Friday, March 8th 2013

NVIDIA Readies Second GK110-based SKU

Following the launch of its consumer-oriented GeForce GTX Titan, NVIDIA is planning its second graphics card based on the 28 nm GK110 silicon, this one for professionals, and featuring in the Quadro family. Likely bearing the model name Quadro K6000, the card is expected to feature 13 of the GK110's 15 streaming multiprocessors (SMXs), which work out to 2,496 CUDA cores, 208 TMUs, 40 ROPs, and a 320-bit wide GDDR5 memory interface, likely holding 5 GB of memory. The configuration could be just about enough for 1 TFLOP/s double-precision floating-point performance, and could hence offer an upgrade over Fermi-based Quadro 6000.


Sources: X-bit Labs, Arab PC World
Add your own comment

20 Comments on NVIDIA Readies Second GK110-based SKU

#1
radrok
Wonder what a full fledged GK110 could achieve, AFAIK there is no full enabled GK110, right?
Posted on Reply
#2
LDNL
Titan X2, x3 the price ... ohh wait that is right when it comes to Quadro :laugh:
Posted on Reply
#3
TheHunter
This looks like a GF110 570GTX variant in the near future.

But if "full" GK110 14SMX costs 1000$, then this crippled 13SMX could very well be a 650-700$ gpu, still to much imo..
Posted on Reply
#4
radrok
by: TheHunter
This looks like a GF110 570GTX variant in the near future.

But if "full" GK110 14SMX costs 1000$, then this crippled 13SMX could very well be a 650-700$ gpu, still to much imo..
I think that if they remove the fancy cooler thingy and reduce the amount of VRAM to 3 gigs they could very well sell one at 400 eur
Posted on Reply
#5
Jack1n
by: TheHunter
This looks like a GF110 570GTX variant in the near future.

But if "full" GK110 14SMX costs 1000$, then this crippled 13SMX could very well be a 650-700$ gpu, still to much imo..
A full GK110 packs 15SMX units,the only card that has all of them enabled is the Tesla variant.
Posted on Reply
#6
radrok
by: Jack1n
A full GK110 packs 15SMX units,the only card that has all of them enabled is the Tesla variant.
http://en.wikipedia.org/wiki/Nvidia_Tesla

AFAIK the highest end of Tesla GPUs is the K20X which has the same SMX units enabled as the GTX Titan, which are 14 (2688 CUDA cores).

There is no 15 SMX unit available if I'm not mistaken.
Posted on Reply
#7
Jack1n
by: radrok
http://en.wikipedia.org/wiki/Nvidia_Tesla

AFAIK the highest end of Tesla GPUs is the K20X which has the same SMX units enabled as the GTX Titan, which are 14 (2688 CUDA cores).

There is no 15 SMX unit available if I'm not mistaken.
Yes that is correct for some reason i thought the k20 had all enabled.
Posted on Reply
#9
jihadjoe
Called the K6000, because it will cost $6000! :laugh:
Posted on Reply
#10
bogami
See how Invidia treats customers and no competition when you can afford to sell the failed processors produced for the big money. What I think: Production of processors avoids making processors failed after the test determines how many were able to reach the desired standards and if a cluster does not work the chip useless for K20 product well then, this is pushed as the best GPU product and sell as much as $ 1000. Already here we suffer for a nice number of cores and now we will be pushed to THIS group failed processors which two clusters inoperative for big money. So far, the TESLA GPU show many weaknesses in the inefficiency of individual cores to Fremen arhitekruri and a strong increase of these is made ​​by several problems such as low FPS in multi GPU option (SLI) has fueled heating due to loading times ... Maybe two processors on the card made ​​better efficiency and, of course, would be $ 1,000 price make it more bearable for second-class chip. I hope you understand the dilemma.:shadedshu
Posted on Reply
#11
Shihabyooo
by: radrok
I think that if they remove the fancy cooler thingy and reduce the amount of VRAM to 3 gigs they could very well sell one at 400 eur
They could probably sell it for less as it is. Could, but [/i]wouldn't[/i]..

by: radrok
Wonder what a full fledged GK110 could achieve, AFAIK there is no full enabled GK110, right?
Wait for the GTX780 (Or Titan 2 or something). Remember Fermi?
GF100 ->GF110.
Posted on Reply
#12
Solidstate89
by: TheHunter
This looks like a GF110 570GTX variant in the near future.

