Overclock.net › Forums › Graphics Cards › NVIDIA › GTX 970s can only use 3.5GB of 4GB VRAM issue
New Posts  All Forums:Forum Nav:

GTX 970s can only use 3.5GB of 4GB VRAM issue - Page 8

post #71 of 2990
With paging disabled it stops loading around 3.9GB and says:



but continues using 3.9GB.



I'll test some games when I have more time.

Edit:

No errors with SLI disabled.
Edited by nersty - 1/14/15 at 2:03pm
Jigglypuff
(9 items)
 
Snorlax
(13 items)
 
 
CPUMotherboardRAMHard Drive
amd 5350 ASRock AM1H-ITX 8gb gskill 3tb wd red 
Hard DriveHard DriveOSPower
3tb wd red sandisk Debian 300w Corsair 
Case
Cooler Master Elite 130 
CPUMotherboardGraphicsRAM
R7 1800X asus ch6 hero Asus Titan X @1.5Ghz g.skill 3200 
Hard DriveCoolingOSMonitor
ocz vertex 3 h220-x, 2x ek-wb, monsta 120mmx80mm windows 10 pro rog swift 
KeyboardPowerCaseMouse
g.skill KM570 evga 750 g2 rosewill blackhawk g502 
  hide details  
Reply
Jigglypuff
(9 items)
 
Snorlax
(13 items)
 
 
CPUMotherboardRAMHard Drive
amd 5350 ASRock AM1H-ITX 8gb gskill 3tb wd red 
Hard DriveHard DriveOSPower
3tb wd red sandisk Debian 300w Corsair 
Case
Cooler Master Elite 130 
CPUMotherboardGraphicsRAM
R7 1800X asus ch6 hero Asus Titan X @1.5Ghz g.skill 3200 
Hard DriveCoolingOSMonitor
ocz vertex 3 h220-x, 2x ek-wb, monsta 120mmx80mm windows 10 pro rog swift 
KeyboardPowerCaseMouse
g.skill KM570 evga 750 g2 rosewill blackhawk g502 
  hide details  
Reply
post #72 of 2990
Me thinks the GTX 970 is not a true 256bit card and more like 192.....
post #73 of 2990
Quote:
Originally Posted by swiftypoison View Post

Me thinks the GTX 970 is not a true 256bit card and more like 192.....
No, if that was the case it would only use 3gb .
The 970 is cut down version of 980 , but both have 256bit bus , my guess is driver is setup for most efficient use amount the 970 can handle .
desktop
(10 items)
 
  
CPUMotherboardGraphicsRAM
Intel i5-3570k Asus PZ77 v-Pro MSI 970 Gaming 4G Crucial Ballistix Tactical Tracer 8GB 
Hard DriveHard DriveCoolingOS
WD Black WD5002AALX 500GB 7200 RPM 32MB Cache ... Samsung 850evo 500GB SATA 6.0Gb/s CM Hyper 212evo Win7 64bit Pro SP1 
PowerCase
Corsair TX650 V2 650W Corsair 500R 
  hide details  
Reply
desktop
(10 items)
 
  
CPUMotherboardGraphicsRAM
Intel i5-3570k Asus PZ77 v-Pro MSI 970 Gaming 4G Crucial Ballistix Tactical Tracer 8GB 
Hard DriveHard DriveCoolingOS
WD Black WD5002AALX 500GB 7200 RPM 32MB Cache ... Samsung 850evo 500GB SATA 6.0Gb/s CM Hyper 212evo Win7 64bit Pro SP1 
PowerCase
Corsair TX650 V2 650W Corsair 500R 
  hide details  
Reply
post #74 of 2990
Quote:
Originally Posted by Edkiefer View Post

my guess is driver is setup for most efficient use amount the 970 can handle .

Then why would the pagefile size have anything to do with this? and the card can handle 3.6, but not 4? sounds unlikely..
Dipsy
(16 items)
 
  
CPUMotherboardGraphicsGraphics
Intel 3930k @ 4.20 Ghz ASUS RAMPAGE Extreme IV x79 gigabyte g1 970 gigabyte g1 970 
RAMHard DriveHard DriveHard Drive
Trident X 32 Gb @ 2400 mhz quad channel 8 * 4 GB Samsung 840 Pro SSD 512GB Samsung 840 Pro SSD 512GB 2 7200 RPM HDD RAID 0 
CoolingOSMonitorKeyboard
Corsair h100i with push/pull Windows 8.1 Samsung 4k U28D590D Corsair k70 RGB 
PowerCaseMouseAudio
Corsair Hx850 Cooler Master HAF X MX 518 seinheiser momentum 
  hide details  
Reply
Dipsy
(16 items)
 
