There is no 970 'memory bug'

[Guest]

The card does have 4gb vram, and it does have a real 256 bit memory bus, its just that with the cut down sm's, the 970 cannot access all 4 gb of ram as efficiently as the full 980 chip, and this isn't really suprising, since the gtx 970 is a cut down version of the gtx 980...

 

 

 

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.

 

Source: http://www.pcper.com/news/Graphics-Cards/NVIDIA-Responds-GTX-970-35GB-Memory-Issue

 

 

 

 

tl;dr/captain obvious

cut down version of gtx 980 doesn't perform as well as gtx 980

[79wjd]

repost :unsure:

http://linustechtips.com/main/topic/297381-nvidia-responds-to-gtx-970-35gb-memory-issue/

[Imakuni]

Sounds... as a very well made excuse. If that WAS the reason, they'd have replied that right away.

[Guest]

Sounds... as a very well made excuse. If that WAS the reason, they'd have replied that right away.

It IS the reason

[DeViLzzz]

Sorry but I am not letting Nvidia off the hook for this one.  The GTX 970 should be using all of the 4 GB of memory when you need it and when people with some cards can't do it they deserve a refund.

[TheSLSAMG]

Sorry but I am not letting Nvidia off the hook for this one.  The GTX 970 should be using all of the 4 GB of memory when you need it and when people with some cards can't do it they deserve a refund.

Then ask both AMD and NVIDIA for refunds for 256-bit 4GB cards in the past not using all of the VRAM due to bus limitations?

 

I don't disagree with you, but people didn't flip their shit so much over that back then like they are over this now.

[Guest]

Sorry but I am not letting Nvidia off the hook for this one.  The GTX 970 should be using all of the 4 GB of memory when you need it and when people with some cards can't do it they deserve a refund.

Nothing's changed since the card's release. the card performs just as well as indicated in reviews and benchmarks, and the card does legitimately have 4gb of vram, and the card can utilize all 4gb of vram. there is nothing to refund. This would be far from the only video card with some kind of performance bottleneck somehwere.

[maizenblue]

This will likely be Nvidias line, or something similar. The fanboys are already making this excuse without Nvidia prompting. In the end Nvidia will either ignore the issue and keep saying we are looking into it if prompted, or come up with the line that TECHNICALLY it does have 4 GB of memory. Unless of course this can be fixed via software, but if its a hardware problem expect this to be how they handle it.

 

Edit: Having read the link, I guess that IS Nvidias line. In other words they are saying there is no fix, it is what it is, and we think the card is fine this way.

 

So basically if you want to access a full 4 gb of VRAM, get a 980 or an r9 290/x.

[Imakuni]

It IS the reason

Proof? "Because Nvidia said it?"

 

Conspiracy theories.....

 

Sorry but I am not letting Nvidia off the hook for this one.  The GTX 970 should be using all of the 4 GB of memory when you need it and when people with some cards can't do it they deserve a refund.

But.... it IS using all the memory?

[EarthboundHero]

Nothing's changed since the card's release. the card performs just as well as indicated in reviews and benchmarks, and the card does legitimately have 4gb of vram, and the card can utilize all 4gb of vram. there is nothing to refund. This would be far from the only video card with some kind of performance bottleneck somehwere.

This doesn't hurt performance, it adds visual stuttering among other things that can't be measured by an FPS counter.

[TheSLSAMG]

Nothing's changed since the card's release. the card performs just as well as indicated in reviews and benchmarks, and the card does legitimately have 4gb of vram, and the card can utilize all 4gb of vram. there is nothing to refund. This would be far from the only video card with some kind of performance bottleneck somehwere.

We're talking huge performance tanking when going over 3.5GB though, which is a significant problem which needs a fix. It has 4GB of VRAM and can't use it without tanking it's performance, it's an issue.

[kpreg]

The card does have 4gb vram, and it does have a real 256 bit memory bus, its just that with the cut down sm's, the 970 cannot access all 4 gb of ram as efficiently as the full 980 chip, and this isn't really suprising, since the gtx 970 is a cut down version of the gtx 980...

 

 

 

 

Source: http://www.pcper.com/news/Graphics-Cards/NVIDIA-Responds-GTX-970-35GB-Memory-Issue

 

 

 

 

tl;dr/captain obvious

cut down version of gtx 980 doesn't perform as well as gtx 980

Why was I sold a card that said 4GB? Shouldn't it have said "3.5GB With 512MB Slow Access Memory?" 256bit 4GB is in the product's title, and that implies that the card has a 256bit wide bus to access the mem. still feel ripped off and nothing will change that unless NVIDIA compensates for this.

[Guest]

Why was I sold a card that said 4GB? Shouldn't it have said "3.5GB With 512MB Slow Access Memory?" 256bit 4GB is in the product's title, and that implies that the card has a 256bit wide bus to access the mem. still feel ripped off and nothing will change that unless NVIDIA compensates for this.

the 970 does not have an asymetrical design like the gtx 660, the card does have a 256-bit bus. The card's memory access is bottlenecked by sm resources, not the bus.

[maizenblue]

Notice the r9 290 doesn't have this same problem versus the 290x.

 

Looks like Nvidia pulled a fast one on people. They already sold a shit ton of 970s so what do they care if a few more tech geeks pass on a new 970 because of this issue. As for disgruntled 970 owners, they know most of them are dedicated to Nvidia, and by the time they are ready to buy a new card a couple years from now they will be too excited about Nvidias next new thing to care or jump ship to AMD.

