NVIDIA Working on New Driver For GeForce GTX 970 To Tune Memory Allocation Problems and Improve Performance

Later on the recent update of specifications of their GeForce GTX 970 graphics cards, it seems like NVIDIA is in pressure level from their consumers who seem betrayed that they didn't get what they payed for. While NVIDIA gave a reasonable brief explanation a few days dorsum on why the GeForce GTX 970 has an issue allocating its entire four GB VRAM pool to games, they did so very late and has at present been revealed that the GPU lacks some major components such as ROPs and Enshroud which were falsely advertised v months agone during launch.

NVIDIA To Fine Tune GeForce GTX 970 Retention Allocation Bug To Improve Performance

While NVIDIA is nether pressure and the GTX 970 owners are flamed up, NVIDIA's employee Peterson has accustomed that his company did mess upward the stats of the GeForce GTX 970 and that they volition soon release a driver which will tune how the memory is allocated by the GeForce GTX 970 in gaming titles that will help  ameliorate performance further up. Information technology's non known how the new GeForce driver will work simply some have suggested that it may be similar to the GeForce 337.fifty BETA commuter by focusing towards all its optimizations towards the GeForce GTX 970 in memory bound conditions such every bit the three.5 GB VRAM edge after which it is reported that games start to lag or stutter. Following is the message from Peterson on GeForce forums:

Annotate #1 -Hey,

Outset, I want you to know that I'thou not but a mod, I work for NVIDIA in Santa Clara.

I totally go why so many people are upset. We messed upward some of the stats on the reviewer kit and we didn't properly explain the retentivity compages. I realize a lot of you guys rely on product reviews to make purchase decisions and we permit you downwards.

It sucks because we're really proud of this matter. The GTX970 is an astonishing card and I genuinely believe it'due south the best card for the money that yous can buy. Nosotros're working on a driver update that volition tune what'due south allocated where in memory to further better performance.

Having said that, I understand that this whole experience might have turned you off to the card. If y'all don't want the menu anymore you should return it and get a refund or substitution. If y'all have any bug getting that washed, let me know and I'll do my all-time to help.

--Peter

Annotate #ii -Really I'm not sure as that'due south not a simple issue with just one crusade. Card memory is not just used for the frame buffer, enough of driver stuff gets loaded into it as well. We're looking at sticking as much of that stuff as possible into the 0.5GB infinite to exit the remainder bachelor.

Annotate #three -The GTX970 really does accept 4GB of memory and tin can admission all of information technology. And we're looking at ways to tweak the commuter to better empathize where to put stuff to make it fifty-fifty faster. But I totally get that information technology might non be the right product for your specific state of affairs. If you actually desire to render it and are getting denied, let me know and I'll practise my all-time to help.

The issue started in late November when users started reporting in through several forums that their GTX 970'south are declining to get past 3.5 GB VRAM in certain games. Afterwards a few days, nosotros were quick to come up up with our own analysis showing that the card can utilize its four GB VRAM which is the total retentiveness available on the PCB but only under highly stressful atmospheric condition. Information technology was not until users who were breaking by the iii.5 GB VRAM started reporting some other issue that their cards tin can become past the 3.five GB buffer just either lag, stutter or show artifacts which made several sites initiate a testing spree with their GeForce GTX 970 cards and asking NVIDIA for a answer on this issue. NVIDIA's outset reply was that the carte du jour had a batten which had two pools of memory, one was the 3.five GB VRAM which was faster and used past gaming applications while other was a 0.5 GB pool which was slow yet notwithstanding faster then system retentiveness. This optimally handled the Maxwell core arrangement that was available on the GeForce GTX 970 GM204 fleck since information technology was disabled SKU.

Just two days afterwards the initial revealing by NVIDIA and five months afterwards the launch of GeForce GTX 970, Jonah Alben, SVP of GPU Applied science, accustomed that the specs of the carte du jour weren't what was showed initially and the specs of the card were more cut down then previously expected. Shown in a block diagram of the GTX 970 GM 204 fleck, he showed that the scrap had just 56 ROPs every bit opposed to 64 and 1792 KB L2 cache every bit opposed to 2048 KB L2 enshroud equally previously advertised. He stated that it was a misunderstanding of specs handling by the NVIDIA marketing and technical team.