  
CPUMotherboardGraphicsGraphics
Intel 3930k @ 4.20 Ghz ASUS RAMPAGE Extreme IV x79 gigabyte g1 970 gigabyte g1 970 
RAMHard DriveHard DriveHard Drive
Trident X 32 Gb @ 2400 mhz quad channel 8 * 4 GB Samsung 840 Pro SSD 512GB Samsung 840 Pro SSD 512GB 2 7200 RPM HDD RAID 0 
CoolingOSMonitorKeyboard
Corsair h100i with push/pull Windows 8.1 Samsung 4k U28D590D Corsair k70 RGB 
PowerCaseMouseAudio
Corsair Hx850 Cooler Master HAF X MX 518 seinheiser momentum 
  hide details  
Reply
post #75 of 2990
Quote:
Originally Posted by tarasis View Post

I was testing SpaceEngine last night and managed to get over 3.5GB a number of times. Apparently 11% of the time I was in the 3600MB-3800MB region, highest I got was 3846MB. There wasn't much off a FPS hit when I was. This is a single GTX 970 at 1680x1050.


I apologize, I meant it just wouldn't go above regularly in that short testing period to see if manually disabling SLI made a difference. In my earlier post on SpaceEngine I noted it going above ~3.5 GBs (by which I mean around 3584 MBs, which is actually closer to 3600 most of the time than 3500 as "3.5 GBs" might imply) briefly at points, but it didn't sustain it and seemed to come back down to around 3600 MBs again after some time. I've been doing some testing again, and if I zip around between planets quickly, I can force it into the upper 3600s and 3700s more easily, but sometimes I get bizarre stuttering or even single-digit/teen FPS while doing it. Space Engine is sometimes going a bit above, but it just seems to want to stabilize more at ~3600 MBs when it can even though I would suspect it to try to and use the full 4 GBs buffer given how slow it is to load in new textures sometimes. But it's hard to tell sometimes with something like Space Engine if crap performance is related to one particular thing.

Anyway, the really odd thing is I'm still topping out in the same range you also generally are (the absolute peak GPU-Z recorded for me was 3844 MBs) even though I'm also running at 2560x1440 as opposed to 1680x1050. That should be enough to push my 970 to use more VRAM and I'm worried the 980 will perform exactly as I expect to in contrast, by constantly using up its full 4 GB (4096 MB) framebuffer. At 1680x1080, I would guess it's easier for your 970 to work with these supposed memory allocation issues, but if I'm stuck using the same amount of VRAM at over twice the resolution and getting performance deficits from it (stuttering, yet less than 99% GPU load), something seems wrong.


Quote:
Originally Posted by TopicClocker View Post

I see exactly what you mean! thumb.gif

I'm going to try and upload a video today, when I was testing I was also recording using Shadowplay whilst running MSI Afterburner to monitor VRAM, RAM usage, frame-rate and frame-time in the Rivatuner On-Screen display.

I ran 1920x1080 Ultra settings, 2560x1620 Ultra settings and 3840x2160 Ultra settings.

I ran the benchmark each time I changed the resolution and noticed that the VRAM usage wasn't willing to budge above 3.5-3.6GB even at 4K, so I thought "huh, that's kinda weird?"

I know how VRAM works and a card will sometimes use more VRAM than is needed or even not use that much, however it seemed quite odd that the VRAM usage didn't increase much if at all even when I ran it at resolutions much higher than 1080p.

From what I've read is that other people have tested games on both cards and the GTX 980 or other 4GB cards will allocate or use over 3.9GB or 4GB.

But then when I got into the game it was alot different.

The frame-rate at 4K was between 20-30fps, at 2560x1620 it was around 30-40fps+ and at 1920x1080 it was 50-60+ fps all of the tests was done with the GPU at stock, with a hint of CPU-bottlenecking past 60 fps.

At 4K I had seen about 3.8-4GB of VRAM usage, and at 2560x1620 I had seen 3.8-3.9ish, and at 1920x1080 3.8-3.9ish aswell IIRC.

4K wasn't smooth understandably trying to run that insane resolution on a single card at Ultra settings, 2560x1620 was alot smoother, and 1920x1080 was flawless, I ran the test twice so I have two sets of gameplay.

I see what you mean about "extreme conditions" as I suppose you could say 4K is kinda of pushing it on a single card, 2560x1620 is sort of too, I would have ran 2560x1440 but I'm not sure why it wasn't showing up in the resolution options.

