• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

Can someone explain this please ( Shared Video Memory )

Soldato
Joined
10 May 2004
Posts
3,480
Location
South UK
Can someone explain the image below to me and what each heading for Memory means?

As you can see, its a GTX 460 with 1gb RAM onboard. The motherboard is an ASUS P8P67 Pro (with no onboard gfx) Socket 1155 and i5 2500k CPU and I have 4gb system RAM.

I had always assumed with desktop motherboards (not laptops) that the gfx card RAM was separate to system RAM and not shared, I have looked through the BIOS but cannot find anything related to RAM and gfx card sharing settings.

Is this normal or should there be no shared memory on an up to date rig?

Im running Vista 64bit.

Many Thanks.

svm.jpg


***Heres a pic from XP, it seems only shared memory is in Vista or Win7.

gfxcard-1.jpg
 
Last edited:
It's nothing to worry about.

The gpu will go to the system memory when it runs out of of dedicated local memory, you'll know it's happening as you will get massive slowdowns.
 
You dont get massive slowdowns from it if you have at least 8 Gb ram. At the most you lose a few FPS over using dedicated Vram instead.

Slowdowns happen when the shared ram reduces your system ram too low for the game, this is a very common problem when using 4 Gb ram and 1 Gb Vram.

But people who have tried using more system ram do not experience anywhere near as much lag when the shared ram is used, and this has actually been sufficient for lots of people to overcome any limitations and eliminate lag spikes / stutter associated with only having 1 Gb Vram at 1200p resolution or less.
 
Last edited:
Windows automatically allocates a certain amount of system memory to be used by the graphics card as required.

As long as the GPU doesn't need more than 1024MB then the shared memory won't be used.

In your case 1024MB dedicated VRAM + 1780 of system RAM = 2804MB available in total.
 
You dont get massive slowdowns from it if you have at least 8 Gb ram. At the most you lose a few FPS over using dedicated Vram instead.

Slowdowns happen when the shared ram reduces your system ram too low for the game, this is a very common problem when using 4 Gb ram and 1 Gb Vram.

But people who have tried using more system ram do not experience anywhere near as much lag when the shared ram is used, and this has actually been sufficient for lots of people to overcome any limitations and eliminate lag spikes / stutter associated with only having 1 Gb Vram at 1200p resolution or less.
Vram on a midrange/high end card is insanely quick compared to snails pace system ram. As soon as the gpu has to reach to the system memory you will hit a slowdown as you drop from 100+gb/s to ~20-30gb/s.

If I fire up metro at 5760x1200 I go past the gpu limits (1.5gb) and start eating into slow system memory and my fps drops to about 5-7. Of which I have 12gb, so more than enough to not have the memory allocation problems with game data.
 
Windows automatically allocates a certain amount of system memory to be used by the graphics card as required.

As long as the GPU doesn't need more than 1024MB then the shared memory won't be used.

In your case 1024MB dedicated VRAM + 1780 of system RAM = 2804MB available in total.

So these values cannnot be changed and it is Windows that determines how much system memory will be shared to top up the gfx memory?
 
This'll be a really thick question perhaps but.. what if your ram speed (mine being 1600mhz) was the same speed as the memory clock on the GPU.. would that then stop the slowdowns?

My card has 1300mhz ram speed (overclocked) and as above my ram is 1600mhz.. wouldnt be a slowdown in that case right?
 
What happens when you run out of VRAM depends on a number of factors - not sure where you pulled the 8gig figure from as the optimal amount of system RAM will depend entirely on OS/other task useage.

The pattern of useage by the GPU will dictate much of what happens i.e. it could have paged 100s of MB out to system RAM but only having to access a couple of MB per frame and you may see no hit at all or it might be only needing a few 10s of MB of system RAM to swap stuff to/from GPU VRAM but due to the high intensity of requirement to access the data result in a massive performance hit... and so on.
 
This'll be a really thick question perhaps but.. what if your ram speed (mine being 1600mhz) was the same speed as the memory clock on the GPU.. would that then stop the slowdowns?

My card has 1300mhz ram speed (overclocked) and as above my ram is 1600mhz.. wouldnt be a slowdown in that case right?

You can't compare directly like that i.e. DDR3 @ 1600MHz system RAM might be capable of 20gig/s bandwidth but the DDR5 @ 1600MHz on your GPU could be pusing 120gig/s - due to a number of things including that the system RAM might be say 64bit wide bus and the GPU 512bit wide bus.
 
Well ....

I really have no clue how this happens.

:confused:

Don't get the point of that, doesn't seem to have anything to do with what I said other than being the same game.