A quick note about the GTX 980 here: it uses a 1KB retentivity access stride to walk beyond the memory bus from left to right, able to striking all 4GB in this chapters. Merely the GTX 970 and its altered design has to do things differently. If y'all walked beyond the memory interface in the exact same style, over the same 4GB chapters, the 7th crossbar port would tend to always get twice as many requests as the other port (because information technology has 2 memories fastened).  In the short term that could be ok due to queuing in the memory path.  Simply in the long term if the 7th port is fully busy, and is getting twice as many requests as the other port, then the other 6 must be but half busy, to match with the ii:1 ratio.  So the overall bandwidth would exist roughly half of peak. This would crusade dramatic underutilization and would foreclose optimal performance and efficiency for the GPU.

Let's be edgeless here: access to the 0.5GB of memory, on its ain and in a vacuum, would occur at 1/7th of the speed of the 3.5GB pool of memory. If you expect at the Nai benchmarks floating around, this is what you lot are seeing.To avert this, NVIDIA divided the retentivity into two pools, a three.5GB pool which maps to 7 of the DRAMs and a 0.5GB pool which maps to the eighth DRAM.  The larger, primary pool is given priority and is then accessed in the expected 1-two-iii-4-five-half-dozen-7-one-2-3-4-5-6-7 blueprint, with equal asking rates on each crossbar port, then bandwidth is balanced and can be maximized. And since the vast majority of gaming situations occur well under the three.5GB retentivity size this determination makes perfect sense. Information technology is those instances where memory above iii.5GB needs to be accessed where things go more interesting.

*To those wondering how peak bandwidth would remain at 224 GB/south despite the division of memory controllers on the GTX 970, Alben stated that it can achieve that speed only when retentivity is being accessed in both pools. via PCPER

The reason for this cut down was that the terminal two 0.v DRAM had to be continued to 2 32-flake memory controllers which notwithstanding were situated across just one L2 enshroud module. This resulted in vigorous sharing and handling for just one L2 cache since the other had to be disabled. To employ the terminal memory cake finer, NVIDIA had to dissever a single slice of DRAM hence converting the card into a iii.5 GB model which only used the last section of the VRAM when information technology was really needed. NVIDIA is now going to fine melody the operation of this specific block and how it manages resource sharing across a set of applications.

It'due south not known when the new driver launches but looking at the amount of heat they are currently getting, the new drivers will exist launched next month.

Petition Launched Against NVIDIA To Refund GTX 970 Cards: A petition has since been launched asking NVIDIA to refund the users who were falsely advertised with wrong specifications at launch. It has gained almost 3000 supporters and the numbers are closing each passing moment. You tin sign the petition by heading to the link below:

https://www.alter.org/p/nvidia-refund-for-gtx-970

[socialpoll id="2249878"]

NVIDIA GeForce GTX 970 and GTX 980 Specifications:

	GeForce GTX 970 (Initial)	GeForce GTX 970 (Corrected)	GeForce GTX 980
Codename	GM204	GM204	GM204
Process	28nm	28nm	28nm
GPU Core	Maxwell	Maxwell	Maxwell
SM Units	thirteen x 128	thirteen x 128	16 x 128
CUDA Cores	1664	1664	2048
ROPS	64	56	64
TMUs	104	104	128
L2 Enshroud	2048KB	1792KB	2048KB
Cadre Clock	1051 MHz	1051 MHz	1126 MHz
Boost Clock	1178 MHz	1178 MHz	1216 MHz
Retentivity	4 GB GDDR5	4 GB GDDR5	four GB GDDR5
Memory Coach	256-scrap	256-chip	256-chip
Retentivity Clock	vii.0 GHz	7.0 GHz	vii.0 GHz
Retentiveness Bandwidth	224.5 GB/south	224.5 GB/s	224.five GB/due south
Texture Fill Charge per unit GT/s	145.0	145.0	172.eight
TDP	148W	148W	165W
Power Connectors	half dozen+half-dozen Pin	six+6 Pin	vi+vi Pin
DirectX 12 Support	Yes	Yes	Yes
Launch	18th September 2022	18th September 2022	18th September 2022
Price	$329 Reference $329+ Custom	$329 Reference $329+ Custom	$549 Reference $549+ Custom

Source: https://wccftech.com/nvidia-working-driver-geforce-gtx-970-tune-memory-allocation-problems-improve-performance/

Posted by: jonesdervive.blogspot.com

Share this post