I'm going to play Dead Rising 3 and Assassin's Creed Unity.

Thanks for doing this.
Edited by Serandur - 1/14/15 at 10:55pm
Ice-nine
(17 items)
 
  
CPUMotherboardGraphicsRAM
i7-6700K Asus Z170-DELUXE Gigabyte G1 GTX 980 Ti Corsair Vengeance LPX DDR4 3000 
Hard DriveHard DriveOptical DriveCooling
Samsung 850 Evo 500 GB Seagate Barracuda 7200 RPM 2 TB Asus DVD Drive DRW-24B1ST/BLK/B/AS Phanteks PH-TC14PE 
OSMonitorKeyboardPower
Windows 8.1 Pro AOC Q2770PQU Corsair K95 RGB - Cherry MX Blue EVGA Supernova G2 750W 
CaseMouseMouse PadAudio
NZXT Phantom 410 Gunmetal Edition Logitech G502 Proteus Core Razer Goliathus Speed Edition Small Cyber Acoustics Satellite CA-3602 
Audio
Oppo PM-3 Closed Back Planar Magnetic Headphones 
  hide details  
Reply
Ice-nine
(17 items)
 
  
CPUMotherboardGraphicsRAM
i7-6700K Asus Z170-DELUXE Gigabyte G1 GTX 980 Ti Corsair Vengeance LPX DDR4 3000 
Hard DriveHard DriveOptical DriveCooling
Samsung 850 Evo 500 GB Seagate Barracuda 7200 RPM 2 TB Asus DVD Drive DRW-24B1ST/BLK/B/AS Phanteks PH-TC14PE 
OSMonitorKeyboardPower
Windows 8.1 Pro AOC Q2770PQU Corsair K95 RGB - Cherry MX Blue EVGA Supernova G2 750W 
CaseMouseMouse PadAudio
NZXT Phantom 410 Gunmetal Edition Logitech G502 Proteus Core Razer Goliathus Speed Edition Small Cyber Acoustics Satellite CA-3602 
Audio
Oppo PM-3 Closed Back Planar Magnetic Headphones 
  hide details  
Reply
post #76 of 2990
@Serandur:

There's another monitoring number that I personally don't consider to be a trustworthy indicator of pretty much anything: GPU usage. You really, really can't say that stuttering at less than 99% GPU usage means VRAM bottlenecking.

I still appreciate the effort, but the monitoring tools that are available for the general public are all coming short of pinpointing the issue (if there is one). The best and pretty much the only valid test method I can see would be for someone with both a 970 and a 980 running a whole bunch of VRAM heavy benchmarks, showing the performance and frametime graphs. Perhaps the best way would be to compare an OC'd 970 to a stock 980, so the GPU power itself would be pretty similar.

If there is something actually wrong here, it must have an effect on performance and the end user experience. Otherwise it's just a misinterpretation of monitoring data. A subjective 970-only test can't get to the bottom of this. This issue has gotten enough attention for it to be worthwhile for some professional hardware site to look into it, I'm hoping someone will bite.
The Air Tunnel
(10 items)
 
  
CPUMotherboardGraphicsRAM
i7-4820K (4500MHz@1.28V) P9X79 Sapphire R9 290 Tri-X New Edition Team Group Vulcan 4x4GB 2133MHz 
Hard DriveCoolingOSPower
2x240GB SSD @RAID 0 Noctua NH-D14 SE2011 Windows 10 Pro Corsair AX750 
CaseAudio
Rosewill Armor EVO hiFace+AudioGD+DT770Pro 
  hide details  
Reply
The Air Tunnel
(10 items)
 
  
CPUMotherboardGraphicsRAM
i7-4820K (4500MHz@1.28V) P9X79 Sapphire R9 290 Tri-X New Edition Team Group Vulcan 4x4GB 2133MHz 
Hard DriveCoolingOSPower
2x240GB SSD @RAID 0 Noctua NH-D14 SE2011 Windows 10 Pro Corsair AX750 
CaseAudio
Rosewill Armor EVO hiFace+AudioGD+DT770Pro 
  hide details  
Reply
post #77 of 2990
Quote:
Originally Posted by specopsFI View Post

@Serandur:

There's another monitoring number that I personally don't consider to be a trustworthy indicator of pretty much anything: GPU usage. You really, really can't say that stuttering at less than 99% GPU usage means VRAM bottlenecking.

