• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

GPUz vRAM use. What is it really showing us?

W1zzard

Administrator
Staff member
Joined
May 14, 2004
Messages
18,597 (3.47/day)
Likes
20,665
Processor Core i7-4790K
Memory 16 GB
Video Card(s) GTX 1080
Display(s) 30" 2560x1600 + 19" 1280x1024
Software Windows 7
#26
Some people argue that "use" means "actively reading/writing from at a specific instant in time" (which has no meaning in this context i'd claim)
 
Joined
Dec 31, 2009
Messages
13,142 (3.97/day)
Likes
7,505
Location
Ohio
System Name Daily Driver
Processor 7960X 4.4GHz 16c/16t 1.18V
Motherboard MSI XPower Gaming Titanium
Cooling MCR320 + Kuplos Kryos NEXT CPU block
Memory GSkill Trident Z 4x8 GB DDR4 3600 MHz CL16
Video Card(s) NVIDIA RTX 2080 Ti FE
Storage 512GB Patriot Hellfire, 512GB OCZ RD400, 640GB Caviar Black, 2TB Caviar Green
Display(s) 27" Acer Predator 2560x1440 144hz IPS + 27" Acer 2560x1440 75Hz IPS
Case Thermaltake P5
Power Supply EVGA 750W Supernova G2
Benchmark Scores Faster than most of you! Bet on it! :)
#27
Some people argue that "use" means "actively reading/writing from at a specific instant in time" (which has no meaning in this context i'd claim)
kk... were on the same page.

Seems wrong to say gpuz and MSI ab arent really viable for vRAM use/allocation... which is what im trying to get to the bottom of.

Do you agree with that sentiment of gpuz and msi ab being fine or at least 'good enough'to display ram use/allocation?
 
Joined
Sep 17, 2014
Messages
7,440 (4.69/day)
Likes
6,381
Location
Duiven, Netherlands
Processor i7 8700k 4.8Ghz @ 1.31v
Motherboard AsRock Fatal1ty K6 Z370
Cooling beQuiet! Dark Rock Pro 3
Memory 16GB Corsair Vengeance LPX 3200/C16
Video Card(s) MSI GTX 1080 Gaming X @ 2100/5500
Storage Samsung 850 EVO 1TB + Samsung 830 256GB + Crucial BX100 250GB + Toshiba 1TB HDD
Display(s) Eizo Foris FG2421
Case Fractal Design Define C TG
Power Supply EVGA G2 750w
Mouse Logitech G502 Protheus Spectrum
Keyboard Sharkoon MK80 (Brown)
Software W10 x64
#28
vRAM. I dont recall the titles, but I've seen it before in reviews. Again, it isnt common, but happens.

The whole point here is the viability of gpuz and msi ab in reading ram use/allocation. And it looks like it is viable, contrary what has been posted a couple of times around tpu.
Its viable but not to the point of being able to state "it uses X on card A, so you need X on card B'. In that, @John Naylor and others got the right idea.

But at the same time, the bigger the gap between allocated maximums and the VRAM capacity on each card, the higher the chances and frequency of swaps, and depending on game and engine, the greater the chance it results in a performance hit.

So yes the readout may be accurate to some degree, but its usefulness is really on a case by case basis I think.
 

W1zzard

Administrator
Staff member
Joined
May 14, 2004
Messages
18,597 (3.47/day)
Likes
20,665
Processor Core i7-4790K
Memory 16 GB
Video Card(s) GTX 1080
Display(s) 30" 2560x1600 + 19" 1280x1024
Software Windows 7
#29
Its viable but not to the point of being able to state "it uses X on card A, so you need X on card B'.
True, good point. Because games detect GPU / GPU arch / VRAM size and run different code paths. This is similar to comparing memory usage for Photoshop 32-bit with Photoshop 64-bit

Seems wrong to say gpuz and MSI ab arent really viable
If they weren't viable such a feature wouldn't exist.
 
