Nvidia Responds To GTX 970 3.5GB Memory Issue
There have been numerous reports of the GTX 970 suffering from a memory allocation issue limiting it to 3.5GB instead of the full 4GB. We've done our own testing and analysis of this issue here at Wccftech.com in a previous piece. And in this article we aim to explain the issue in totality, why it occurs, what it means and how it affects users.
This whole debacle started when GTX 970 users began reporting an issue with memory allocation on their cards. The issue specifically was around the GTX 970 not being able to address all of the 4GB of on-board GDDR5 memory that Nvidia has equipped its cards with. GTX 970 users continuously reported maximum memory usage hitting a wall at 3.5GB. Shortly afterwards we went ahead and tested out the GTX 970 to try and see where this issue is stemming from. What we had discovered was quite interesting and goes hand in hand with Nvidia's latest response to this issue. But before we discuss our findings let's see what Nvidia has to say first.
Nvidia Responds to GTX 970 Memory Allocation Issue
In a recent statement to PCPer.com Nvidia admitted the existence of a memory allocation behavior that led so many GTX 970 users to worry. The company also offered a brief explanation of the phenomenon and why it occurs.
"The GeForce GTX 970 is equipped with 4GB of dedicated graphics memory. However the 970 has a different configuration of SMs than the 980, and fewer crossbar resources to the memory system. To optimally manage memory traffic in this configuration, we segment graphics memory into a 3.5GB section and a 0.5GB section. The GPU has higher priority access to the 3.5GB section. When a game needs less than 3.5GB of video memory per draw command then it will only access the first partition, and 3rd party applications that measure memory usage will report 3.5GB of memory in use on GTX 970, but may report more for GTX 980 if there is more memory used by other commands. When a game requires more than 3.5GB of memory then we use both segments.
We understand there have been some questions about how the GTX 970 will perform when it accesses the 0.5GB memory segment. The best way to test that is to look at game performance. Compare a GTX 980 to a 970 on a game that uses less than 3.5GB. Then turn up the settings so the game needs more than 3.5GB and compare 980 and 970 performance again.
Here’s an example of some performance data:
|GTX 980||GTX 970|
|Shadow of Mordor|
|<3.5GB setting = 2688x1512 Very High||72 FPS||60 FPS|
|>3.5GB setting = 3456x1944||55 FPS (-24%)||45 FPS (-25%)|
|<3.5GB setting = 3840x2160 2xMSAA||36 FPS||30 FPS|
|>3.5GB setting = 3840x2160 135% res||19 FPS (-47%)||15 FPS (-50%)|
|Call of Duty: Advanced Warfare|
|<3.5GB setting = 3840x2160 FSMAA T2x, Supersampling off||82 FPS||71 FPS|
|>3.5GB setting = 3840x2160 FSMAA T2x, Supersampling on||48 FPS (-41%)||40 FPS (-44%)|
On GTX 980, Shadows of Mordor drops about 24% on GTX 980 and 25% on GTX 970, a 1% difference. On Battlefield 4, the drop is 47% on GTX 980 and 50% on GTX 970, a 3% difference. On CoD: AW, the drop is 41% on GTX 980 and 44% on GTX 970, a 3% difference. As you can see, there is very little change in the performance of the GTX 970 relative to GTX 980 on these games when it is using the 0.5GB segment."
Nvidia's statement explains our own findings earlier. In our analysis we discovered that the GTX 970 can actually address and use all available 4GB of VRAM and not just 3.5GB. However the performance penalty for using that last 0.5GB of VRAM is still interesting. As Nvidia explains it and as we had suspected earlier, the issue is essentially due to the fact that the GTX 970 is based on a cut-down GM204-200 GPU die instead of the full GM204 chip like its bigger brother the GTX 980.
The GM204-200 GPU has three SMM units disabled, for a total of 13/16 SMMs. This in turn disables some components of the memory sub-system. This leaves the chip with less resources to manage the same amount of memory as the GTX 980. Which is why Nvidia resorted to the dual segment (3.5GB-0.5GB) memory arrangement. This in turn insures that the larger 3.5GB segment doesn't suffer any performance penalty.
Nvidia doesn't provide data to compare how a 970 would perform had it not exhibited this memory behavior. So I went ahead and calculated how the performance of the GTX 970 would've been affected had this memory allocation phenomenon was absent. I've done this by applying the performance scaling ratio of the GTX 980 ,which doesn't exhibit this behavior, to the GTX 970.
|Wccftech||GTX 970 Performance in %||GTX 970 Performance Without Memory Allocation Issue in %|
|Shadow of Mordor||100%||102%|
|Call of Duty Advanced Warfare||100%||104%|
Simply put the GTX 970 is only able to fully utilize 3.5GB out of the 4GB of available VRAM optimally. While the card can still access all 4GB of VRAM, the last 0.5GB of memory is not accessed or managed as efficiently as the rest of the available memory. Which leads to the reported performance degradation when an application needs to access more than 3.5GB. In the examples Nvidia provided to the press the performance penalty is in the mid single digit percentages. However we don't fully know yet if in other usage scenarios the memory issue will prove to be more detrimental. As there's evidence that it has a more perceptual impact on frame time consistency rather than FPS which can be perceived as the stuttering or choppiness reported by Nvidia users on the GeForce forums.
Most users should not face this issue except on higher resolutions such as 4K or in the case of multi-monitor gaming. In which case we would actually recommend buying a graphics card with more memory anyway. AMD already has 8GB R9 290X cards in the market which were introduced three months ago. Earlier rumors indicated that Nvidia was also launching 8GB variants of the GM204 based 900 series cards in November of last year but that unfortunately did not happen. Samsung SDRAM schedules indicate that Nvidia could launch 8GB variants of their 900 series cards this quarter.
Although with rumors looming of a GM200 launch taking place in two months time and AMD's HBM equipped Fiji XT coming soon after. I'd suspect that users would want to see how well products based on these new GPUs perform in the market and then make a decision.