NVIDIA’s GTX 960 was launched amidst a flurry of reviews but its actual performance has been a bit hard to accurately nail down. On one hand its 1080P benchmarks were admirable but those metrics were obtained with pre-overclocked cards like ASUS’ STRIX which tended to inflate benchmark numbers. We wanted to get our hands on a reference-clocked card to get some idea of baseline tendencies and that’s exactly what PNY stepped up to deliver.
While the vast majority of GTX 960 cards use higher clock frequencies and heavily upgraded heatsink assemblies, they can also cost a fair bit more than NVIDIA’s stated starting price of $199. There have been a few examples (namely EVGA’s GTX 960 SC) where rebates have brought even these cards’ respective costs down to those of reference-based versions. However, the promotions are due to run out which leaves just a few SKUs left at NVIDIA’s advertised price. So what does $199 actually get you? That’s what we’re aiming to find out.
On paper at least, the reference GTX 960 really doesn’t have all that much going for it. It uses a fully enabled GM206 core but that only grants 1024 CUDA cores, 32 ROPs and 64 Texture units. While there are technical features within the Maxwell architecture which insure substantial performance enhancements over the Kepler-based GTX 760 on a per SM-basis, they can only go so far.
The most notable shortfall is in the memory bandwidth where the GTX 960 makes due with just a 128-bit bus spread over a pair of 64-bit controllers. Once again the Maxwell architecture does feature some memory efficiencies which increase theoretical bandwidth but even then the peak output of its GDDR5 pipelines remains behind the previous generation. If you want to know more about the architecture and how it really compares, check out our launch-day GTX 960 review.
Despite what sounds to be like a drastic specification shortfall in relation to the card it’s supposed to replace, the GTX 960 is supposed to surpass its processor in most respects. By how much? We’re going to find out in this reference versus reference showdown.
PNY’s reference GTX 960 is slightly different than their competitors’ solutions since instead of using an ITX-centric short design they have basically reinvigorated the GTX 760’s blower-style design. This is a great layout for exhausting hot air outside the case but there won’t be any ultra low temperature results. It also extends the card’s overall length to 9 ½” which certainly isn’t SFF friendly but compatibility issues should be minimal at best.
This card may not be all that much to look at but the price is certainly right provided you can find it for PNY’s SRP of $199. Another addition is PNY’s lifetime warranty (upon registration of course) which is now a unique feature among NVIDIA’s current crop of board partners.
NVIDIA mentioned to us that most board partners’ GTX 960 overclocking will likely be limited by their choice of input power connectors. In PNY’s case, they have decided to use a very basic single 6-pin which may reduce the amount of additional Power Limit headroom, thus things back on this front.
The selection of output connectors is quite wide and covers all the usual bases. There’s a trio of DisplayPorts and single HDMI / DVI connectors which provides native triple-screen support.
While the vast majority of GTX 960 cards use higher clock frequencies and heavily upgraded heatsink assemblies, they can also cost a fair bit more than NVIDIA’s stated starting price of $199. There have been a few examples (namely EVGA’s GTX 960 SC) where rebates have brought even these cards’ respective costs down to those of reference-based versions. However, the promotions are due to run out which leaves just a few SKUs left at NVIDIA’s advertised price. So what does $199 actually get you? That’s what we’re aiming to find out.
On paper at least, the reference GTX 960 really doesn’t have all that much going for it. It uses a fully enabled GM206 core but that only grants 1024 CUDA cores, 32 ROPs and 64 Texture units. While there are technical features within the Maxwell architecture which insure substantial performance enhancements over the Kepler-based GTX 760 on a per SM-basis, they can only go so far.
The most notable shortfall is in the memory bandwidth where the GTX 960 makes due with just a 128-bit bus spread over a pair of 64-bit controllers. Once again the Maxwell architecture does feature some memory efficiencies which increase theoretical bandwidth but even then the peak output of its GDDR5 pipelines remains behind the previous generation. If you want to know more about the architecture and how it really compares, check out our launch-day GTX 960 review.
Despite what sounds to be like a drastic specification shortfall in relation to the card it’s supposed to replace, the GTX 960 is supposed to surpass its processor in most respects. By how much? We’re going to find out in this reference versus reference showdown.
PNY’s reference GTX 960 is slightly different than their competitors’ solutions since instead of using an ITX-centric short design they have basically reinvigorated the GTX 760’s blower-style design. This is a great layout for exhausting hot air outside the case but there won’t be any ultra low temperature results. It also extends the card’s overall length to 9 ½” which certainly isn’t SFF friendly but compatibility issues should be minimal at best.
This card may not be all that much to look at but the price is certainly right provided you can find it for PNY’s SRP of $199. Another addition is PNY’s lifetime warranty (upon registration of course) which is now a unique feature among NVIDIA’s current crop of board partners.
NVIDIA mentioned to us that most board partners’ GTX 960 overclocking will likely be limited by their choice of input power connectors. In PNY’s case, they have decided to use a very basic single 6-pin which may reduce the amount of additional Power Limit headroom, thus things back on this front.
The selection of output connectors is quite wide and covers all the usual bases. There’s a trio of DisplayPorts and single HDMI / DVI connectors which provides native triple-screen support.