Joined
Nov 13, 2007
Messages
6,760 (1.65/day)
Likes
2,096
Location
Austin Texas
System Name Lucille
Processor Intel i7 7820X Delidded @ 5 Ghz HT OFF / 3.1 Ghz Mesh
Motherboard MSI X299 Tomahawk
Cooling 240mm Corsair H105 Intake
Memory 32 GB Quad 3838 Mhz DDR4 17-17-7-34-340-2T
Video Card(s) Gigabyte GTX 1080 Ti Gaming
Storage 1Tb Samsung 960 Pro m2, 1TB Samsung 850 Pro SSD
Display(s) Dell 24" 2560x1440 144hz, G-Sync @ 165Hz
Case NZXT S340 Elite Black w Magnetic Side Panel
Audio Device(s) Arctis 7
Power Supply FSP HydroG 750W
Mouse $15 500Mhz polling silent click no brand mouse
Keyboard tenkeyless
Software Windows 10 64 Bit
Benchmark Scores 1730 Multi CB, 225 Single CB, superpi 1M - 7.15s
#30
a 1060 3/6 GB vs 580 4/8 GB would be a perfect guinea pig test for this.
 
Joined
Dec 31, 2009
Messages
13,142 (3.97/day)
Likes
7,505
Location
Ohio
System Name Daily Driver
Processor 7960X 4.4GHz 16c/16t 1.18V
Motherboard MSI XPower Gaming Titanium
Cooling MCR320 + Kuplos Kryos NEXT CPU block
Memory GSkill Trident Z 4x8 GB DDR4 3600 MHz CL16
Video Card(s) NVIDIA RTX 2080 Ti FE
Storage 512GB Patriot Hellfire, 512GB OCZ RD400, 640GB Caviar Black, 2TB Caviar Green
Display(s) 27" Acer Predator 2560x1440 144hz IPS + 27" Acer 2560x1440 75Hz IPS
Case Thermaltake P5
Power Supply EVGA 750W Supernova G2
Benchmark Scores Faster than most of you! Bet on it! :)
#31
Thank you. Hopefully moving forward we dont see this again as a reply to suggesting gpuz to see memory use/allocation.

...Actually you can not. I hesitate to say that people are wrong when this statement is made .... they have simply been miinformed...
 
Joined
Mar 18, 2015
Messages
1,316 (0.94/day)
Likes
748
Location
Long Island
#32
Right... and I think overall I either misunderstood the point or the analogies the guy was trying to make... here is the support....

I can refute the statement that cards with larger memory will request more memory' as a blanket statement. It seems to depend in title and amount of vram on the card. But I've seen 4/8/11 GB cards allocate the same amount of memory in most titles... does everyone else share that sentiment?
Thank you ! for conforming exactly what I am saying. It most certainly,as you said, will "depend on title and amount of vram on the card". So if the the game sees 4 GB, it might allocate 2.5 GB ... if it sees 8 GB it might allocate 4.5 GB.

I have posted multiple references multiple times, perhaps reading the would have helped. Let's start here 1st with the RAM amount issue:

http://alienbabeltech.com/main/gtx-770-4gb-vs-2gb-tested/3/

There is one last thing to note with Max Payne 3: It would not normally allow one to set 4xAA at 5760×1080 with any 2GB card as it ***claims to require 2750MB***. However, when we replaced the 4GB GTX 770 with the 2GB version, the game allowed the setting. And there were no slowdowns, stuttering, nor any performance differences that we could find between the two GTX 770s.
Now the entire review contradicts what everyone was saying back then .... "make sure you get the 4 GB model". Those test shows without any shadow of date that 2 or 4GB was irrelevant. The only time that VRAM mattered was when settings were high enough as to make the game was unplayable . Whoo hooo 4 GB model got me 11.8% more fps in Sleeping Dogs at 2560 x 1600 ... who cared, you went from 13.5 fps to 15.9, ... unplayable with 2 or 4 GB. Same results with 6xx series by Puget Sound, same with, same rsults at Guru3D with 9xx, same results by Extremetech, same results in every review of this type I have ever read....same resukts with 10xx here on TPU. Most games are not affected at 1080p and most of those at 1440p. No doubt if ya search long and hard enough you can find a game which bucks the results, hard to explain why, but a porr console port is one known cause as the process gives little concern to such things.