I still appreciate the effort, but the monitoring tools that are available for the general public are all coming short of pinpointing the issue (if there is one). The best and pretty much the only valid test method I can see would be for someone with both a 970 and a 980 running a whole bunch of VRAM heavy benchmarks, showing the performance and frametime graphs. Perhaps the best way would be to compare an OC'd 970 to a stock 980, so the GPU power itself would be pretty similar.

If there is something actually wrong here, it must have an effect on performance and the end user experience. Otherwise it's just a misinterpretation of monitoring data. A subjective 970-only test can't get to the bottom of this. This issue has gotten enough attention for it to be worthwhile for some professional hardware site to look into it, I'm hoping someone will bite.

GPU usage is a trustworthy indicator of whether the GPU's processing speed is the issue or not, I didn't say it was anything more, not on its own. For SpaceEngine I said as much that the performance issues' cause can't be properly pinpointed because it is switching things into and out of both VRAM and system memory while potentially stressing one or two CPU cores as well, but textures also loaded more slowly once past the 3.5 GB point as well in my case. However, if, as in Skyrim for example, the setting and all major factors including rendering settings with the exception of MSAA remain the same, VRAM allocation caps out at the same place, and GPU usage is kept below full yet freezing/stuttering appears, it is not a stretch to suspect the VRAM as a culprit. I stated before that overfilling a card's framebuffer has a very distinct set of consequences and those are similar to what I'm seeing on my own PC in some of these games.

Of course this isn't proving anything and I agree a more professional examination of it needs to occur (with a 980 too); it's merely conjecture and testing, but that's not pointless if only for the sake of detailed information describing the issue. This is affecting user experience for me. VRAM always has (of course not exclusively) hampered my experience in the same ways; erratic hanging/freezing and transitional stutter/pop-in. They can be very minor as far as some people are concerned, but I'm easily bothered by any such inconsistencies.

Ultimately I agree it needs proper testing, but I just need to clear up that I don't think I am proving anything and am not intending to, I'm just sharing further experiences and discussing it; not claiming to make any conclusions.

I really hope a proper site bite and they had better include frametimes. tongue.gif
Edited by Serandur - 1/15/15 at 12:34am
Ice-nine
(17 items)
 
  
CPUMotherboardGraphicsRAM
i7-6700K Asus Z170-DELUXE Gigabyte G1 GTX 980 Ti Corsair Vengeance LPX DDR4 3000 
Hard DriveHard DriveOptical DriveCooling
Samsung 850 Evo 500 GB Seagate Barracuda 7200 RPM 2 TB Asus DVD Drive DRW-24B1ST/BLK/B/AS Phanteks PH-TC14PE 
OSMonitorKeyboardPower
Windows 8.1 Pro AOC Q2770PQU Corsair K95 RGB - Cherry MX Blue EVGA Supernova G2 750W 
CaseMouseMouse PadAudio
NZXT Phantom 410 Gunmetal Edition Logitech G502 Proteus Core Razer Goliathus Speed Edition Small Cyber Acoustics Satellite CA-3602 
Audio
Oppo PM-3 Closed Back Planar Magnetic Headphones 
  hide details  
Reply
Ice-nine
(17 items)
 
  
CPUMotherboardGraphicsRAM
i7-6700K Asus Z170-DELUXE Gigabyte G1 GTX 980 Ti Corsair Vengeance LPX DDR4 3000 
Hard DriveHard DriveOptical DriveCooling
Samsung 850 Evo 500 GB Seagate Barracuda 7200 RPM 2 TB Asus DVD Drive DRW-24B1ST/BLK/B/AS Phanteks PH-TC14PE 
OSMonitorKeyboardPower
Windows 8.1 Pro AOC Q2770PQU Corsair K95 RGB - Cherry MX Blue EVGA Supernova G2 750W 
CaseMouseMouse PadAudio
NZXT Phantom 410 Gunmetal Edition Logitech G502 Proteus Core Razer Goliathus Speed Edition Small Cyber Acoustics Satellite CA-3602 
Audio
Oppo PM-3 Closed Back Planar Magnetic Headphones 
  hide details  
Reply
post #78 of 2990
Quote:
Originally Posted by Serandur View Post

GPU usage is a trustworthy indicator of whether the GPU's processing speed is the issue or not, I didn't say it was anything more, not on its own. For SpaceEngine I said as much that the performance issues' cause can't be properly pinpointed because it is switching things into and out of both VRAM and system memory while potentially stressing one or two CPU cores as well, but textures also loaded more slowly once past the 3.5 GB point as well in my case. However, if, as in Skyrim for example, the setting and all major factors including rendering settings with the exception of MSAA remain the same, VRAM allocation caps out at the same place, and GPU usage is kept below full yet freezing/stuttering appears, it is not a stretch to suspect the VRAM as a culprit. I stated before that overfilling a card's framebuffer has a very distinct set of consequences and those are similar to what I'm seeing on my own PC in some of these games.

