970 memory allocation issue revisited

Passive aggressive much?

I notice you did not address the issues I pointed out to your response.

Again, not saying that the recording is graphical evidence in support of the supposed 3.5GB issue. Just saying that it seems a tad coincidental. Of course, it could just be how shadowplay records freezing.. however..

The fact that while glitching shadowplay recorded movement, whilst the on screen was just a freeze... I don't know. I'm not sure what that signifies, if anything.

 

was probably combined with microsoft and nvidia only to resolve this in windows10

at 7 and 8.1 will remain bug to force users to migrate to Windows 10 if they want to use the 4GB of gtx970 without loss in speed

 

It means usable in whatever way the operating system and D3D sees fit to use it.

I've lost count of how many times you've claimed the card only has 3.5GB vram. It's OK, we get it.Can we please cut the wild speculation and concentrate on resolving this one way or another.

 

Zotac 970 4GB (not the Omega/Extreme)

Aero Disabled - Monitor Disconnected.

:bang:

Code:

Nai's Benchmark, edited by VultureX
  Device: GeForce GTX 970 (4.00 GB)
  Memory Bus Width (bits): 256
  Peak Theoretical DRAM Bandwidth (GB/s): 224.320000

Allocating Memory . . .
Chunk Size: 128 MiByte
Allocated 30 Chunks
Allocated 3840 MiByte
Benchmarking DRAM
DRAM-Bandwidth of Chunk no. 0 (0 MiByte to 128 MiByte):155.85 GByte/s
DRAM-Bandwidth of Chunk no. 1 (128 MiByte to 256 MiByte):155.85 GByte/s
DRAM-Bandwidth of Chunk no. 2 (256 MiByte to 384 MiByte):155.91 GByte/s
DRAM-Bandwidth of Chunk no. 3 (384 MiByte to 512 MiByte):155.85 GByte/s
DRAM-Bandwidth of Chunk no. 4 (512 MiByte to 640 MiByte):155.90 GByte/s
DRAM-Bandwidth of Chunk no. 5 (640 MiByte to 768 MiByte):155.85 GByte/s
DRAM-Bandwidth of Chunk no. 6 (768 MiByte to 896 MiByte):155.92 GByte/s
DRAM-Bandwidth of Chunk no. 7 (896 MiByte to 1024 MiByte):155.74 GByte/s
DRAM-Bandwidth of Chunk no. 8 (1024 MiByte to 1152 MiByte):155.87 GByte/s
DRAM-Bandwidth of Chunk no. 9 (1152 MiByte to 1280 MiByte):155.93 GByte/s
DRAM-Bandwidth of Chunk no. 10 (1280 MiByte to 1408 MiByte):155.88 GByte/s
DRAM-Bandwidth of Chunk no. 11 (1408 MiByte to 1536 MiByte):155.91 GByte/s
DRAM-Bandwidth of Chunk no. 12 (1536 MiByte to 1664 MiByte):155.87 GByte/s
DRAM-Bandwidth of Chunk no. 13 (1664 MiByte to 1792 MiByte):155.94 GByte/s
DRAM-Bandwidth of Chunk no. 14 (1792 MiByte to 1920 MiByte):155.76 GByte/s
DRAM-Bandwidth of Chunk no. 15 (1920 MiByte to 2048 MiByte):155.87 GByte/s
DRAM-Bandwidth of Chunk no. 16 (2048 MiByte to 2176 MiByte):155.93 GByte/s
DRAM-Bandwidth of Chunk no. 17 (2176 MiByte to 2304 MiByte):155.88 GByte/s
DRAM-Bandwidth of Chunk no. 18 (2304 MiByte to 2432 MiByte):155.96 GByte/s
DRAM-Bandwidth of Chunk no. 19 (2432 MiByte to 2560 MiByte):155.94 GByte/s
DRAM-Bandwidth of Chunk no. 20 (2560 MiByte to 2688 MiByte):155.89 GByte/s
DRAM-Bandwidth of Chunk no. 21 (2688 MiByte to 2816 MiByte):155.91 GByte/s
