New rumors today imply that Nvidia is preparing to release a mainstream version of its Maxwell GPU in late January. Nvidia debuted Maxwell in three desktop GPUs in 2014 — the lower-end GTX 750 Ti in February, and the high-end GTX 980 and GTX 970 in September. The GTX 970 was the crown jewel of that arrangement, as far as the price-performance ratio is concerned. Now, the company is reportedly prepping a new card for the mainstream market — the GTX 960.
Specs on the card, at present, are a bit of a muddle. Today’s report indicates that the chip will use a 128-bit memory bus, paired with a new GM206 GPU core and 2GB of RAM. Previous reports, however, had pointed to a larger chip, a 256-bit memory bus, and a 4GB memory buffer using a cut-down version of the same GM204 that powers the GTX 980 and 970.
I suspect there’s truth to both rumors. Right now, the GTX 970 sells for $ 329, with 1664 CUDA cores, an 8GB frame buffer, and a 256-bit memory bus. Nvidia could trim the die to create a GTX 960 (or 960 Ti) by disabling more cores, but the GTX 970 is already a fair bit smaller than the 980, with 13 SMMs enabled (1664 cores) as opposed to 16 (2048 cores) on the full chip. Given that Nvidia typically targets both the $ 180-$ 200 and the $ 250-$ 280 market with separate SKUs, it makes sense to assume that the company will introduce products to address both these segments. The question is, how will it hit them?
When we reviewed the GTX 750 Ti this past spring, we noted that while the card’s power efficiency and performance-per-watt was excellent, AMD ultimately had a stronger offering from its own hardware. The GTX 750 Ti’s 128-bit memory bus appeared to handicap it in 1080p gaming — there were several cases where the older GTX 650 Ti Boost was faster than the card that ostensibly replaced it.
This implies that the GTX 960, when it does launch, will either offer a larger L2 cache than the GTX 750 Ti, draw on faster memory (the GTX 750 Ti used 1350MHz GDDR5, but GDDR5 clocked at 1750MHz is available on the market), or will simply use a wider memory bus. A 2GB frame buffer, possibly with a 4GB OEM option, probably makes the most sense for the current space — though given the RAM requirements on next-gen games, Nvidia could plausibly push the envelope on this front a bit.
How this plays out will depend on how much trouble Nvidia wants to make for AMD, and what kind of yields it’s getting on 28nm Maxwell hardware. After the GTX 980 launched, AMD responded with aggressive price cuts on the R9 290 and R9 290X. Combined with the R9 285 that it launched in August, Team Red currently has a solidly competitive lineup at the $ 329 price point and below. Maxwell’s major claim to fame, however, is its performance-per-watt — and that’s a gain that AMD’s current crop of GCN cards simply can’t match. If Nvidia gets aggressive with pricing this new card, it could put significant pressure on AMD’s competitive structure in the important mainstream market.
Now read: Why are high-end graphics cards so big?