NVIDIA Publicizes “NVIDIA Titan V” Video Card: GV100 for $3000, On Sale Now
|December 8, 2017||Posted by BLOGGER under HACKER-TECH|
Out of nowhere, NVIDIA has revealed the NVIDIA Titan V at the present time at the 2017 Neural Data Processing Methods convention, with CEO Jen-Hsun Huang flashing out the card on stage. A mere 7 months after Volta changed into offered with the Tesla V100 accelerator and the GV100 GPU internal it, NVIDIA continues its breakneck rush by releasing the GV100-powered Titan V, on hand for sale at the present time. Aimed at a decidedly extra compute-oriented market than ever earlier than, the 815 mm2 behemoth die that is GV100 is now on hand to the broader public.
|NVIDIA Compute Accelerator Specification Comparability|
|Titan V||Tesla V100
|Memory Clock||1.7Gbps HBM2||1.75Gbps HBM2||1.4Gbps HBM2||eleven.4Gbps GDDR5X|
|Memory Bus Width||3072-bit||4096-bit||4096-bit||384-bit|
|Single Precision||thirteen.Eight TFLOPS||14 TFLOPS||9.three TFLOPS||12.1 TFLOPS|
|Double Precision||6.9 TFLOPS
|One hundred ten TFLOPS||112 TFLOPS||N/A||N/A|
|Manufacturing Direction of||TSMC 12nm FFN||TSMC 12nm FFN||TSMC 16nm FinFET||TSMC 16nm FinFET|
For the spec sheet we indulge in long gone forward and lined it up in opposition to NVIDA’s diverse Pascal cards, and for precise motive. While the Titan series of cards would per chance also impartial indulge in began life as a prosumer card in 2013, since then NVIDIA’s GPU designs indulge in change into an increasing number of divergent between compute and graphics. And although the earlier Titan Xp changed into constant with the extra graphics-centered GP102 GPU, the card itself changed into primarily (but now not totally) pitched as an entry-stage compute card, for prospects who wished a (moderately) cheap design to execute FP32 compute and neural network inferencing in workstations and minute clusters.
The Titan V, by extension, sees the Titan lineup in the slay switch loyalties and begin the usage of NVIDIA’s excessive-quit compute-centered GPUs, in this case the Volta architecture primarily primarily based V100. The tip consequence is that moderately than being NVIDIA’s top prosumer card, the Titan V is decidedly extra centered on compute, particularly as a consequence of the combo of the worth label and the extraordinary characteristic situation that comes from the usage of the GV100 GPU. Which isn’t to command that you just can too’t execute graphics on the card – this is aloof very noteworthy a video card, outputs and all – but NVIDIA is in the beginning promoting it as a workstation-stage AI compute card, and by extension specializing in the GV100 GPU’s extraordinary tensor cores and the extensive neural networking performance advantages they give over earlier NVIDIA cards.
In this sense the Titan V is a return to make of kinds to the expert side of prosumer for the Titan household. Certainly one of the most traditional claims to fame for the normal Titan changed into its excessive performance in genuinely expert FP64 compute workloads, one thing that changed into lost on the later Titan X and Titan Xp. By switching to NVIDIA’s genuinely expert excessive-quit compute GPUs, the Titan V regains its beforehand lost compute capabilities, the complete whereas also gaining all the compute capaiblities NVIDIA has offered since then. It’s no mistake that Jen-Hsun offered the card at a neural networking convention, because it is a extensive chunk of the expert computing viewers that NVIDIA is concentrating on with the card.
Curiously, evaluating it to the PCIe Tesla V100, I’m taken aback by impartial how shut the cards are in aspects and performance. NVIDIA has confirmed that the Titan V gets the GV100 GPU’s stout, unrestricted FP64 compute and tensor core performance. To the top seemingly of our records (and from what NVIDIA will comment on) it doesn’t appear that they’ve artificially disabled any of the GPU’s core aspects. What does separate the Titan from the Tesla then from a performance standpoint is moderately straightforward: memory capacity, memory bandwidth, and the lack of NVLink functionality. There are also a option of smaller variations between the cards that abet to grunt apart them between server and workstation – a lot like passive versus full of life cooling, NVLink, and the increase policies – but in every other case for prospects who are running a minute quantity of cards, the Titan V’s characteristic situation is remarkably shut to the noteworthy costlier Tesla V100’s, which is a genuinely attention-grabbing pattern because it goes to existing impartial how assured NVIDIA is that this won’t undermine Tesla gross sales.