Clearly, in the case of Max Payne, the GPU was requesting 2.750 GB and therefore would not install with only 2 GB, and yet it played with no discernable difference .... no o significant difference in performance, image quality, or user experience between them. Same was found to be true with other links previous posted for Puget Sound, Guru3D, ExtremeTech and Techpowerups. The simple fact is most games are not affected at 1080p with 3 GB, the ones that are are anomalies.... look at it this way, if 3 GB isn't enough or 1080p, then no card exists that has enough VRAM for 2160 ... aka 4 x 1080p. If 3 GB id no good at 1080p, then with 4 times the pixels, 4 x 3GB is no good for 2160p. The data is there, just have to read it.

In Wizzard's MSI 1060 3Gb review, he noted unexpected performance drops in Hitman and Tomb Raider, no explanation was presented nor do we have a means (that I can see) in determining why that is so given there are multiple differences between the cards. Is it the shaders ? is it the RAM, no way to tell. But it is very clearly stated that "Other games seem completely unaffected by having 3 GB less VRAM at their disposal, especially at 1080p". I can't see any ambiguity in Wizzard's statement. As was shown , repeatedly, if 3 GB is an issue at 1080p, then it absolutely must follow that it will be a bigger issue at 1440p. Wizzard's test results show no significant difference in performance over the 18 game test suite between 1080p and 1440p, both showing 6% . Even when we get to 2160p, where VRAM does have a perfomane impact, the games, most of the games are unplyable and the other barely so as the GPU can not keep up.

With regard to "VRAM Usage" as used in available utilities, nvidia is paraphrasing Indigo Montoya (Princess Bride) saying " that word doesn't mean what you think it means". The link to nVidia's statement has been posted multiple times and no one has refuted it. It does provide the only explanation I have heard that "works" with the results above. If not, the other thread which birthed this one, has some issues as the OP say 2.0 to 2.5 GB IIRC, I'd like to see how it would possible show the same if installed witha 2 GB card. I'll check .... I have a pair of 560 2Gb laying around, I will see what happens.

https://www.extremetech.com/gaming/...y-x-faces-off-with-nvidias-gtx-980-ti-titan-x

We spoke to Nvidia’s Brandon Bell on this topic, who told us the following: “None of the GPU tools on the market report memory usage correctly, whether it’s GPU-Z, Afterburner, Precision, etc. They all report the amount of memory requested by the GPU, not the actual memory usage. Cards will larger memory will request more memory, but that doesn’t mean that they actually use it. They simply request it because the memory is available.
Note the following sentence in the article quoting Mr. Bell "Our own testing backed up this claim; VRAM monitoring is subject to a number of constraints." call it what you want ... allocation or usage but clearly just because you ***see*** 2.75 GB of RAM usage in a utility on a 4GB card does not mean the game won't run at the same fs, quality etc. on a 2 GB card.... regardless of what name you give it, it clearly is having no impact. Let's change the semantics, and focus on the point, the allegation is that seeing usage of 2.X GB in GPU_z proves that the game is being impacted. Clearly, in the referenced thread, GPUz is not going to show 2.X GB or VRAM usage if run on a 2 GB card. The number in any utility proves nothing of the sort. Max Payne showed this definitively and cinclusively .... the other 40 or so games in Alienbabeltech (7xx) showed mo significant hit. Same with Guru3D (9xx) , same with Puget Sound (6xx), and other TPUs tests showed this in 16 /18 games. Because of the difference in shaders, have not eliminated all other impacts on those two.