But if "full" GK110 14SMX costs 1000$, then this crippled 13SMX could very well be a 650-700$ gpu, still to much imo..
It's a Quadro card, it'll be more than that. Expect 2-3 grand for it, if not more.

Remember, when you pay for a Quadro or FirePro card you're not paying for the hardware; you're paying for the ISV-certified drivers.
Posted on Reply
#13
TheHunter
^
Im referring to Geforce variant with those or similar specs., of course this Quadro K6000 will be more then 1000$, more like +2.5K $.

by: Jack1n
A full GK110 packs 15SMX units,the only card that has all of them enabled is the Tesla variant.
Yep, that's why i used "" by the word full. ;)
Posted on Reply
#14
NeoXF
by: radrok
I think that if they remove the fancy cooler thingy and reduce the amount of VRAM to 3 gigs they could very well sell one at 400 eur
And call it what? TITAN LE? That would never happen.

Anyway, THEORETICALLY it would have either 2,5, 3,75 or 5GB of VRAM, not 3 or 4, because of the bus.
Posted on Reply
#15
radrok
by: NeoXF
And call it what? TITAN LE? That would never happen.

Anyway, THEORETICALLY it would have either 2,5, 3,75 or 5GB of VRAM, not 3 or 4, because of the bus.
The GTX 580 has a 384bit bus width and there are 3GB versions.

The GTX Titan has a 384bit bus, why is it different? Logic tells me that it can be made 3GB.


http://www.geforce.co.uk/hardware/desktop-gpus/geforce-gtx-580/specifications

http://www.geforce.co.uk/hardware/desktop-gpus/geforce-gtx-titan/specifications

That was an unnecessary and incorrect nitpick :p
Posted on Reply
#18
jihadjoe
by: radrok
The GTX 580 has a 384bit bus width and there are 3GB versions.

The GTX Titan has a 384bit bus, why is it different? Logic tells me that it can be made 3GB.


http://www.geforce.co.uk/hardware/desktop-gpus/geforce-gtx-580/specifications

http://www.geforce.co.uk/hardware/desktop-gpus/geforce-gtx-titan/specifications

That was an unnecessary and incorrect nitpick :p
Unnecesary, perhaps. But he isn't incorrect. If you read the whole of the first post, btarunr said the card would likely be 320-bit, not 384. So the RAM amounts will be more like the 570, less like the 580.

The GTX570 has a 320 bit bus, and there are 2.5GB versions
The 13-sm GK110, if it has a 320 bit bus can easily have 5GB VRAM using chips with double the density compared to what was in the GTX570.
Posted on Reply
#19
HumanSmoke
by: NeoXF
Anyway, THEORETICALLY it would have either 2,5, 3,75 or 5GB of VRAM, not 3 or 4, because of the bus.
Theoretically, the framebuffer could be virtually anything. Between Nvidia's memory controllers being able to support asymmetric memory configurations (different density chips on the same package), and the ability of the GDDR5 controllers to operate in clamshell mode there are a hell of a lot of configurations available.

Wouldn't surprise me if Nvidia loaded the Quadro with a high (and round) number for marketing purposes as well as intended workload.
Posted on Reply
#20
NeoXF
by: HumanSmoke
Theoretically, the framebuffer could be virtually anything. Between Nvidia's memory controllers being able to support asymmetric memory configurations (different density chips on the same package), and the ability of the GDDR5 controllers to operate in clamshell mode there are a hell of a lot of configurations available.

Wouldn't surprise me if Nvidia loaded the Quadro with a high (and round) number for marketing purposes as well as intended workload.
Agreed, but it would most likely be something more 320bit-friendly. So 2,5GB at minimum... actually, if something like this for consumer would ever come to market, I'd thin 5GB would seems fit best against the TITAN's 6GB... well, maybe with 2,5 or 3,75GB as a more "value" version.

All in all, I don't see this happening. But if it would, it would probably be something GTX 700 or 800 related. As would a AMD Radeon HD 8000 or 9000 series, by the end of the year...
Posted on Reply
Add your own comment