NVIDIA TITAN X Review: The Pascal Beast Unleashed

NVIDIA dropped a bomb a couple of weeks back and shocked almost everyone in the enthusiast community with the surprise announcement of the new, Pascal-based TITAN X. If you missed our original coverage, the story went something like this: NVIDIA CEO Jen-Hsun Huang lost a bet, and out of the blue went and gave away a few TITAN X cards at a special AI event that was taking place at Stanford University. That may sound a little far-fetched, but the way this launch shook out, it seems somewhat plausible.

Regardless, as soon as the announcement was posted, we’re sure many of you had two questions in mind – how fast is it? And what’s the deal with the GP102? Thankfully, we can answer both of those questions and more for you today. We’ve just finished putting the brand new NVIDIA TITAN X through its paces and have the full scoop for you laid out on the pages ahead. First up though, some specifications and a quick tour of the card...

NVIDIA TITAN X (Pascal) Specifications & Features

	NVIDIA TITAN X
Graphics Processing Clusters	6
Streaming Multiprocessors	28
CUDA Cores (single precision)	3584
Texture Units	224
ROP Units	96
Base Clock	1417MHz
Boost Clock	1531MHz
Memory Clock (Data rate)	5005MHz (Effective Speed - ~10Gbps)
L2 Cache Size	3072KB
Total Video Memory	12228 MB GDDR5X
Memory Interface	384-Bit
Total Memory Bandwidth	480 GB/s
Texture Filtering Rate (Bilinear)	317.4 GigaTexels/sec
Fabrication Process	16 nm
Transistor Count	12 Billion
Connectors	3 x Display Port, 1 x Dual-Link DVI, 1 x HDMI
Form Factor	Dual Slot
Power Connectors	One 8-Pin, One 6-Pin
Recommended Power Supply	600 Watts
Thermal Design Power (TDP)	250 Watts
Thermal Threshold	94°C
Price	$1299 MSRP - Find Them At NVIDIA.Com

---

In light of its specifications, the new TITAN X seems like an absolute beast. In comparison to a GeForce GTX 1080, for example, the TITAN X has 1,204 additional CUDA cores (3,584 vs. 2,560), more texture units (224 vs. 160), and more ROPs (96 vs. 64). It also has more L2 cache (3,072KB vs. 2,048KB) and a wider memory bus (384-bit vs. 256-bit).

	TITAN X (New)	GTX 1080	GTX 980 Ti	TITAN X (Old)	GTX 980
NVIDIA GPU	GP102	GP104	GM200	GM200	GM204
Architecture	Pascal	Pascal	Maxwell	Maxwell	Maxwell
Number of Cores	3584	2560	2816	3072	2048
Core Clock	1417 MHz	1607 MHz	1000 MHz	1000 MHz	1126 MHz
Memory	12GB	8GB	6GB	12GB	4GB
Memory Clock	10000 MHz	10000 MHz	7000 MHz	7000 MHz	7000 MHz
Memory Interface	384-bit G5X	256-bit G5X	384-bit	384-bit	256-bit
Memory Bandwidth	480 GB/s	320 GB/s	336 GB/s	336 GB/s	224 GB/s
TDP	250 watts	180 watts	250 watts	250 watts	165 watts
Peak Compute	11.0 TFLOPS	8.2 TFLOPS	5.63 TFLOPS	6.14 TFLOPS	4.61 TFLOPS
Transistor Count	12B	7.2B	8.0B	8.0B	5.2B
Process Tech	16nm FinFET	16nm FinFET	28nm	28nm	28nm
MSRP	$1,200	$599	$649	$999	$499

 The new TITAN X, however, has lower base and boost GPU clocks. Its effective data rate for the memory is the same though, and the TITAN X has more memory on board too -- 12GB vs. 8GB on the GeForce GTX 1080. We should note that the GP102 as it exists on the TITAN X is not the full implementation of the chip. Two SMs have been disabled. The full GP102 has a grand total of 3,840 CUDA cores, as evidenced by the Quadro P6000 that was announced a few days ago.


The GPU at the heart of the TITAN X has a base clock of 1,417MHz and a boost clock of 1,531MHz. The GDDR5X memory on the card is clocked at 5GHz for an effective data rate of 10Gbps. At its reference clocks, the TITAN X offers up to 480GB/s of memory bandwidth and a peak texture fillrate of 317.4 GigaTexels/s, all within a 250 watt power envelope. The card requires a single 8-pin and single 6-pin supplemental power feed.

Aesthetically, the new TITAN X borrows heavily from the GTX 1080, minus some silver and plus some additional black. What you see pictured above is the only version of the TITAN X that will be for sale. Partners won't be offering the TITAN X and it will be available only through NVIDIA.com.  The cooler, however, may (or may not) change at some point to remove the GeForce GTX branding.

Like the GTX 1080, the new TITAN X has a faceted fan-shroud with black and silver elements, a thin, sleek modular back-plate, and an aggressive overall design language that looks great to us. Should you use TITAN X cards in SLI, in adjacent slots, the rear portion of the backplate is also easily removable to allow for better airflow between the cards (we have additional pictures in the gallery below if you'd like to see what the back of the card looks like with half the plate removed).

NVIDIA's TITAN X is outfitted with a die cast aluminum body which is machine finished and heat treated for additional strength and rigidity. The top edge of the card we received features the signature, lighted "GeForce GTX" logo, but -- as we've mentioned -- that may not be permanent.

The thermal solution on the TITAN X is just like the 1080's. It features a radial fan and a relatively large vapor chamber linked to a dense heatsink fin array, also like previous-gen flagship cards. The fan pulls air in from above, blows it across the heatsink, and then vents the heated air outside the chassis. At the rear of the card, a second, small heatsink helps wick heat away from the VRM.

The TITAN X features a 7+2-­phase dual-­FET power array for its GPU and memory, extra capacitance in its filtering circuitry, and a power delivery network on its PCB that has been optimized for low impedance. All told, these elements result in increased power efficiency and peak-to-peak voltage noise reduction over previous-gen solutions.

Like NVIDIA's other Pascal-based graphics cards, the outputs on the TITAN X consist of a trio of DisplayPorts, an HDMI 2.0b output, and a dual-link DVI output. The DisplayPorts are 1.2 certified and DP 1.3/1.4 ready, which enables support for 4K displays at 120Hz, 5K displays at 60Hz, and 8K displays at 60Hz (using two cables and multi-stream transport). Up to four display outputs can be used simultaneously for multi-monitor or VR setups.

Enough with the technical jargon; let's hit the benchmarks...