If there's evidence to the contrary, I'd be anxious to read it. The Max Payne experience clearly shows that the GPU requests more than it actually **needs** as it assigns 2.75 GB and won't allow game to run at those settings if not there. Play the semantic game if ya will but on a 2 GB card, it clearly is not impacted by not having the 2.75 GB the GPU insists on allocating for the game. ... so strictly so that it will not allow you to run it. If Mr. Bell is wrong abut allocation versus usage, then explain Max Payne. Until given, will have to take it that nVidia knows their product. Of all those commenting on the issue, Mr. Bell has the best resume. In the end, the semantics do not matter, whether it's allocation, usage or keebler eleves, the numbers reported by any utility are no indication whatsover that the game's performance is impacted when uitility reads 2.5 GB on a 3 GB card or even a 2 GB card. Whether it's using it, allocating it or whatever, is irrelevant ... if it is [pick a word you prefer] X.Y GB in the utility, that is no indication as to whether it's [pick a word you prefer] it because it needs that as a minimum or just grabbing in "just in case" ... or whatever. It's not indicating that performance is in any way impacted if less is available.. and THAT is the only point of relevance.


a 1060 3/6 GB vs 580 4/8 GB would be a perfect guinea pig test for this.
I'm gonna see if any of the kids has COD3 in the library. Will come back w/ results if its there.
 
Last edited:
Joined
Dec 31, 2009
Messages
13,142 (3.97/day)
Likes
7,505
Location
Ohio
System Name Daily Driver
Processor 7960X 4.4GHz 16c/16t 1.18V
Motherboard MSI XPower Gaming Titanium
Cooling MCR320 + Kuplos Kryos NEXT CPU block
Memory GSkill Trident Z 4x8 GB DDR4 3600 MHz CL16
Video Card(s) NVIDIA RTX 2080 Ti FE
Storage 512GB Patriot Hellfire, 512GB OCZ RD400, 640GB Caviar Black, 2TB Caviar Green
Display(s) 27" Acer Predator 2560x1440 144hz IPS + 27" Acer 2560x1440 75Hz IPS
Case Thermaltake P5
Power Supply EVGA 750W Supernova G2
Benchmark Scores Faster than most of you! Bet on it! :)
#33
So if the the game sees 4 GB, it might allocate 2.5 GB ... if it sees 8 GB it might allocate 4.5 GB.
This varies by title and is not accurate as a blanket statement. Some exhibit this behavior, some do not.

Ive followed up multiple times with additional questions which were not answered... multiple @JOhnNaylors... etc... I had to clear it up. Please understand I am not the only one who had follow up questions to your dissertations that were not answered. Perhaps kindly respond to the follow ups as well...maybe you have notifications off? :)


There is not reason to correct a statement such as "check GPUz/MSI AB for vRAM" use. I'm not trying to split hairs between what is going on under the covers. Sure... in some titles, it will take more than it needs AT THAT MOMENT. But that doesn't mean it won't use it eventually, as was referenced earlier in this thread. In the end, we know these applications show vRAM used/allocated accurately. At least accurately enough for horseshoes and hand grenades. I think the idea of the NVIDIA statement may be lost in the words.
Some people argue that "use" means "actively reading/writing from at a specific instant in time" (which has no meaning in this context i'd claim)
Nobdy gives a darn about what it is crunching this second. What is has to crunch later that is stored already, if it isn't there, incurs some kind of latency. So it is expected that both 'pools' of memory (currently active and ready when needed, lol) are included. The level of penalty that may or may not occur from swapping out or not having that data will vary by titles. But again, these applications give users an idea of vRAM use/allocation.