[MattDaemontools]

This is what happens when you let marketing morons go wild. The 980 isn't a 165W card and the 970 can't use all of its VRAM properly. Also what the frack is wrong with Nvidia saying the Tegra X1 is a 1TFlop chip when they are actually counting the 16bit half precision performance which was abandoned over a decade ago. When in fact the single percision FP32 performance is 500GFlops at PEAK, so not even when it goes into actual products with real-world thermal and power limits.

[XiaoPineApple]

like the wanshow said, hope is a firmware issue ans not hardware. If mass recall....

the money ive spend on the card + waterblock...

Tear...

[xAcid9]

It's not a memory bug but instead a misinformation/false advertising. 

[kpreg]

the 970 does not have an asymetrical design like the gtx 660, the card does have a 256-bit bus. The card's memory access is bottlenecked by sm resources, not the bus.

My bad. Can you elaborate on what a SM is, I feel kind of dumb.

[Guest]

My bad. Can you elaborate on what a SM is, I feel kind of dumb.

SM (also referred to as 'SMM' in maxwell) = Streaming multiprocessor

 

In maxwell Each SMM has 128 CUDA cores (aka stream processors).

 

With the GTX 970 they cut down the card by severing a trio of SMM units, while leaving the rest of the card pretty much the same as the gtx 980. This is the reason for the less efficient memory access.

 

This article does a good job explaining how the reduced SMM count in the 970 effects performance compared to the 980:

 

http://techreport.com/blog/27143/here-another-reason-the-geforce-gtx-970-is-slower-than-the-gtx-980

 

 

The pixel fillrate can be linked to the number of ROPs for some GPUs, but it’s been limited elsewhere for years for many Nvidia GPUs. Basically there are 3 levels that might have a say at what the peak fillrate is :

• The number of rasterizers
• The number of SMs
• The number of ROPs

On both Kepler and Maxwell each SM appears to use a 128-bit datapath to transfer pixels color data to the ROPs. Those appears to be converted from FP32 to the actual pixel format before being transferred to the ROPs. With classic INT8 rendering (32-bit per pixel) it means each SM has a throughput of 4 pixels/clock. With HDR FP16 (64-bit per pixel), each SM has a throughput of 2 pixels/clock.

On Kepler each rasterizer can output up to 8 pixels/clock. With Maxwell, the rate goes up to 16 pixels/clock (at least with the currently released Maxwell GPUs).

So the actual pixels/cycle peak rate when you look at all the limits (rasterizers/SMs/ROPs) would be :

GTX 750 : 16/16/16

GTX 750 Ti  : 16/20/16

GTX 760 : 32/24/32 or 24/24/32 (as there are 2 die configuration options)

GTX 770 : 32/32/32

GTX 780 : 40/48/48 or 32/48/48 (as there are 2 die configuration options)

GTX 780 Ti : 40/60/48

GTX 970 : 64/52/64

GTX 980 : 64/64/64

Extra ROPs are still useful to get better efficiency with MSAA and so. But they don’t participate in the peak pixel fillrate.

That’s in part what explains the significant fillrate delta between the GTX 980 and the GTX 970 (as you measured it in 3DMark Vantage). There is another reason which seem to be that unevenly configured GPCs are less efficient with huge triangles splitting (as it’s usually the case with fillrate tests).

So the GTX 970's peak potential pixel fill rate isn't as high as the GTX 980's, in spite of the fact that they share the same ROP count, because the key limitation resides elsewhere. When Nvidia hobbles the GTX 970 by disabling SMs, the effective pixel fill rate suffers.

[kpreg]

SM (also referred to as 'SMM' in maxwell) = Streaming multiprocessor

 

In maxwell Each SMM has 128 CUDA cores (aka stream processors).

 

With the GTX 970 they cut down the card by severing a trio of SMM units, while leaving the rest of the card pretty much the same as the gtx 980. This is the reason for the less efficient memory access.

 

This article does a good job explaining how the reduced SMM count in the 970 effects performance compared to the 980:

 

http://techreport.com/blog/27143/here-another-reason-the-geforce-gtx-970-is-slower-than-the-gtx-980

So why isn't there a deticated memory controller? I feel like I'm missing something. 

[Guest]

So why isn't there a deticated memory controller? I feel like I'm missing something. 

There are four 64 bit memory controllers. What happens is each SM uses 'crossbars' to access memory channels. The issue here is that because of the disabled SMs there's an increased load on fewer crossbars and efficiency is lost.

[iDeJayDzn]

Am i the only one right in thinking that Nvidia don't want to fix it because the 970 may gain a performance increase to the 980 therefore counting the 980 as an unusal buy and making the 970 power to price ratio higher than any other card?

 

Literally all it sounds like.

[kpreg]

There are four 64 bit memory controllers. What happens is each SM uses 'crossbars' to access memory channels. The issue here is that because of the disabled SMs there's an increased load on fewer crossbars and efficiency is lost.

I see. Thanks for explaining!

But this just sounds like a simple oversight to me, an unseen effect. 

cmon nvidia, get it toghether

[-rascal-]

There are four 64 bit memory controllers. What happens is each SM uses 'crossbars' to access memory channels. The issue here is that because of the disabled SMs there's an increased load on fewer crossbars and efficiency is lost.

If THIS causes a significant performance drop, then it is a hardware flaw / poor engineering / implementation.

I can ask my Engineering professors, and Engineers out in the field, and they would most likely say the same.

[Guest]

If THIS causes a significant performance drop, then it is a hardware flaw / poor engineering / implementation.

I can ask my Engineering professors, and Engineers out in the field, and they would most likely say the same.

It is a flaw in the design, the way they cut it down certainly isn't perfect, but overall the 970 is still a very efficient card, and this whole 'memorygate' thing is pretty overblown to say the least.