BENNETT RING inspects NVIDIA’s latest gadget.
Performance gamers, welcome at last to Maxwell. The successor to NVIDIA’s incredibly popular Kepler architecture, Maxwell first snuck its way into desktop gaming PCs by way of the GTX 750/Ti mainstream graphics card back in February of this year. It was a strange way to introduce the company’s latest
desktop design, as the company tends to roll out new architectures at the upper echelons first. As a result it didn’t make much of a splash, but the good news is that NVIDIA has now introduced the next version of the Maxwell architecture to the speedy end of town. Welcome to the GeForce GTX 980 and its cutdown sibling, the GTX 970.
THE GTX 980
The GeForce GTX 980 is currently the fastest single GPU product in NVIDIA’s lineup, and is built around the new GM204 GPU. It’s designed to replace the GTX 780 Ti, but don’t expect it to wipe the floor with it – as with past NVIDIA launches, it offers an incremental performance increase over the prior generation. This version of the chip comes with 2048 CUDA cores, 128 Texture Units and 64 ROPs. The GPU is clocked at a base speed of 1126MHz, but using NVIDIA’s Boost technology increases up to 1216MHz when the going gets tough. NVIDIA has limited the GTX 980 to 4GB of onboard GDDR5 memory, but it’s clocked at a blazing speed of 7GHz, running over a 256-bit memory bus. Surprisingly, the GTX 980 is built on the same 28nm manufacturing process as the last generation; it seems only Intel still has the technical knowhow to build chips on a 22nm process.
This makes the power savings offered by Maxwell even more impressive. Traditionally chip makers have had to use a smaller manufacturing process to cut energy requirements, but Maxwell uses clever design tricks instead. The result is a processor that NVIDIA claims delivers twice the performance per Watt of the last generation. The way it did this isn’t easily summarised, but it began by designing Maxwell first and foremost for mobile use, a complete shift in the way NVIDIA designed its chips. This makes perfect sense considering the massive growth rates amongst mobile gamers compared to desktop, and power savings are of crucial importance to this audience. Compared to the GTX 780, which ran with a TDP of 250W, the new GTX 980 drops this to just 165W. Now, that’s not quite half the Wattage, but the 980 churns through 5 TFLOPS compared to 4 TFLOPS on the 780, hence the claim about doubling of performance per Watt.
The GTX 980 requires two six-pin power plugs to supplement the power supply over the PCIe bus, and once again we see NVIDIA wheel out its sexy aluminium cooler design. Those of you with Perspex windows will find the reference cooler is easily the most attractive when compared to after-market coolers, if not quite as powerful. It’s nice to see NVIDIA has also changed up the outputs on the GTX 980, now packing three DisplayPort outputs, an HDMI port and a dual-link DVI port. Best of all is that the HDMI port is of the newest 2.0 standard, making this the first graphics card on the market to support 4K resolution at 60fps over HDMI. This will be a game-changer for those who want to hook their PC up to a 4K TV, as most TVs are now relying upon HDMI 2.0 to hit such 60Hz at 4K resolution.
Prices for the GeForce GTX 980 are currently starting around the $730 mark for the cheapest of brands, with premium products fetching a whopping $830. Once again it appears NVIDIA isn’t too worried about bang for buck, preferring to leverage top tier performance and features to make sales.
THE GTX 970
The GTX 980 might be a potent piece of kit, but if you’re like us you probably care more about the GTX 970. NVIDIA’s X70 products have always offered the best bang for buck at the performance level, and it appears that the GTX 970 won’t be any different. In the past, we’ve seen the second tier chip ship with some of the internals disabled, which slows performance slightly at stock speed. However, this gives the chip a lower operating temperature, which in turn makes it even more overclockable than its big brother. Avid tweakers have usually been able to overclock the X70 versions of NVIDIA products to within a percent or two of the X80 product, while saving a couple of hundred dollars in the process.