You said you need to have 8gb min in order for the gpu to not slow down horribly when leaching system ram. I have more than that and it happens, thus it's not that reason. If I turn the settings/res down enough to get it just under the 1.5gb cap I get 35fps+ again.

Odd way to measure memory too, 24 Gb of ram is only 3 GB. :D
 
Even more so that the 6x4gb is both slower and has lower minimums, while both the 6x4gb and 6x2gb(who goes and buys 6 sticks of both capacitys for any reason?) use what seem like incredibly unlikely timings, 6-7-6 @ 1600Mhz and 1t, very unlikely and 1t with 6x4gb sticks that have far slower timings?

As for the OP< don't worry about it, its normal, I forget exactly how it works as its been a while but AFAIK its just memory it COULD use in some situations, but basically won't, so meh.

Most situations going to system memory = performance goes bye bye, but as Rroff said, there are situations where only a tiny portion is in system memory and it doesn't kill performance too badly but chances of that happening are fairly slim in general. Even when performance like average FPS is roughly the same you'll be getting jerky stuttery performance. Fact is its pretty hard to go over the limit with most cards because again in general most cards have pretty much enough memory for the overall power the gpu has. a 460gtx isn't really a card you want to game at 1920x1080 with 4xmsaa in, or 2560x1600 with no aa, at 1920x1080 with little to no aa the 460gtx will do pretty good and it won't be using too much memory anyway.
 
Don't get the point of that, doesn't seem to have anything to do with what I said other than being the same game.

Sorry, I should explain ...

1 Gb Vram and 24 Gb Cas 9 = 1800ish points
2 Gb Vram = same as above
1 Gb Vram and 12 Gb Cas 6 ram = 2100ish points, and a 5 fps higher average.

Ram speed / timings were having a bigger impact in Metro than Vram does, I dont know because more Vram should help more at those settings rather than better ram.

This kind of made me think that maybe shared ram actually does work. I still wouldnt ever claim that it is better than dedicated Vram, but for some reason with 12 Gb of fast cas 6 ram, performance with only 1 Gb Vram is better than 2 Gb Vram.

I should have seen slowdown with extra Vram data going into the shared ram, but I never do. Even with the cas 9 4 Gb sticks, I never get lag spikes or stutter in games which are 'known to eat more than 1 Gb Vram' :p
 
Last edited:
Thats ******** tbh something wrong with your testing methodology or just some random weird setting/incompatibility somewhere.
 
Thanks for replies guys.

I think I found my answer after searching Google, its something to do with the behaviour of Vista (and presumably Win7) over Win XP and how memory is allocated depending on rig specs.

See this thread - http://www.tomshardware.co.uk/forum/233624-34-disabling-shared-system-memory

As quoted on the thread, this text was taken from the microsoft document which is available here - http://download.microsoft.com/download/9/c/5/9c5b2167-8017-4bae-9fde-d599bac8184a/GraphicsMemory.doc

In WDDM, the operating system can accurately account for each of the graphics memory contributors and report available memory precisely through new APIs. The following are some of the clients that use this reporting:

Windows System Assessment Tool (WinSAT) checks for the available graphics memory and takes the action to turn off or on the Premium Aero Glass experience based on the amount of available memory.

The Desktop Windows Manager (dwm.exe) is depends on the exact state of the available graphics memory on WDDM systems.

DirectX games and other graphics applications for Windows Vista need to be able to get accurate values describing the state of the graphics memory in the system. An inaccurate graphics memory number could drastically change the game experience for the user, for example.

Thus, Windows Vista enables the critically important capability of reporting the correct amount of graphics memory to the end-user.
 
Thats ******** tbh something wrong with your testing methodology or just some random weird setting/incompatibility somewhere.

It was an accidental discovery, I just wanted to see the difference with my 2 Gb and 4 Gb ram modules.

I have done the Metro benchmarks on two set of 24 Gb Cas 9 ram now (recently got some G Skills that went faulty within 3 weeks and they got refunded), and there was absolutely nothing I could do to match the results of the 2 Gb modules.

What it does prove however, is that the metro benchmark is a BS way of comparing stuff :p
 
Some games do work by actually managing video data rather than just fire and forget upload to VRAM and caching it in system memory, using things like texture streaming, etc. however no idea if metro is one of those and it would make no sense that 1gig + more RAM even with tighter timings/more bandwidth works better than 2gig if its running out of VRAM with 2gig.

At best its some oddity of Metro benchmark as you said.
 
I just looked up the shared memory in XP, heres a pic. No shared memory is shown so I guess its only a feature of Vista or Win 7.

gfxcard-1.jpg
 
Back
Top Bottom