Transferring on and diving into the numbers, Titan V aspects eighty streaming multiprocessors (SMs) and 5120 CUDA cores, the identical quantity as its Tesla V100 siblings. The diversities arrive with the memory and ROPs. In what’s clearly a salvage allotment for NVIDIA, undoubtedly one of the most card’s four memory partitions has been decrease, leaving Titan V with 12GB of HBM2 linked thru a 3072-bit memory bus. As each memory controller is related to a ROP partition and 768 KB of L2 cache, this in flip brings L2 all the design in which down to four.5 MB, as effectively as reducing down the ROP depend.
Via clockspeeds, the HBM2 has been downclocked a itsy-bitsy bit to 1.7GHz, whereas the 1455MHz boost clock genuinely fits the 300W SXM2 variant of the Tesla V100, even supposing that accelerator is passively cooled. Particularly, the amount of tensor cores indulge in now not been touched, even supposing the legitimate One hundred ten DL TFLOPS rating is decrease than the 1370MHz PCIe Tesla V100, because it would seem that NVIDIA is the usage of a clockspeed decrease than their boost clock in these calculations.
For the card itself, it aspects a vapor chamber cooler with copper heatsink and sixteen energy phases, fascinated by the 250W TDP that has change into identical outdated with the one GPU Titan models. Output-wise, the Titan V brings three DisplayPorts and 1 HDMI connector. And as for card-to-card dialog, PCB itself appears to indulge in NVLink connections on the highest of the PCB, but these discover to had been intentionally blocked by the screen to prevent their employ and are presumably disabled.
As talked about earlier, NVIDIA is unsurprisingly pushing this as a compute accelerator card, particularly focused on that Titan V aspects tensor cores, and keeping the TITAN branding rather than GeForce TITAN. But there are those of us who know better than to rating americans won’t tumble $3000 to employ one of the most smartly-liked Titan card for gaming, and whereas gaming is now not the fundamental (and even secondary) focal level of the card, you also is now not going to rating NVIDIA denying it. In that sense the Titan V is going to be handled as a jack-of-all-trades card by the company.
To that quit, no gaming performance files has been disclosed, but NVIDIA has confirmed that the card uses the frequent GeForce driver stack. Now whether those drivers indulge in genuinely be optimized for the GV100 is one other subject fully; Volta is a brand unique architecture, markedly so at times. Speaklng totally off the cuff here, for graphics workloads the card has extra sources than the Titan Xp in practically each meaningful metric, on the opposite hand additionally it is a smaller distinction on paper than you might have confidence.
As for NVIDIA’s supposed market of compute and AI customers, the Titan V will likely be supported by NVIDIA GPU Cloud, which entails a option of deep studying frameworks and HPC-related tools.
If the golden screen didn’t already imply so, the Titan V would per chance also be carving out a brand unique peep-watering be conscious level, losing in at $2999 and on sale now at the NVIDIA retailer. NVIDIA has, thus some distance, been promoting Tesla V100 products as quick as they can set them, so I am now not going to be taken aback if the Titan V sees a equivalent fate. The $3000 be conscious label is moderately excessive, even by Titan requirements, but with the rare Tesla V100 PCIe card going for spherical $10,000, the Titan V is markedly more cost effective. No doubt in some respects I am taken aback NVIDIA is promoting a GV100 card for therefore itsy-bitsy; these are GV100 salvage components that don’t fabricate the decrease for Tesla – so the different would be throwing them away – on the opposite hand it impartial goes to existing how assured NVIDIA is that it’s now not going to undermine the Tesla household.
At any rate, for NVIDIA expert customers who had been seeking to dip their toes into Volta but did now not desire a stout-fledged Tesla card, the Titan V is clearly going to be a favored card. Over the excellent two years NVIDIA’s AI efforts had been firing on all cylinders, and by bringing a GV100 card all the design in which down to impartial $3000, request to rating them crack originate the market that noteworthy extra. I dare command the hypothesis of the “prosumer” Titan has died with this card, but for the growing expert compute market, this appears to be like to be precisely the roughly card that plenty of developers had been waiting for.