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.
So the VRAM capacity of the 970s is segmented into 3.5 and 0.5 GBs where as the 980 has a sole segment resulting in 4 GBs? Sounds like a cheesy design strategy.
Why would anyone care about this? The 970 brings awesome performance/price ratio either way. How it does that internally, isnt really that interesting. Only for syntethic testers maybe.
It's not false advertising. The fact remains, the 970 has 4GB of VRAM on board. That's it, don't argue about this further.
The issue is how the 4GB of VRAM is being used, NOT that the 970 having a missing 512mb of VRAM.
Quote:
Originally Posted by LocutusH
Why would anyone care about this? The 970 brings awesome performance/price ratio either way. How it does that internally, isnt really that interesting. Only for syntethic testers maybe.
People are angry that the 4GB of VRAM doesn't work as it should and that Nvidia kept quiet about it. A few days ago a rep was saying they are "looking" into the issue, but really it's more like how they should word out their PR statement instead of actually doing anything specifically to the 970.
Why would anyone care about this? The 970 brings awesome performance/price ratio either way. How it does that internally, isnt really that interesting. Only for syntethic testers maybe.
Why would anyone care about this? The 970 brings awesome performance/price ratio either way. How it does that internally, isnt really that interesting. Only for syntethic testers maybe.
Except, the driver picks up on benchmarks and you never notice the difference, except in games.Post #90
Many reviewers resort to making conjunctures out of a repertoire of games quite distant from reaching 4GB ever, too.
While i see your point, there is no reason to call it 3.5gb. It has 4gb, and all 4gb is accessible. There was no word about a slower part tough.
Either way, anyone who bought a 970, did it because of its performance in nowadays games. Is this affected by this new found issue? NO. It still brings the fps numbers everyone wanted...
It's not false advertisement but it is deceptive advertisement, which may hurt Nvidia sales in the long run. If something is marketed as 4GB VRAM you reasonably expect it to use up to 4GB of VRAM effectively, not 3.5GB effective and the last .5 poorly. I'm okay with the latter as long as it's marketed to perform that way, but Nvidia has kept that quiet for obvious reasons.
224GB/s for the full 4GB while that is not the case so that would qualify as false advertising because the actually bandwidth is 1/10th of that and that means it is significantly off.
Quote:
Originally Posted by LocutusH
Why would anyone care about this? The 970 brings awesome performance/price ratio either way. How it does that internally, isnt really that interesting. Only for syntethic testers maybe.
Bandwidth is cut to a point where hiccups occur.
Quote:
Originally Posted by benbenkr
People are angry that the 4GB of VRAM doesn't work as it should and that Nvidia kept quiet about it. A few days ago a rep was saying they are "looking" into the issue, but really it's more like how they should word out their PR statement instead of actually doing anything specifically to the 970.
While i see your point, there is no reason to call it 3.5gb. It has 4gb, and all 4gb is accessible. There was no word about a slower part tough.
Either way, anyone who bought a 970, did it because of its performance in nowadays games. Is this affected by this new found issue? NO. It still brings the fps numbers everyone wanted...
No. People wanted 4GB of vRAM. A lot of GTX780 owners got it for the extra 1 GB and not 512MB. Its false advertisement if it's true. It's like putting 1.5GB of vRAM in GTX570 and saying only 1.25GB is being used. As a 3.5GB card GTX970 would have lost sales to R9 290 easily.
It's not clear to me if this means the main 3.5GB has lower bandwidth because the last 0.5GB is not used most of the time? Is that 0.5GB a physical chip or a range of adresses on all 8 memory chips?
What would have been a honest start? To say its 3.5GB, when it does have 4GB? Or to tell that 3.5GB is fast, and 0.5GB is slower, because the GPU is stripped down from the 980, and it just cant do more?
And there also the fact, that users found a lot of other, older cards, that are also affected by this slowing down upper memory part. In fact a lot of GPU-s, even 290-s with that specific syntethic benchmark. Some more, some less. I think both GPU companies are using this techique to address memory. It is always highly dependent on the GPU architecture, if it can use exactly 4GB, or 2GB, or just a bit less.
It's not clear to me if this means the main 3.5GB has lower bandwidth because the last 0.5GB is not used most of the time? Is that 0.5GB a physical chip or a range of adresses on all 8 memory chips?
I am not 100% how memory BW is allocated but it would mean the card has 7/8th the memory bandwidth if it has 3.5GB. That would be 256-Bit down to 224-Bit
What would have been a honest start? To say its 3.5GB, when it does have 4GB? Or to tell that 3.5GB is fast, and 0.5GB is slower, because the GPU is stripped down from the 980, and it just cant do more?
And there also the fact, that users found a lot of other, older cards, that are also affected by this slowing down upper memory part. In fact a lot of GPU-s, even 290-s with that specific syntethic benchmark. Some more, some less. I think both GPU companies are using this techique to address memory. It is always highly dependent on the GPU architecture, if it can use exactly 4GB, or 2GB, or just a bit less.
Say that 3,5GB ran at 224GB/s and the other 0,5GB runs and therefor cuts the entire amount of memory back to a puny 20GB/s bandwidth.
Also that bench is meant to be run headless otherwise it isn't an indication of anything all all cards would fail to meet their bandwidth numbers at those ranges of memory usage. Now there are reports of hiccups so there are implications. Also not the Nvidia given numbers are named "example" as if they were not representative and this might be due to users reporting hiccups despite the frame rate being constant.
Status
Not open for further replies.
You have insufficient privileges to reply here.
Related Threads
?
?
?
?
?
Ask a question
Ask a question
Overclock.net
27.8M posts
541.2K members
Since 2004
A forum community dedicated to overclocking enthusiasts and testing the limits of computing. Come join the discussion about computing, builds, collections, displays, models, styles, scales, specifications, reviews, accessories, classifieds, and more!