Nvidia has finally revealed the secret of its latest graphics cards and no we are not talking about the GTX 1080 Ti. While that graphics card is scheduled to come out soon, fact remains that it is a poorly kept secret and everyone knows it will make an appearance sometime in 2017. We are speaking of the GP100 Pascal architecture and the Quadro GP100 – first PCIe graphics card in the Pascal architecture for the Quadro range. Some might argue that the Tesla P100 is the first in the Pascal line-up but this isn’t a commercial card since it is only available on Nvidia DGX-1 server, which by the way costs a whopping $129,000.
For those of you unaware of the Tesla P100, this specific graphics card is proprietary from Nvidia and it works only on its own servers, more specifically the DGX-1 server. Now, the first editions of this card relied on custom NVLink mezzanine connectors from Nvidia instead of a PCIe slot. A little while later though Nvidia did introduce the PCIe variant but without any display output. If Nvidia can make a PCIe version of their proprietary graphics card, then it stands to reason that they can add display outputs and that is what the GP100 brings to the consumers. Think of it as a toned done version of the P100 Tesla that operates off a PCIe bridge and has its display outputs.
Where Tesla is predominantly a computing based solution, Quadro from Nvidia is geared more towards professional multimedia and graphics works such as 3D rendering, CAD/CAM, video editing and a lot more in between. Now while the region of overlap between graphics and multimedia work with gaming has grown, there is still a clear demand for the gaming series of graphics card as well as the QUADRO range of professional graphics cards.
Now the next question about the $6000 card is whether it trumps the previous version of QUADRO, the P6000 or does it come close to its performance? Let’s find out.
New Memory Module
If you were to create a Venn diagram showcasing the areas where the GP100 overlaps on the performance of the P6000, it would have a large area of intersection but the one place where it would stand far apart is in the memory module. The GP100 relies on HBM2 memory rather than GDDR5X and this gives it higher memory bandwidth. At 720GB/s, it is way higher than the P6000’s 480GB/s transfer rates. However, memory bandwidth isn’t the only thing that defines a graphics card and it is in other areas that the figures get a little confusing.
The P6000 features a fully functional GP102 GPU, which means it uses the complete 3840 CUDA cores at a clock speed of 1530 MHz thus producing around 12 TFLOPS. To do this it uses 24GB of GDDR5X memory. The Quadro GP100 however packs in only 16GB of HBM2 memory with 3584 CUDA cores operating at around 1430 MHz that together deliver around 10.3 TFLOPS of performance. Basically, you get greater memory bandwidth but less computing power and memory. Because of the tradeoffs though, the GP100 uses significantly less power as opposed to the P6000.
At present the P6000 Quadro retails for about $5000, which is a significant slash in prices from the original price of $7,000. It is safe to assume that the Quadro GP100 will perfectly slot into the $7000 price range though going by the initial stats, but considering the slight drop in overall performance, it is possible for a lower price tag. The P6000 on the other hand should continue to retail at around the $5000 mark even with the GP100 launch since it is still a quite capable card that matches the GP100 blow for blow, outdoing it some performance segments.
Now, being a gaming oriented website, all this talk of QUADRO is immaterial to us unless the GP100 technology is scheduled to make an appearance in the GTX commercial gaming series. Can it make an appearance as a Titan? Well, as it stands this seems highly unlikely. The Titan Pascal X is already giving phenomenal performance with its 12GB GDDR5X GP102 processor that produces around 11 TFLOPS. The need for higher performance of memory bandwidth at the moment is limited and companies that are interested in this sort of jump in memory performance are limited. In other words, GP100 might not receive a consumer variant anytime soon simply because the Titan XP performance is almost at par with it.
No graphics card article is ever complete without discussing AMD cards. Good news for AMD lovers, the Vega will release with HBM2 memory with 8GB in total. The reason for such limited releases in this particular memory technology is the price. More expensive than GDDR5 memory, AMD trying to keep true to its affordable nametag, packs in only two stacks of 4GB HBM2 memory. This does beg the question as to why Nvidia could not reduce its memory stacks on the GP100 for a slightly lower memory module with 8GB in total? This card could have outgrown the popularity of the P6000 with its lower price tag plus equal performance.
As it stands, Nvidia probably will not pump out another graphics card with HBM2 memory module at least until AMD comes out with its Volta model, which is scheduled for the later part of 2018.
Performance Of The P6000 And Why It Won’t Be Outdated Because Of GP100
Speaking in terms of pure gaming, professional graphics cards were never really designed with this intention. Yes, they have similar stats and at times even better parts but they are not game optimized – this is the number one complaint we hear from gamers. Well, if benchmarks are anything to go by, the P6000 blasts high end gaming GPUs to smithereens.
A popular choice among benchmarks, the NVIDIA TITAN X Pascal actually falls way behind the P6000 QUADRO in gaming benchmarks. This is not surprising though, since the TITAN X Pascal has only 3548 cores while the GP102 based card has all its CUDA cores enabled. When it comes to gaming parlance, the full gamut of cores and superior architecture together help the P6000 zoom past the Titan X Pascal in games.