Thanks for poking your head in. :)
 
Last edited:
Joined
Jan 8, 2017
Messages
3,706 (4.99/day)
Likes
2,823
System Name Good enough
Processor AMD Ryzen R7 1700X - 4.0 Ghz / 1.350V
Motherboard ASRock B450M Pro4
Cooling Scythe Katana 4 - 3x 120mm case fans
Memory 16GB - Corsair Vengeance LPX
Video Card(s) OEM Dell GTX 1080
Storage 1x Samsung 850 EVO 250GB , 1x Samsung 860 EVO 500GB
Display(s) 4K Samsung TV
Case Zalman R1
Power Supply 500W
#34
It most certainly,as you said, will "depend on title and amount of vram on the card". So if the the game sees 4 GB, it might allocate 2.5 GB ... if it sees 8 GB it might allocate 4.5 GB.
Just stop with this nonsensical argument. A game does not allocate memory based on what it "sees", it uses the memory that it needs. Outside of that it may cache or not data in what is left over.
 

W1zzard

Administrator
Staff member
Joined
May 14, 2004
Messages
18,597 (3.47/day)
Likes
20,665
Processor Core i7-4790K
Memory 16 GB
Video Card(s) GTX 1080
Display(s) 30" 2560x1600 + 19" 1280x1024
Software Windows 7
#35
Clearly, in the case of Max Payne, the GPU was requesting
The VRAM requirement numbers you see in the settings for games are always estimated, and usually completely wrong

I can't see any ambiguity in Wizzard's statement.
I don't remember what I wrote, but I wasn't intentionally vague. It's easy to figure out whether the perf diff is because of shaders or memory. If it's significantly bigger than what the other benchmarks show, then it's due to memory.

Nobdy gives a darn about what it is crunching this second. What is has to crunch later that is stored already, if it isn't there, incurs some kind of latency. So it is expected that both 'pools' of memory (currently active and ready when needed, lol) are included. The level of penalty that may or may not occur from swapping out or not having that data will vary by titles. But again, these applications give users an idea of vRAM use/allocation.
I know what you're saying, but this "later" argument would mean that the game has to load every single asset into VRAM, every time, because it'll be used at some point in your start-to-finish playthrough

They all report the amount of memory requested by the GPU.
Is that the actual quote? The GPU does not request anything.

Cards will larger memory will request more memory
Same as before and completely wrong, the card has no control over what memory my application requests
 
Last edited:
Joined
Dec 31, 2009
Messages
13,142 (3.97/day)
Likes
7,505
Location
Ohio
System Name Daily Driver
Processor 7960X 4.4GHz 16c/16t 1.18V
Motherboard MSI XPower Gaming Titanium
Cooling MCR320 + Kuplos Kryos NEXT CPU block
Memory GSkill Trident Z 4x8 GB DDR4 3600 MHz CL16
Video Card(s) NVIDIA RTX 2080 Ti FE
Storage 512GB Patriot Hellfire, 512GB OCZ RD400, 640GB Caviar Black, 2TB Caviar Green
Display(s) 27" Acer Predator 2560x1440 144hz IPS + 27" Acer 2560x1440 75Hz IPS
Case Thermaltake P5
Power Supply EVGA 750W Supernova G2
Benchmark Scores Faster than most of you! Bet on it! :)
#36
the game has to load every single asset into VRAM, every time, because it'll be used at some point in your start-to-finish playthrough
Does it though?

Im not trying to be difficult here...but why would it have to do that?
 

