $100 to produce given that 28nm must have dropped in cost by now, depends what yields are like for such a monster

Engineering boards for gm200 are supposedly based around a 20nm core, would be a bit odd though as for a professional card/hardware that might end up in enterprise use they'd usually want good availability.

 

[Orangey]

Can you even get a single chip off a 20nm wafer for a 600mm2 part

 

[andybird123]

Can you even get a single chip off a 20nm wafer for a 600mm2 part

It wouldnt still be 600mm at 20nm would it
If it is 20nm that would make it something like ~450mm

Theres so many different rumours flying around, it is difficult to even guess which it will be, i'd assumed the lack of a 20nm-HP process would make it unsuitable for high power gpu parts

 

[andybird123]

Is that a normal profit margin? They'll be doing well when it comes out, people are thirsty for 4K and new shiny, so probably sell as well or better than Titan.

Finalised wafers cost around $5000, you would get about 89-93 ~600+mm2 dies from a 300mm wafer, but then you need to factor in yields

Last i heard, AMD pay for working cores but a higher price per core, where as Nvidia buy wafers and do all the testing and binning themselves, so if nvidia get the process working for them they can actually drastically lower their costs, equally if they get it wrong their costs rocket.

They could be as low as $60 per core for titan2, and $45 for gtx970/980

I'm extrapolating, so basically this is guess work, but I would be very surprised if the actual values aren't somewhere between the two

 

[Quartz]

If you notice in the link the spec table also has this err new card running PCI-E 2.0

It could just be plugged into an older motherboard with an older CPU.

ETA there is already a Nvidia Quadro card with 12 GB VRAM, the K6000.