This site may earn affiliate commissions from the links on this page. Terms of employ.

GTX970

When Nvidia launched the GTX 970 and 980 back in the autumn of 2014, Maxwell quickly rose to prominence thanks to strong performance and excellent power efficiency. Early reviews showed good results on both cards, only deeper dives into the GTX 970 post-launch turned upward troubling behavior: The card strongly preferred to limit its available memory to three.5GB, not the 4GB information technology actually carried. A grade-action lawsuit was launched a month after the problem came to light, and Nvidia has apparently resolved that case every bit of today.

Nvidia volition pay each GTX 970 owner that applies an average of $30 (at that place's no data yet on how one applies for the refund) in improver to paying $one.three million in chaser fees, according to Overclockers3D. While $thirty isn't huge, it presumably accounts for the estimated value of the finer-missing 512MB of RAM.

For those of you lot who don't recall the details of the effect, the problem was this: The GTX 970's internal retentivity configuration disabled one of its crossbar switches, as shown in the diagram beneath:

diagram2

GTX 970'due south memory subsystem

The "SM" blocks in this diagram are the actual processor cores, while the L2 blocks refer to the L2 cache. "MC" blocks are the memory controllers. The bit is designed so that whatsoever SM block could talk to whatsoever of the retentiveness controllers, just three of the SM blocks and one of the L2 cache blocks is grayed out in this diagram. The SM blocks are disabled to striking the GTX 970's core count target, but the terminal block of L2 enshroud does double duty, serving 2 SM blocks rather than just one. Disabling L2 cache blocks in this fashion also reduced the GTX 970's ROP count to 56 downwardly from 64 (every bit Tech Study noted at the time, actual throughput was actually a maximum of 52 pixels per clock due to associated limitations in the crossbar).

Nvidia PR claimed that the GTX 970's original specs were miscommunicated to reviewers, and I believe them. There's a difference between scarlet picking results and arguments to favor whatsoever visitor's given position on a topic, and flatly misrepresenting the capabilities of 1's hardware in an effort to drive sales under imitation pretenses. The erstwhile is expected, the latter is criminal. The cyberspace effect of this problem was to leave the GTX 970 with a 512MB retentiveness buffer that was technically available to games if absolutely required, but could only be accessed at a fraction of the speed of main retention, as shown in the results beneath:

Nvidia memory test

Practically speaking, the impact on the vast majority of users was minimal. Benchmarks and tests showed that in that location were games that could trip the GTX 970, but this ofttimes only occurred at the limits of playable frame rates in whatever case. Overall, the GTX 970 sold well and became one of the almost popular GPUs of the previous generation. Given that recent cost cuts have left information technology priced as depression as $260 (equally of this writing), information technology'southward not a bad deal even now — though generally speaking I'd however recommend either the RX 480 or the GTX 1060, depending on your preferences. While the GTX 970 should be fine for 1080p and below in the indefinite future, there'south always the take chances that this RAM problem will bite every bit VRAM requirements continue to scale up.