Nvidia Announces GeForce GTX TITAN X based on GP102 GPU with GDDR5X
Read this well, announced, not launched. So out of no where Nvidia announced the 1200 USD GeForce GTX TITAN X based on Pascal architecture. The product is expected to launch August two and will only available from selected system builders.
The announcement is unexpected, though there was talk about the product, nobody expected a release before Christmas. Nvidia has been ramping out Pascal SKUs like crazy lately. The new Titan X will hold the GP102, a 12 billion transistor GPU and it can squeeze out 11 TFLOPS of performance. The GP10 would be huge at almost 600mm^2. It matches the Tesla P100 spec for the bigger part, aside from HBM2 memory.
|
||||
Products | Tesla P100 | GP100 | Titan X | GTX 1080 |
GPU | GP100 (Pascal) | GP100 (Pascal) | GP102 (Pascal) | GP104 (Pascal) |
SMs | 56 | 60 | 56 | 40 |
TPCs | 28 | 30 | 28 | 20 |
FP32 CUDA Cores / SM | 64 | 64 | 64 | 64 |
FP32 CUDA Cores / GPU | 3584 | 3840 | 3584 | 2560 |
Base Clock | 1328 MHz | ~1328 MHz | 1417 MHz | 1607 MHz |
GPU Boost Clock | 1480 MHz | ~1480 MHz | 1531 MHz | 1733 MHz |
Texture Units | 224 | 240 | 224 | 160 |
Memory Interface | 4096-bit HBM2 | 4096-bit HBM2 | 384-bit | 256-bit |
Memory Size | 16 GB | 16 GB | 12 GB | 8 GB |
L2 Cache Size | 4096 KB | 4096 KB | 4096 KB | 2048KB |
Register File Size / SM | 256 KB | 256 KB | 256 KB | 256 KB |
Register File Size / GPU | 14336 KB | 14336 KB | 14336 KB | 10240 KB |
TDP | 300 Watts | ~300 Watts | 250 Watts | 180 Watts |
Transistors | 15.3 billion | 15.3 billion | 12 billion | 7.2 Billion |
Manufacturing Process | 16-nm | 16-nm | 16-nm | 16-nm |
The Pascal-based GP102 features 3,584 shader processor cores, clocked at 1.53GHz (previous-gen Titan X has 3,072 CUDA cores clocked at 1.08GHz) thus 56 streaming multiprocessors, 224 TMUs, 96 ROPs, and a 384-bit GDDR5X memory interface with 12 GB of memory.
- 12-billion transistors
- 11 TFLOPs FP32 (32-bit floating point)a
- 44 TOPS INT8 (new deep learning inferencing instruction)
- 3,584 CUDA cores at 1.53GHz (versus in previous TITAN X)
- High performance engineering for maximum overclocking
- 12GB of GDDR5X memory (480GB/s)
This card is not powered by the same GPU and HBM2 memory, but the GP102 is paired to GDDR5X memory at 10 Gbps. Titan X will be available Aug. 2 for $1,200 direct from NVIDIA.com in North America and Europe, and from select system builders. Availability in Asia will come at a later date.
Nvidia announces GeForce GTX 1060 at $249 for the 6GB models - 07/07/2016 02:41 PM
Moments ago Nvidia announced their GeForce GTX 1060, the new mid-range graphics card is now official. Fairly soon performance reviews will be allowed, today it remains to be just the announcement an...
Nvidia Announces PCIe version Tesla P100 - 06/20/2016 01:01 PM
Nvidia released the PCI-Express version of the P100 in the form of the Tesla P100. The hpc-card was revealed back in April already however was based on a Mezzanine-connector....
Nvidia Announces the GeForce GTX 1070 and 1080 - 05/07/2016 09:11 PM
Right then, time for a proper news post about it. Nvidia hours ago announced the GeForce GTX 1070 and 1080. Both cards are obviously Pascal architecture based and made on a 16nm FinFet design. The ru...
NVIDIA and Samsung Agree to Settle All Outstanding IP Litigations - 05/03/2016 04:59 PM
NVIDIA and Samsung have agreed to settle all pending intellectual property litigation between the two companies, NVIDIA announced today....
Nvidia announces Tesla P100 data-center GPU - 04/05/2016 06:37 PM
Nvidia just announced a HPC data-center module based on Pascal GPU architecture. The Tesla P100 is a daughter add-in board that holds 150B transistors, the GPU itself however will hold 15B transistors...
Senior Member
Posts: 6956
Joined: 2008-10-27
I still don't see it -- although PC Per lists it as GP102.
I also don't see how GP100 isn't suited at all for consumer gaming. The specs of this chip are identical to the PCI-E GP100 board, aside from swapping the HBM2 to GDDR5x.
It's basically a memory controller change, which is the reason why it's got a different chip designation.
Regardless, the point is that there is a bunch of wasted FP64 die space on this card that isn't on the GP104. Which means that Nvidia could refresh with a Ti variant next year with over 4k cores. This is basically the Titan model to the Titan Black/780Ti.
Where did you see shots of the actual die and the blocks identified? And come on, it isn't just 'simply a memory controller change' to go from board mounted GDDR memory to HBM. It is an entirely different design on the die.
Senior Member
Posts: 1255
Joined: 2003-04-26
about right. glad i ditched my overpriced 1080s
Senior Member
Posts: 2449
Joined: 2015-03-20
I was hoped Nvidia will pull HBM2,but CUDA core count at 3584 looks awesome,I will wait on benches in IRAY and Octane too and then I can decide,but looks like will be beast in rendering
Price of 1200USD looks like is still on OK side,I was expected price around £1500-£1800
Thanks,Jura
Senior Member
Posts: 510
Joined: 2009-11-24
NO HBM memory? why
Senior Member
Posts: 13235
Joined: 2004-05-16
I still don't see it -- although PC Per lists it as GP102.
I also don't see how GP100 isn't suited at all for consumer gaming. The specs of this chip are identical to the PCI-E GP100 board, aside from swapping the HBM2 to GDDR5x.
It's basically a memory controller change, which is the reason why it's got a different chip designation.
Regardless, the point is that there is a bunch of wasted FP64 die space on this card that isn't on the GP104. Which means that Nvidia could refresh with a Ti variant next year with over 4k cores. This is basically the Titan model to the Titan Black/780Ti.