Last week, NVIDIA unveiled their plans for the GeForce GTX 1060 Founders Edition, a limited run of their custom cooler design that will be eventually replaced by the (arguably) better designed coolers from their partners. The GTX 1060 is the company’s answer to AMD’s Radeon RX 480, launched just a few weeks ago. Aside from the kick-ass Millennium Falcon-like design on the front, the GTX 1060 has a reasonably low price (for AIB cards) set at $249 for a 6GB model, while the Founders Edition you see here will go on sale later this month for $299. Is it enough to run up against Polaris 10? I’m not sure yet. NVIDIA’s always been able to pull out a mid-range surprise when cornered, but this time it’s not exactly an apples to oranges comparison.
NVIDIA GeForce GTX 1060 (GP106) comparison
|GTX 980||GTX 1060||Radeon RX 480||GTX 970||GTX 960|
|GPU family name||Maxwell||Pascal||Polaris 10||Maxwell||Maxwell|
|Single-precision throughput||4.6 TFLOPS||3.8 TFLOPS||5.1 TFLOPS||3.5 TFLOPS||2.3 TFLOPS|
|Base clock||1127 MHz||1506 MHz||1120 MHz||1050 MHz||1127 MHz|
|Boost clock||1216 MHz||1709 MHz||1266 MHz||1178 MHz||1178 MHz|
|Shader module count||16||10||36||13||8|
|Memory||4GB GDDR5||6GB GDDR5||4/8GB GDDR5||4GB GDDR5||2GB GDDR5|
|Memory bus width||256-bit||192-bit||256-bit||256-bit||128-bit|
|Memory bandwidth||224 GB/s||192 GB/s||256 GB/s||224 GB/s||112 GB/s|
|Outputs||Displayport 1.2a, HDMI 2.0, DVI||Displayport 1.4, HDMI 2.0b, DVI||Displayport 1.4, HDMI 2.0b, DVI||Displayport 1.2a, HDMI 2.0, DVI||Displayport 1.2a, HDMI 2.0, DVI|
|HDCP 2.2 support||Yes||Yes||Yes||Yes||Yes|
|Thermal design power (TDP)||165W||120W||150W||145W||120W|
|Power connectors||2x 6-pin||1x 6-pin||1x 6-pin||2x 6-pin||1x 6-pin|
On paper, the GTX 1060 doesn’t seem like a solid replacement for the GTX 980, which has recently dropped down as low as $320 as US-based retailers start to clear out stock. It also seems to be just ahead of the GTX 970, as the single-precision throughput only increases by 3.0 GFLOPS, while the core count decreases. There are hefty jumps in clock speed to make up for the deficit, and there’s an extra 2GB of VRAM to make up for the GTX 970’s now-baseline 4GB. Against the Radeon RX 480, it doesn’t look that good, doesn’t it? But we’ve seen time and again how NVIDIA’s architecture seems to extract more performance with less execution units, and AMD’s GCN architecture has had an atypically low performance-per-clock ratio in the past – it’s always born a high rating in terms of theoretical throughput, but resource utilisation has been an issue for gaming workloads. This might not be a clear-cut race for either card, but NVIDIA has the benefit of stronger brand recognition to help it along.
However, when comparing the GTX 960, we see some hefty improvements waiting in the wings. There’s a 65% jump in raw throughput on the cards, more memory bandwidth, and more VRAM, hinting that NVIDIA will be pushing the GTX 1060 as their card for everyday 1080p and 1440p gaming. The GTX 960 was a great card when it launched, but it’s been getting slower as more games push up the system requirements, and the low memory bandwidth did it no favours. It’s finally time for the people’s champion to retire. At 3.42% of all Steam owners, nearly six million people own and use a GTX 960.
What is both interesting and puzzling about the reference design for the GTX 1060 is the placement of the power connector. In previous board designs, and with literally every single other design from NVIDIA’s and AMD’s partners, the connector has been mounted on the circuit board directly, soldered in place. There are good reasons for that placement, notably that it means you won’t be breaking or bending the cooler shroud itself if you end up reinserting the connector often. You’re also able to install custom coolers of your own choosing, including full coverage water blocks.
With the GTX 1060 Founders Edition, the connector on the board is the same size as a four-pin PWN fan connector. There’s an extension lead that plugs onto those pins and runs into the cooler shroud, where the power connector is located on the end of the cooler. NVIDIA has definitely done this for aesthetic reasons, because it makes more sense to have it on the top of the board rather than at the end, but it means that you’ll have to make your own replacement connectors if you end up installing a different cooler.
There’s also no SLI on this product. That’s a big break from previous designs at this price point. Even if most people were never going to use it anyway, there’s still more than a handful of people out there who would buy one of these cards now, and nab a second one on a sale a year down the line to boost performance. The GeForce GTX 960 supported two-way SLI, and so did the GTX 950.
Some people might point to something like DirectX 12 multiadapter as a possible workaround, which in explicit mode will use any two or more GPUs together as a single unit. That’s definitely something that can happen, which means that the lack of a connector won’t be a problem, but in that particular instance all communication is happening over the PCI-Express bus. The onus is also on the game developers to code support for it in their games or game engines. Those studios that already have the requisite expertise will do it and do it well (DICE and Crytek, for example), while other studios won’t touch it at all.
AMD, on the other hand, isn’t saddling their Polaris 10 GPUs with this limitation. Crossfire multi-GPU runs over the PCI-Express based XDMA now, so there’s also no physical limit to how many GPUs could be in one array (only drivers stop you from running eight-way Crossfire). It remains to be seen how much of a drawback this turns out to be. NVIDIA might surprise everyone with a PCI-Express version of SLI that doesn’t need a bridge, but I don’t think they will.
The NDA for sales of the GTX 1060 goes up on 19 July 2016, and reviews will go up a little bit before that (I’m guessing sometime this week, perhaps Thursday or Friday). It’ll be interesting to see how this launch proceeds. The GeForce GTX 1070 and GTX 1080 have been suffering from stock issues since launch and NVIDIA can’t keep up with demand, but with the RX 480 out in the market, NVIDIA will be pressured more to get this card into the market with good stock levels.