A Beautiful Graphics Card
It was a surprise when it launched, it was a surprise when it showed up on our door. How about a review?
As a surprise to nearly everyone, on July 21st NVIDIA announced the existence of the new Titan X graphics cards, which are based on the brand new GP102 Pascal GPU. Though it shares a name, for some unexplained reason, with the Maxwell-based Titan X graphics card launched in March of 2015, this is card is a significant performance upgrade. Using the largest consumer-facing Pascal GPU to date (with only the GP100 used in the Tesla P100 exceeding it), the new Titan X is going to be a very expensive, and very fast gaming card.
As has been the case since the introduction of the Titan brand, NVIDIA claims that this card is for gamers that want the very best in graphics hardware as well as for developers and need an ultra-powerful GPGPU device. GP102 does not integrate improved FP64 / double precision compute cores, so we are basically looking at an upgraded and improved GP104 Pascal chip. That’s nothing to sneeze at, of course, and you can see in the specifications below that we expect (and can now show you) Titan X (Pascal) is a gaming monster.
| Titan X (Pascal) | GTX 1080 | GTX 980 Ti | TITAN X | GTX 980 | R9 Fury X | R9 Fury | R9 Nano | R9 390X | |
|---|---|---|---|---|---|---|---|---|---|
| GPU | GP102 | GP104 | GM200 | GM200 | GM204 | Fiji XT | Fiji Pro | Fiji XT | Hawaii XT |
| GPU Cores | 3584 | 2560 | 2816 | 3072 | 2048 | 4096 | 3584 | 4096 | 2816 |
| Rated Clock | 1417 MHz | 1607 MHz | 1000 MHz | 1000 MHz | 1126 MHz | 1050 MHz | 1000 MHz | up to 1000 MHz | 1050 MHz |
| Texture Units | 224 | 160 | 176 | 192 | 128 | 256 | 224 | 256 | 176 |
| ROP Units | 96 | 64 | 96 | 96 | 64 | 64 | 64 | 64 | 64 |
| Memory | 12GB | 8GB | 6GB | 12GB | 4GB | 4GB | 4GB | 4GB | 8GB |
| Memory Clock | 10000 MHz | 10000 MHz | 7000 MHz | 7000 MHz | 7000 MHz | 500 MHz | 500 MHz | 500 MHz | 6000 MHz |
| Memory Interface | 384-bit G5X | 256-bit G5X | 384-bit | 384-bit | 256-bit | 4096-bit (HBM) | 4096-bit (HBM) | 4096-bit (HBM) | 512-bit |
| Memory Bandwidth | 480 GB/s | 320 GB/s | 336 GB/s | 336 GB/s | 224 GB/s | 512 GB/s | 512 GB/s | 512 GB/s | 320 GB/s |
| TDP | 250 watts | 180 watts | 250 watts | 250 watts | 165 watts | 275 watts | 275 watts | 175 watts | 275 watts |
| Peak Compute | 11.0 TFLOPS | 8.2 TFLOPS | 5.63 TFLOPS | 6.14 TFLOPS | 4.61 TFLOPS | 8.60 TFLOPS | 7.20 TFLOPS | 8.19 TFLOPS | 5.63 TFLOPS |
| Transistor Count | 11.0B | 7.2B | 8.0B | 8.0B | 5.2B | 8.9B | 8.9B | 8.9B | 6.2B |
| Process Tech | 16nm | 16nm | 28nm | 28nm | 28nm | 28nm | 28nm | 28nm | 28nm |
| MSRP (current) | $1,200 | $599 | $649 | $999 | $499 | $649 | $549 | $499 | $329 |
GP102 features 40% more CUDA cores than the GP104 at slightly lower clock speeds. The rated 11 TFLOPS of single precision compute of the new Titan X is 34% higher than that of the GeForce GTX 1080 and I would expect gaming performance to scale in line with that difference.
Titan X (Pascal) does not utilize the full GP102 GPU; the recently announced Pascal P6000 does, however, which gives it a CUDA core count of 3,840 (256 more than Titan X).
A full GP102 GPU
The complete GPU effectively loses 7% of its compute capability with the new Titan X, although that is likely to help increase available clock headroom and yield.