DRAM-Bandwidth of Chunk no. 22 (2816 MiByte to 2944 MiByte):155.88 GByte/s
DRAM-Bandwidth of Chunk no. 23 (2944 MiByte to 3072 MiByte):155.88 GByte/s
DRAM-Bandwidth of Chunk no. 24 (3072 MiByte to 3200 MiByte):22.35 GByte/s
DRAM-Bandwidth of Chunk no. 25 (3200 MiByte to 3328 MiByte):22.35 GByte/s
DRAM-Bandwidth of Chunk no. 26 (3328 MiByte to 3456 MiByte):22.35 GByte/s
DRAM-Bandwidth of Chunk no. 27 (3456 MiByte to 3584 MiByte):27.52 GByte/s
DRAM-Bandwidth of Chunk no. 28 (3584 MiByte to 3712 MiByte):12.21 GByte/s
DRAM-Bandwidth of Chunk no. 29 (3712 MiByte to 3840 MiByte):12.11 GByte/s
Benchmarking L2-Cache
L2-Cache-Bandwidth of Chunk no. 0 (0 MiByte to 128 MiByte):398.81 GByte/s
L2-Cache-Bandwidth of Chunk no. 1 (128 MiByte to 256 MiByte):398.70 GByte/s
L2-Cache-Bandwidth of Chunk no. 2 (256 MiByte to 384 MiByte):398.77 GByte/s
L2-Cache-Bandwidth of Chunk no. 3 (384 MiByte to 512 MiByte):398.83 GByte/s
L2-Cache-Bandwidth of Chunk no. 4 (512 MiByte to 640 MiByte):398.77 GByte/s
L2-Cache-Bandwidth of Chunk no. 5 (640 MiByte to 768 MiByte):398.83 GByte/s
L2-Cache-Bandwidth of Chunk no. 6 (768 MiByte to 896 MiByte):398.86 GByte/s
L2-Cache-Bandwidth of Chunk no. 7 (896 MiByte to 1024 MiByte):398.80 GByte/s
L2-Cache-Bandwidth of Chunk no. 8 (1024 MiByte to 1152 MiByte):398.69 GByte/s
L2-Cache-Bandwidth of Chunk no. 9 (1152 MiByte to 1280 MiByte):398.77 GByte/s
L2-Cache-Bandwidth of Chunk no. 10 (1280 MiByte to 1408 MiByte):398.79 GByte/s
L2-Cache-Bandwidth of Chunk no. 11 (1408 MiByte to 1536 MiByte):398.77 GByte/s
L2-Cache-Bandwidth of Chunk no. 12 (1536 MiByte to 1664 MiByte):398.78 GByte/s
L2-Cache-Bandwidth of Chunk no. 13 (1664 MiByte to 1792 MiByte):398.91 GByte/s
L2-Cache-Bandwidth of Chunk no. 14 (1792 MiByte to 1920 MiByte):398.80 GByte/s
L2-Cache-Bandwidth of Chunk no. 15 (1920 MiByte to 2048 MiByte):398.95 GByte/s
L2-Cache-Bandwidth of Chunk no. 16 (2048 MiByte to 2176 MiByte):398.83 GByte/s
L2-Cache-Bandwidth of Chunk no. 17 (2176 MiByte to 2304 MiByte):398.92 GByte/s
L2-Cache-Bandwidth of Chunk no. 18 (2304 MiByte to 2432 MiByte):398.61 GByte/s
L2-Cache-Bandwidth of Chunk no. 19 (2432 MiByte to 2560 MiByte):398.79 GByte/s
L2-Cache-Bandwidth of Chunk no. 20 (2560 MiByte to 2688 MiByte):398.77 GByte/s
L2-Cache-Bandwidth of Chunk no. 21 (2688 MiByte to 2816 MiByte):398.70 GByte/s
L2-Cache-Bandwidth of Chunk no. 22 (2816 MiByte to 2944 MiByte):398.73 GByte/s
L2-Cache-Bandwidth of Chunk no. 23 (2944 MiByte to 3072 MiByte):398.73 GByte/s
L2-Cache-Bandwidth of Chunk no. 24 (3072 MiByte to 3200 MiByte):71.21 GByte/s
L2-Cache-Bandwidth of Chunk no. 25 (3200 MiByte to 3328 MiByte):71.21 GByte/s
L2-Cache-Bandwidth of Chunk no. 26 (3328 MiByte to 3456 MiByte):71.22 GByte/s
L2-Cache-Bandwidth of Chunk no. 27 (3456 MiByte to 3584 MiByte):86.84 GByte/s
L2-Cache-Bandwidth of Chunk no. 28 (3584 MiByte to 3712 MiByte):12.16 GByte/s
L2-Cache-Bandwidth of Chunk no. 29 (3712 MiByte to 3840 MiByte):12.17 GByte/s
Presione una tecla para continuar . . .

