Friday, October 15th 2010
NVIDIA to Counter Radeon HD 6970 ''Cayman'' with GeForce GTX 580
AMD is undertaking its product development cycle at a breakneck pace, NVIDIA trailed it in the DirectX 11 and performance leadership race by months. This November, AMD will release the "Cayman" GPU, its newest high end GPU, the expectations are that it will outperform the NVIDIA GF100, that is a serious cause for concern, for the green team. It's back to its old tactics of talking about GPUs that haven't even taken shape, to try and water down AMD's launch. Enter, the GF110, NVIDIA's new high-end GPU under design, on which is based the GeForce GTX 580.
The new GPU is speculated to have 512 CUDA cores, 128 TMUs, and a 512-bit wide GDDR5 memory interface holding 2 GB of memory, with a TDP of close to that of the GeForce GTX 480. In the immediate future, there are prospects of a more realistic-sounding GF100b, which is basically GF100 with all its 512 CUDA cores enabled, while retaining its 384-bit GDDR5 memory interface, 64 TMUs, and slightly higher TDP than that of the GTX 480.
Sources:
3DCenter.org, PCGH
The new GPU is speculated to have 512 CUDA cores, 128 TMUs, and a 512-bit wide GDDR5 memory interface holding 2 GB of memory, with a TDP of close to that of the GeForce GTX 480. In the immediate future, there are prospects of a more realistic-sounding GF100b, which is basically GF100 with all its 512 CUDA cores enabled, while retaining its 384-bit GDDR5 memory interface, 64 TMUs, and slightly higher TDP than that of the GTX 480.
195 Comments on NVIDIA to Counter Radeon HD 6970 ''Cayman'' with GeForce GTX 580
By close i assume they do mean over 300w? whats that a minimum of two 8 pin pci-e connectors?
But still i hope nvidia get going on these fast, more cards from both sides can only mean good things for the consumers (me :laugh:).
8 Pin PCI-E - 150W
6 pin PCI-E - 75W
total 300W :D And that won't be enuff I think so yeah, 2 8 pins!
Nvidia is still trying to appeal to the lowest common denominator by making the fastest card any cost. Sales of the 5XXX series should have taught them that most people are a bit more discerning and will weigh up heat/temps/computer noise and cost.
This thing with a 512 bit bus is going to cost more than the 480 and will most likely be over $800 outside the US. Looks like another fail.
Look it's all, the GTX 480 is so good this and so good that. This new GTX 5blah will so own AMD's next 6series.
Let the adults know when the school yard spat is over and we can talk about realities of performance. Of course the 480 kicks the 5870's ass, it gobbles up much more power to do so.
AMD's task is and has been to deliver more fps per watt. Nvidia has went for all out power.
It'll be the same again. When this mystical card gets released it will probably be very compromised in some areas to allow it to actually work and it'll still be more power hungry.
The reason Nvidia hasn't got the single card performance crown is bacause to date, nobody has got a hold of the power issues. Don't hold your breath for this card.
And my final prediction. AMD 6970 will topple gtx 480 as single fastest chip. Then some time down the line NV will release their next puppy and it will become king, and on and on and on.
This is good news and i hope for the best so we can get some solid competition going, lets get some of these prices down!!!!:rockout:
and i want prive drops,
and that new speculated gtx580 will be damn fast
The only probable GF110 specs are one of these (in order of probability):
1 - 3/2 (three halves) of a GF104, that is 3 GPC (clusters)
specs: 576 SP, 96 TMU, 384 bit, core 750 mhz, < 500 mm^2
performance: GTX480 + 25%
Posibility of a dual GF104 card.
2- GF104 with 4 SIMDs per SM instead of 3 (64 SPs instead of 48), 2 GPC
specs: 512 SP, 64 TMU, 256 bit (it would accompanied with 6 gbps memory for about the same bandwidth as GF100), core 750-800 Mhz, << 400 mm^2
performance: ~ GTX480 +/- 5%
There would be a dual gpu card based on this one.
3- Combination of both, #2 but with 3 GPC or #1 with 4 SIMD.
specs: 768 SP, 96 TMU, 384 bit, core 650-700 Mhz, ~550 mm^2
performance: GTX480 + 50%
#3 becomes posible thanks to the fact that TSMC 40nm is said to have exceeded 55nm yields, and Nvidia not fecking up the fabric like they did with GF100.
Also #2 and #3 can easily exist at the same time, as well as a dual card based on #2, and a card based on the same but with 1 GPC only and 128 bit. That way:
GF110#2 (512 SP part) >>> Bart at a higher cost
GF110#3 (768 SP part) >>> Cayman at a higher cost
2xGF110#2 (2x512 SP) == Antilles at similar cost I don't see Nvidia saying anything anywhere, all I see is a pair of websites speculating, based on the fact that speaking about a posible Nvidia response now that AMD is releasing HD6000 is going to be inflamatory and obtain them a lot of clicks.
And 1x512 SP part should be faster and higher power than Bart so 2x512 SP is hard to make.
That's terrific news about the new GF110 Fermi card. But I read somewhere online stated that GeForce GTX 580 will have 580 Cuda cores not 512 Cuda cores. Here is a statement below:
( NVIDIA announced a next-generation GeForce GTX 500 series , the first high-end models named GeForce GTX 580 based on the Fermi architecture.
The GeForce GTX 580 features 580 CUDA cores to go along with its "580" moniker and have a whopping 2560MB of 384-bit GDDR5 memory
"Although we're very proud of the GTX 480," NVIDIA President and CEO Jen-Hsun Huang said, "the 400 series is merely a tease for what Fermi can really accomplish. When we release the 500 series later this year, I think everyone will be pleasantly surprised. ) By Softpedia News.
Anyone care to explain this....In my opinion I would rather buy a 580 cuda cores card instead of 512 cuda cores card.
GF100: 512 SP, 64 TMU, 64 SFU, 382 bit ==> 530 mm^2
GF104: 384 SP, 64 TMU, 64 SFU, 256 bit ==> 332 mm^2
GF110#2: 512 SP, 64 TMU, 64 SFU, 256 bit ==> (?) 375 mm^2
GF110#3: 768 SP, 96 TMU, 64 SFU, 384 bit ==> (?) 375/2 * 3 = 560 mm^2*
* GT200 was 576 mm^2 and both AMD and Nvidia, as well as TSMC are saying that TSMC's 40nm node is finally as good as 55nm or 65 nm.