So, what has NVIDIA disabled to make the GTX 970 a more affordable product? The first major change is the number of CUDA cores, dropping from 2048 in the 980 down to 1664 in the GTX 970, almost a 20% decrease. The number of texture units has also been axed, down from 128 to 104, but the ROPs are identical, at 64. As expected, the GPU base speed has dropped slightly, down to 1050MHz, as has the Boost speed, at 1178MHz. Surprisingly though NVIDIA has kept the memory bandwidth intact, using the exact same 256-bit bus driving 4GB of onboard memory at 7GHz. This should see the 970 performing closer to the 980 at higher resolutions than prior generations, as memory bandwidth is often the limiting factor when it comes to resolution.
The biggest change, and the one we know you’ll love the most, is the price. With prices of the cheapest GTX 970s starting at just $449, it’s a vastly more affordable proposition. That’s a saving of almost 40%, which is even more remarkable if the 970 can overclock to within spitting distance of the 980. Given the low energy requirements of the 970, we’ve got a feeling it can, so let’s put both cards to the test to see just what they can do.
TO THE GRINDSTONE
We couldn’t wait to see what NVIDIA’s new products could do, though our expectations were tempered by the knowledge that over the last few years its new goodies tend to only extend the lead over the last generation by ten percent or so. We slapped together our usual test-bench, with an i7-4770K CPU at the heart, overclocked to a stable 4.4GHz to ensure the GPUs were the bottleneck. 8GB of DDR4 1800MHz memory provided short term memory, while a Corsair 256GB SSD stored our benchmarks. We Gigabyte’s GTX 980 and 970 for testing, but downclocked these factory overclocked products to reference speeds, to see just what NVIDIA’s original design is capable of.
Before we delve into the numbers, we have to point out just how quiet NVIDIA’s new products are. The 700 series was by no means a leaf-blower, but the 900 series is truly something else. It was next to impossible to hear the GPU fan over the sound of our CPU cooler. At load, the 980 piped in at just 45dB, while the 970 was an even quieter 43dB. Even more remarkable were the operating temperatures. After running all three benchmarks we left both cards running Grid Autosport in looping mode, in an open testbench. The 970 peaked at just 61C, making it the perfect candidate for life in a tightly enclosed case, such as a Micro ATX or Small Form Factor PC. The 980 didn’t fare quite so well, hitting 77C, but it’s still a long way off the 95C temps we were all making do with just two years ago.
So, onto the numbers. The 980 is now the undisputed king of single-GPU performance. In most tests it outpaces the prior speed king, the GTX 780 Ti, by around 10 to 15%. We know that this card had a small lead over AMD’s R9 290X, which propels the GTX 980 to the top of the speed heap. However, given that
it’s only 10% faster than the 780 Ti, neither of these products are the smartest upgrade path for those who already own the last generation of products. Gamers with cards approaching their third birthday should see a potential performance increase.
We were extremely impressed with the results of the GTX 970, which performed much closer to the GTX 980 than expected. Things got even more exciting once we put the cards under the overclocking vice though. The Gigabyte GTX 970 managed to hit a top stable GPU speed of 1420MHz, which brings it almost to the same performance levels of the GTX 980 – none too shabby for a 40% cost saving. Memory overclocking wasn’t as spectacular, topping out at 7.5GHz. The GTX 980 reached slightly faster speeds, with a stable GPU speed of 1450MHz, while the memory maxed out at the identical speed of 7.5GHz.
SMARTER, FASTER, COOLER
While both cards aren’t a huge leap over the prior generation, they’re de nitely worth considering for owners of older cards. The fact that NVIDIA managed to deliver even faster performance while using less energy is to be commended, especially considering it did so without a process shrink. Where top-tier cards in the past could double as a portable BBQ, it’s fantastic to see such high levels of performance in cool, quiet packages, opening up top-tier power to a range of smaller and more mobile uses.