 

I never claimed it has 3.5. I said, only, that if these tests are accurate (which I'm not sure they are), that only 3.5ish GB are actually usable.

I've also said (or at least implied) that I'm not sure if there is an actual issue, or that the issue is what it appears to be.

However, arguing from the point that if there is, in fact, an issue where some half a GB of video memory performs well under par, that the fact that yes, you can stuff some information in there if you try hard enough, does not make that memory "usable", and that the description of the cards I bought would not, in fact, match the advertised specifications.

However, if everything is fine, great.

I'm tempted to buy a 980 just to make comparisons, but I already spent my play money this week, would have to wait a week or two.

I'm as happy to put this issue to bed as anyone else, be it in favor of one camp or another.

I just don't like the idea that I may have paid for something I didn't get.

If it turns out I got what I paid for, all the better, as it will save me RMA turnaround time or at the very least, being required to do BIOS updates, etc.

 

yeah, I'm good and see no reason to upgrade the 780tiSC at this time. am disappointed to see the 970 perform under specs. hope a driver can correct these issues

 

It looks like my hunch which I posted here was correct.

4GB VRAM = 4096 MB
4096 X 52 / 64 = 3328 MB



Notice that the bandwidth drops dramatically as the VRAM usage hits 3328 MB. So that means that the problem is really the hardware limitation. It's not a bug, so it can't be fixed with any update or patch.

 

my findings if you try to run this test when using NVidia surround 3 x AOC 1080p monitors all connected via DP 1.2 connections it starts to run then the Nvidia driver stops responding although with out Nvidia surround enabled I get pretty much the same results as everyone else I think it can use a little less ram when using 3 monitors, not sure if any of this adds up to anything but just thought I post my findings using DP 1.2 connectors/cables

 

My 780ti Results, note I have no iGPU and have no idea what headless mode is

Also my gpu drivers failed and recovered at the end of the test, not sure if that would skew the results at all.

Code:

Nai's Benchmark, edited by VultureX
  Device: GeForce GTX 780 Ti (3.00 GB)
  Memory Bus Width (bits): 384
  Peak Theoretical DRAM Bandwidth (GB/s): 336.000000

