NVIDIA GeForce GTX 1070: TITAN X performance at less than half the cost
A golden age for PC gaming
This is a glorious time to be a PC gamer. When NVIDIA first unveiled the GeForce GTX 1080 and 1070 in early May, it wasted no time in talking up the new cards as being faster than last year’s single-GPU king, the GeForce GTX TITAN X. And as we saw in our review, the GeForce GTX 1080 more than lived up to those claims, a fairly impressive feat in and of itself, considering that the new Pascal-based consumer flagship costs only US$699 compared to the TITAN X’s US$999.
But how about the GeForce GTX 1070? NVIDIA ended up walking back its claims about the card a little, clarifying that the latter card would offer TITAN X levels of performance, instead of being outright faster. Be that as it may, it’s still quite a tantalizing prospect, given that the card costs just US$449 (approx. RM1,844) for the Founders Edition. Ultimately, it looked like NVIDIA had just made a GeForce GTX TITAN X at less than half the price, complete with all the goodies and new features – Simultaneous Multi-Projection, GPU Boost 3.0, and Fast Sync, to just name a few – that came with the Pascal architecture and 16nm FinFET manufacturing process.
We mentioned in our GeForce GTX 1080 review that NVIDIA might just have defined a new performance class with that card, where ultra-enthusiast performance was trickling down to so-called 'high-end' segment. NVIDIA looks to continue this trend with the GeForce GTX 1070 – the company’s GTX x70 cards have traditionally straddled the line between the mainstream and enthusiast markets, but the GeForce GTX 1070 looks to be doing this in a whole new way.
For the first time, gamers who buy a GTX x70 card can look forward to – at the very least – similar performance to last generation’s undisputed performance flagship. If that’s not the very definition of progress, we don’t know what is. Once again, the high-end gaming segment is showing why it continues to be bright spot in an otherwise gloomy PC market.
Meet the GeForce GTX 1070
The GeForce GTX 1070 is based on the same GP104 GPU as the GTX 1080, but with certain functional units disabled. While the latter card boasted a total of 20 Streaming Multiprocessors (SMs), the GeForce GTX 1070 has just 15 SMs. Each SM houses 128 CUDA cores, so this brings the number of CUDA cores to 1,920, 25 percent fewer than the flagship Pascal card.
NVIDIA didn’t provide us with a block diagram for the GeForce GTX 1070, but we’ve reproduced that of the GeForce GTX 1080 here for easier reference.
Each Graphics Processing Cluster (GPC) features 16 render output units (ROPs) and 40 texture mapping units (eight in each SM). But because the GeForce GTX 1070 actually still has 64 render output units (ROPs) – the same as its bigger brother – and 120 texture mapping units (TMUs), it looks like NVIDIA hasn’t disabled a single GPC entirely on the GTX 1070. Instead, it appears to have selectively put certain SMs out of action in each GPC, resulting in the same number of ROPs as the GeForce GTX 1080, but 25 percent fewer TMUs.
And thanks to the new Pascal architecture, the GeForce GTX 1070 is equipped with fairly aggressive clock speeds as well, to the tune of a 1,506MHz base clock and 1,683MHz boost clock.
Unlike the GeForce GTX 1080 however, the GTX 1070 doesn’t use GDDR5X memory, relying instead on the more traditional GDDR5 variant. Its 8GB of GDDR5 memory is clocked at 8,000MHz, but both Pascal cards share the same 256-bit memory bus. For the GeForce GTX 1070, this translates into a total available bandwidth figure of 256GB/s, a fair bit behind the GTX 1080’s 320GB/s.
That aside, NVIDIA says that the 8Gbps data rate is the highest memory speed of any GDDR5 GPU on the market. Furthermore, because of Pascal’s improved memory compression engine that offers far more efficient color compression, the GeForce GTX 1070 can avail itself of around 20 percent more effective memory bandwidth.
More impressive is the card’s Thermal Design Power (TDP), which is the maximum amount of power the cooling system is expected to dissipate. This sits at 150W for the GeForce GTX 1070, a mere 5W more than GeForce GTX 970. And considering how much more powerful the GTX 1070 is than its Maxwell-based predecessor, we begin to see the real benefits of moving to the smaller and more efficient 16nm process node. We’d also like to point out that the GeForce GTX Titan X has a TDP of 250W, and considering that the GTX 1070 delivers similar levels of performance, the huge stride that Pascal has taken in terms of efficiency gains becomes truly apparent.
Here’s a table comparing the specifications of both Pascal cards.