The new Titan X will feature 12GB of GDDR5X memory, not HBM as the GP100 chip has, so this is clearly a unique chip with a new memory interface. NVIDIA claims it has 480 GB/s of bandwidth on a 384-bit memory controller interface running at the same 10 Gbps as the GTX 1080.
Other than these changes, and corresponding improvements in texture units and ROP count, there really isn’t anything architecturally different in the Pascal-based Titan X over a GeForce GTX 1080. Just more, better and faster. If you are new to NVIDIA’s latest Pascal architecture, product features and what the move to 14nm nets them, you definitely should read our GeForce GTX 1080 review that covers all of that!
What will you be asked to pay for this performance? $1200, going on sale today, and only on NVIDIA.com, at least for now. Considering the prices of GeForce GTX 1080 cards with such limited availability, the $1200 price tag MIGHT NOT seem so insane. That's higher than the $999 starting price of the Titan X based on Maxwell in March of 2015 - the claims that NVIDIA is artificially raising prices of cards in each segment will continue, it seems.
The NVIDIA Titan X (Pascal) Graphics Card
Our time was short with the new Titan X, as our team prepares for a three week whirlwind of events, but we wanted to get a quick review of this beast out the door ASAP.
The new Titan X features the same design language started with the GTX 1080, a rif on the now aging design for NVIDIA reference products. This includes a blower style cooler with an illuminated GeForce GTX logo along the top of the card (interestingly, one of only a few places I see referencing GeForce with this product) and a window to see the heatsink under the shroud.
Rotating the card around the back we find a full cover backplate on the Titan X with an optional segment on the back half you can remove to improve airflow on adjacent graphics cards in SLI. The backplate even has a custom Titan X stamp on it.
Though the shroud design is shared with the GTX 1080, the Titan X goes with a black out color scheme and a chrome “TITAN X” logo along the front.
Display connectivity remains unchanged: three full size DisplayPort connections, one HDMI 2.0a and a dual-link DVI connection for legacy displays.
With a 250 watt TDP, the card includes both a 6-pin and an 8-pin external power connection. This is more than enough to hit 250 watts but allows the card to draw as much as 300 watts when overclocked.
Titan X (Pascal) includes a set of SLI connections to support the new high bandwidth SLI connections, though still only in 2-Way SLI officially.
Ryan, too bad there is no
Ryan, too bad there is no benchmark of other games such as Six Siege, Overwatch etc.
Think that many of these games is crucial to people if they are seriously buying this GPU.
I know you can’t benchmark every game, but should atleast go for the most popular ones.
Ryan, first off – THANK YOU!
Ryan, first off – THANK YOU! You and your staff did an outstanding job putting together this review. I can easily see there was a lot of work done here. I also understand that not every review is perfect, so I may be a bit more forgiving with regards to any mistakes made – though I really didn’t see any and you approached this review with a ‘just the facts’ mentality. Some things I would like to make note of based upon the information and data provided within your Titan X review, other reviews I have read thus far, and my current PC hardware configuration (two GTX980Ti cards in SLI):
1) Titan X performance is near the performance of two GTX980Ti cards in SLI, let alone two GTX980 cards, which by the way can’t even achieve correct playable frame rates at 4K resolutions due more to the limitation of the VRAM (only 4GB each card).
2) Almost 50% of games today do not scale well with 2-way SLI. 3-way and higher is even worse. This alone is a valid argument for those seeking the best performance without all the technical issues that SLI induces to buy a Titan X. Using a single card means no micro stutter, frame rendering lag, required need for a SLI HB bridge, and of course the fact that double the performance is not achieved in 99% of games currently on the market.
3) A single GTX1080 can not play a vast majority of games at 4K resolutions without having to turn down some settings, and buying two GTX1080 cards to do so will cost you as much as a single Titan X AND you will still have the issues induced by SLI, especially more so with DX12 games.
Based upon these observations, one would conclude that if you are an avid enthusiast PC gamer and play games at the higher resolutions, the Titan X is the best buy for the returned level of performance and least amount of technical issues and limitations associated with running two or more cards in an SLI configuration. One could also argue that if there were a need to ‘grow’ in performance capability, then worst case you could always add a second Titan X card 😉
FP16 performance in GP102 is
FP16 performance in GP102 is just 1.5% of FP32 performance.
This year GPUs will be up to twice as efficient!