rtwjunkie

PC Gaming Enthusiast
Supporter
Joined
Jul 25, 2008
Messages
11,161 (2.91/day)
Likes
15,797
Location
Louisiana -Laissez les bons temps rouler!
System Name Bayou Devil
Processor Core i7-4790k 4.4Ghz @ 1.18v
Motherboard ASUS Z97 Deluxe
Cooling All air: 2x140mm Fractal exhaust; 3x 140mm Cougar Intake; Enermax T40F CPU cooler
Memory 2x 8GB Mushkin Redline DDR-3 1866
Video Card(s) MSI GTX 1080Ti Gaming X
Storage 1x 500 MX500 SSD; 2x 2TB WD Black; 1x4TB WD Black;1x 2TB WD Green (eSATA)
Display(s) HP 27q 27" IPS @ 2560 x 1440
Case Fractal Design Define R4 Black w/Titanium front -windowed
Audio Device(s) Soundblaster Z
Power Supply Seasonic X-850
Mouse Coolermaster Sentinel III
Keyboard Logitech G610 Orion mechanical (Cherry Brown switches)
Software Windows 10 Pro 64-bit (Start10 & Fences 3.0 installed)
#37
But I've seen 4/8/11 GB cards allocate the same amount of memory in most titles... does everyone else share that sentiment?
What I have seen between 4, 6, 8 and 11 GB cards is that it is on a per title basis. Just because a game fills 3.9GB of VRAM on a 4GB card, doesn’t mean it will fill 10.9 on an 11. It will allocate more, for example it might reserve/use 6.5. That tells me the game devs are programming some games to use more if available, but not necessarily the entire amount available as the VRAM size increases.

Again, my observations are not scientific, but are good enough anecdotal evidence to suit me.
 

W1zzard

Administrator
Staff member
Joined
May 14, 2004
Messages
18,597 (3.47/day)
Likes
20,665
Processor Core i7-4790K
Memory 16 GB
Video Card(s) GTX 1080
Display(s) 30" 2560x1600 + 19" 1280x1024
Software Windows 7
#38
Joined
Dec 31, 2009
Messages
13,142 (3.97/day)
Likes
7,505
Location
Ohio
System Name Daily Driver
Processor 7960X 4.4GHz 16c/16t 1.18V
Motherboard MSI XPower Gaming Titanium
Cooling MCR320 + Kuplos Kryos NEXT CPU block
Memory GSkill Trident Z 4x8 GB DDR4 3600 MHz CL16
Video Card(s) NVIDIA RTX 2080 Ti FE
Storage 512GB Patriot Hellfire, 512GB OCZ RD400, 640GB Caviar Black, 2TB Caviar Green
Display(s) 27" Acer Predator 2560x1440 144hz IPS + 27" Acer 2560x1440 75Hz IPS
Case Thermaltake P5
Power Supply EVGA 750W Supernova G2
Benchmark Scores Faster than most of you! Bet on it! :)
#39
Right. Its title dependant.

In my experience, most stay about the same vram use/allocation regardless of GPU capacity. I wish I could recall the couple titles I saw vram 'scale' with available capacity...but it was a minority. :)
 

crazyeyesreaper

Chief Broken Rig
Staff member
Joined
Mar 25, 2009
Messages
9,176 (2.56/day)
Likes
4,033
Location
04578
#41
vRAM. I dont recall the titles, but I've seen it before in reviews. Again, it isnt common, but happens.

The whole point here is the viability of gpuz and msi ab in reading ram use/allocation. And it looks like it is viable, contrary what has been posted a couple of times around tpu.
One of those games is the Total War series, Since around Shogun 2 or so the game would auto lower settings automatically without telling the user if the Vram limit was reached. Its why they have an unlimited vram check box.
 

W1zzard

Administrator
Staff member
Joined
May 14, 2004
Messages
18,597 (3.47/day)
Likes
20,665
Processor Core i7-4790K
Memory 16 GB
Video Card(s) GTX 1080
Display(s) 30" 2560x1600 + 19" 1280x1024
Software Windows 7
#42
would auto lower settings automatically without telling the user if the Vram limit was reached. Its why they have an unlimited vram check box.
Ah yes, I've seen that in very few games, too, but all I cared about had a way to "unlock" the settings
 

crazyeyesreaper

Chief Broken Rig
Staff member
Joined
Mar 25, 2009
Messages
9,176 (2.56/day)
Likes
4,033
Location
04578
#43
Yeah in the total war games it was added after people got pissed off wondering why Ultra settings looked like shit, when in reality the game was just auto lowering in game quality settings per battle based on Vram allocation lol.
 
Top