Of course this isn't proving anything and I agree a more professional examination of it needs to occur (with a 980 too); it's merely conjecture and testing, but that's not pointless if only for the sake of detailed information describing the issue. This is affecting user experience for me. VRAM always has (of course not exclusively) hampered my experience in the same ways; erratic hanging/freezing and transitional stutter/pop-in. They can be very minor as far as some people are concerned, but I'm easily bothered by any such inconsistencies.

Ultimately I agree it needs proper testing, but I just need to clear up that I don't think I am proving anything and am not intending to, I'm just sharing further experiences and discussing it; not claiming to make any conclusions.

I really hope a proper site bite and they had better include frametimes. tongue.gif

I don't agree with the bolded part. It would be true if we could monitor the usage for each part of the graphics pipeline but as a single number, GPU usage can't properly tell whether there is an architectural bottleneck within the GPU or a VRAM bottleneck or a system bottleneck somewhere else in the rig. The monitoring tools we have at our disposal are just not accurate enough to really see what's going on here. You've gone to great lengths to shine light on this, let's hope someone with more resources goes all the way with it. thumb.gif
The Air Tunnel
(10 items)
 
  
CPUMotherboardGraphicsRAM
i7-4820K (4500MHz@1.28V) P9X79 Sapphire R9 290 Tri-X New Edition Team Group Vulcan 4x4GB 2133MHz 
Hard DriveCoolingOSPower
2x240GB SSD @RAID 0 Noctua NH-D14 SE2011 Windows 10 Pro Corsair AX750 
CaseAudio
Rosewill Armor EVO hiFace+AudioGD+DT770Pro 
  hide details  
Reply
The Air Tunnel
(10 items)
 
  
CPUMotherboardGraphicsRAM
i7-4820K (4500MHz@1.28V) P9X79 Sapphire R9 290 Tri-X New Edition Team Group Vulcan 4x4GB 2133MHz 
Hard DriveCoolingOSPower
2x240GB SSD @RAID 0 Noctua NH-D14 SE2011 Windows 10 Pro Corsair AX750 
CaseAudio
Rosewill Armor EVO hiFace+AudioGD+DT770Pro 
  hide details  
Reply
post #79 of 2990
Hello smile.gif

I only have ~ 4GB vram usage in Mordor benchmark when PF is ON = 8gigs, max seettings in game, HD Texture Pack and insane resolution 150%= 5760x3240 (all DSR NVCP active)

proof: 6f78df381825336.jpg

Everything under with PF on or off gave me ~3500 megabytes vram usage

Kombustor 3.5.1 x64 error out of memory with PF off

with PF ON: 6993b8381841193.jpg

Edit1:

MENU FHD bdaf3c381827058.jpg

MENU 150% 35f841381827155.jpg

Benchmark 150%

336235381827306.jpg

924479381827365.jpg

My PC:
3770k@4.5Ghz
MSI z77g45
2x4GB Corsair Vengeance 2133mhz
thermaltake smart SE 730W
gigabyte gtx 970 G1 rev 1.0 - samsung memory - bios 84.04.1F.00.60 - drivers 347.09 WHQL - windows 7 HP x64

So... is it normal? THX and help me if it's problem with my GPU
Edited by lamparcicho - 1/15/15 at 3:53am
post #80 of 2990
Quote:
Originally Posted by specopsFI View Post

I don't agree with the bolded part. It would be true if we could monitor the usage for each part of the graphics pipeline but as a single number, GPU usage can't properly tell whether there is an architectural bottleneck within the GPU or a VRAM bottleneck or a system bottleneck somewhere else in the rig. The monitoring tools we have at our disposal are just not accurate enough to really see what's going on here. You've gone to great lengths to shine light on this, let's hope someone with more resources goes all the way with it. thumb.gif


I think its safe to say the bottleneck is on the card itself and somehow involves the interface of the vram
But i do agree that with our current software we can't monitor everything, like power delivery to memory chips, etc etc

What we do know is that it has to be something that is done differently on the 980 (they are essentially the same chip)
Edited by cocopopsonfire - 1/15/15 at 3:59am
New Posts  All Forums:Forum Nav:
  Return Home
  Back to Forum: NVIDIA
Overclock.net › Forums › Graphics Cards › NVIDIA › GTX 970s can only use 3.5GB of 4GB VRAM issue