Allocating Memory . . .
Chunk Size: 128 MiByte
Allocated 22 Chunks
Allocated 2816 MiByte
Benchmarking DRAM
DRAM-Bandwidth of Chunk no. 0 (0 MiByte to 128 MiByte):288.82 GByte/s
DRAM-Bandwidth of Chunk no. 1 (128 MiByte to 256 MiByte):292.04 GByte/s
DRAM-Bandwidth of Chunk no. 2 (256 MiByte to 384 MiByte):292.62 GByte/s
DRAM-Bandwidth of Chunk no. 3 (384 MiByte to 512 MiByte):292.21 GByte/s
DRAM-Bandwidth of Chunk no. 4 (512 MiByte to 640 MiByte):292.73 GByte/s
DRAM-Bandwidth of Chunk no. 5 (640 MiByte to 768 MiByte):291.90 GByte/s
DRAM-Bandwidth of Chunk no. 6 (768 MiByte to 896 MiByte):292.27 GByte/s
DRAM-Bandwidth of Chunk no. 7 (896 MiByte to 1024 MiByte):296.41 GByte/s
DRAM-Bandwidth of Chunk no. 8 (1024 MiByte to 1152 MiByte):296.89 GByte/s
DRAM-Bandwidth of Chunk no. 9 (1152 MiByte to 1280 MiByte):297.07 GByte/s
DRAM-Bandwidth of Chunk no. 10 (1280 MiByte to 1408 MiByte):297.06 GByte/s
DRAM-Bandwidth of Chunk no. 11 (1408 MiByte to 1536 MiByte):296.55 GByte/s
DRAM-Bandwidth of Chunk no. 12 (1536 MiByte to 1664 MiByte):296.01 GByte/s
DRAM-Bandwidth of Chunk no. 13 (1664 MiByte to 1792 MiByte):296.15 GByte/s
DRAM-Bandwidth of Chunk no. 14 (1792 MiByte to 1920 MiByte):296.51 GByte/s
DRAM-Bandwidth of Chunk no. 15 (1920 MiByte to 2048 MiByte):296.95 GByte/s
DRAM-Bandwidth of Chunk no. 16 (2048 MiByte to 2176 MiByte):296.91 GByte/s
DRAM-Bandwidth of Chunk no. 17 (2176 MiByte to 2304 MiByte):296.67 GByte/s
DRAM-Bandwidth of Chunk no. 18 (2304 MiByte to 2432 MiByte):296.27 GByte/s
DRAM-Bandwidth of Chunk no. 19 (2432 MiByte to 2560 MiByte):297.69 GByte/s
DRAM-Bandwidth of Chunk no. 20 (2560 MiByte to 2688 MiByte):298.24 GByte/s
DRAM-Bandwidth of Chunk no. 21 (2688 MiByte to 2816 MiByte): 7.86 GByte/s
Benchmarking L2-Cache
L2-Cache-Bandwidth of Chunk no. 0 (0 MiByte to 128 MiByte):548.07 GByte/s
L2-Cache-Bandwidth of Chunk no. 1 (128 MiByte to 256 MiByte):547.94 GByte/s
L2-Cache-Bandwidth of Chunk no. 2 (256 MiByte to 384 MiByte):548.20 GByte/s
L2-Cache-Bandwidth of Chunk no. 3 (384 MiByte to 512 MiByte):548.01 GByte/s
L2-Cache-Bandwidth of Chunk no. 4 (512 MiByte to 640 MiByte):548.53 GByte/s
L2-Cache-Bandwidth of Chunk no. 5 (640 MiByte to 768 MiByte):548.27 GByte/s
L2-Cache-Bandwidth of Chunk no. 6 (768 MiByte to 896 MiByte):548.18 GByte/s
L2-Cache-Bandwidth of Chunk no. 7 (896 MiByte to 1024 MiByte):548.36 GByte/s
L2-Cache-Bandwidth of Chunk no. 8 (1024 MiByte to 1152 MiByte):549.02 GByte/s
L2-Cache-Bandwidth of Chunk no. 9 (1152 MiByte to 1280 MiByte):548.36 GByte/s
L2-Cache-Bandwidth of Chunk no. 10 (1280 MiByte to 1408 MiByte):547.85 GByte/s
L2-Cache-Bandwidth of Chunk no. 11 (1408 MiByte to 1536 MiByte):548.35 GByte/s
L2-Cache-Bandwidth of Chunk no. 12 (1536 MiByte to 1664 MiByte):548.97 GByte/s
L2-Cache-Bandwidth of Chunk no. 13 (1664 MiByte to 1792 MiByte):548.99 GByte/s
L2-Cache-Bandwidth of Chunk no. 14 (1792 MiByte to 1920 MiByte):548.77 GByte/s
L2-Cache-Bandwidth of Chunk no. 15 (1920 MiByte to 2048 MiByte):548.73 GByte/s
L2-Cache-Bandwidth of Chunk no. 16 (2048 MiByte to 2176 MiByte):548.01 GByte/s
L2-Cache-Bandwidth of Chunk no. 17 (2176 MiByte to 2304 MiByte):548.95 GByte/s
L2-Cache-Bandwidth of Chunk no. 18 (2304 MiByte to 2432 MiByte):548.92 GByte/s
L2-Cache-Bandwidth of Chunk no. 19 (2432 MiByte to 2560 MiByte):548.08 GByte/s
Kernel launch failed: the launch timed out and was terminated
Kernel launch failed: the launch timed out and was terminated
Press any key to continue . . .

EDIT: Result with SLi disabled seems more relavent.

 

So this means either they decide not to address the problem and we're out in the cold, or they address it by replacing hardware?

 

Thanks.

Now comparing some data.

This is GTX 780 from Loophole35 :
DRAM-Bandwidth of Chunk no. 0 (0 MiByte to 128 MiByte):249.40 GByte/s
DRAM-Bandwidth of Chunk no. 1 (128 MiByte to 256 MiByte):251.34 GByte/s
DRAM-Bandwidth of Chunk no. 2 (256 MiByte to 384 MiByte):250.22 GByte/s
DRAM-Bandwidth of Chunk no. 3 (384 MiByte to 512 MiByte):249.63 GByte/s
DRAM-Bandwidth of Chunk no. 4 (512 MiByte to 640 MiByte):249.36 GByte/s
DRAM-Bandwidth of Chunk no. 5 (640 MiByte to 768 MiByte):249.41 GByte/s
L2-Cache-Bandwidth of Chunk no. 0 (0 MiByte to 128 MiByte):395.03 GByte/s
L2-Cache-Bandwidth of Chunk no. 1 (128 MiByte to 256 MiByte):395.01 GByte/s
L2-Cache-Bandwidth of Chunk no. 2 (256 MiByte to 384 MiByte):395.05 GByte/s
L2-Cache-Bandwidth of Chunk no. 3 (384 MiByte to 512 MiByte):395.02 GByte/s
L2-Cache-Bandwidth of Chunk no. 4 (512 MiByte to 640 MiByte):395.15 GByte/s
L2-Cache-Bandwidth of Chunk no. 5 (640 MiByte to 768 MiByte):395.02 GByte/s

This is GTX 780 TI from anxious_f0x:
DRAM-Bandwidth of Chunk no. 0 (0 MiByte to 128 MiByte):288.82 GByte/s
DRAM-Bandwidth of Chunk no. 1 (128 MiByte to 256 MiByte):292.04 GByte/s
DRAM-Bandwidth of Chunk no. 2 (256 MiByte to 384 MiByte):292.62 GByte/s
DRAM-Bandwidth of Chunk no. 3 (384 MiByte to 512 MiByte):292.21 GByte/s
DRAM-Bandwidth of Chunk no. 4 (512 MiByte to 640 MiByte):292.73 GByte/s
DRAM-Bandwidth of Chunk no. 5 (640 MiByte to 768 MiByte):291.90 GByte/s
L2-Cache-Bandwidth of Chunk no. 0 (0 MiByte to 128 MiByte):548.07 GByte/s
L2-Cache-Bandwidth of Chunk no. 1 (128 MiByte to 256 MiByte):547.94 GByte/s
L2-Cache-Bandwidth of Chunk no. 2 (256 MiByte to 384 MiByte):548.20 GByte/s
L2-Cache-Bandwidth of Chunk no. 3 (384 MiByte to 512 MiByte):548.01 GByte/s
L2-Cache-Bandwidth of Chunk no. 4 (512 MiByte to 640 MiByte):548.53 GByte/s
L2-Cache-Bandwidth of Chunk no. 5 (640 MiByte to 768 MiByte):548.27 GByte/s

Both using :
Chunk Size: 128 MiByte
Allocated 22 Chunks
Allocated 2816 MiByte

Notice the bandwidth of DRAM and L2-Cache.
Then have a look on GTX 780 and GTX 780 TI architecture diagrams(both advertised as 384bit).

EDIT: 780 has one GPC less than 780 TI. Both have 6 memory controllers.

From DRAM comparison, it does look like 384bit is in name only. (Effective bit is lower)
From L2-Cache comparison, 384bit is actually useful for data that can fit within L2 cache.

Anyone?

Conclusion?

 

Yes. I bet NVidia knew this all along, that's why they cancelled the release of 8GB variant. NVidia was hoping that people won't notice the few unused MBs, but the problem will be exposed if the 8GB variant reaches the consumers.

 

What games can I run on my GTX 980M to help the investigation?

Already disproved by my 980M testing earlier. If it was in anyway connected to NVIDIA cutting down the number of SMM's in the chip then ALL cut down GM204's would be affected -> 970 (13:1664), 980M (12:1536), 970M (10:1280), 965M (8:1024).

 

Any information on your 980M memory chip arrangement on the board?

 

This is interesting.

http://www.reddit.com/r/pcmasterrace/comments/2tfybe/investigating_the_970_vram_issue/

I dont have DA:I or WatchDogs, but anyone who does and can replicate these tests on their 970s and if results are close to above findings, would be great.

 

This problem only effecs 970 users ?, then i must warn a friend of mine before he buys the 970 lol.

He was going to buy it this monday, i app him right away now not to do it.

 

I think Wolfenstein uses a lot of VRAM right? I will try kicking it up to max AA and 1620p when I get home to see what happens.

 

No really.
skyrim
http://abload.de/img/tesv_2015_01_20_04_28mju9b.png
http://abload.de/img/tesv_2015_01_20_04_53joj4e.png
watchdogs
http://abload.de/img/watch_dogs_2015_01_11w4sxz.jpg
http://abload.de/img/watch_dogs_2015_01_11mmsqt.jpg

 

address it with new hardware called the 970Ti 4Ggig lol joking :/ but who knows wot Nvidia can/will do.

at a guess if it cant be patched maybe Nvidia will just pass it off as an hardware limitation of the hardware ?

 

This isn't my own picture but it is a picture of the 980M layout. It appears I was wrong when I said 4 memory modules before, it's actually 8 by the looks of it:

i.imgur.com/R1RANhB.jpg