Graphics memory, Windows, and total RAM interaction when gaming

[CakeMonster]

I only have 16Gb RAM since my 32Gb set has an issue with my mobo which forces me to only use two sticks (total 16Gb) (long annoying story). If I buy a new GPU with 8Gb VRAM soon, and in a situation where Windows uses 8Gb+ already, could I get problems with a VRAM heavy game? I seem to remember that you should have free RAM equal to your VRAM.

 

[Bacon1]

You should be fine

 

[SlowSpyder]

I doubt you will run out of vram anytime soon with 8GB of it.

 

[Termie]

...

I seem to remember that you should have free RAM equal to your VRAM.

This is one of those rumors that has spread like wildfire on the 'net despite being completely wrong.

Higher VRAM actually leads to lower system RAM usage.

 

[CakeMonster]

I guess that settles that. Thanks!

I never lived by it but from what I can remember there was something about system RAM caching/shadowing/mirroring VRAM? Just a vague memory but I see people discussing it once in a while.

 

[SlowSpyder]

I have 16GB of system ram and an 8GB card, it is a non-issue.

 

[Stuka87]

This is one of those rumors that has spread like wildfire on the 'net despite being completely wrong.

Higher VRAM actually leads to lower system RAM usage.

Sorry but this statement is wrong.

If you look at system requirements for video cards, they often say to have the same or more system memory than video card memory. This because of how caching works as the driver moves things in and out of vram.

980Ti requirements:
• PCI Express-compliant motherboard with one dual-width x16 graphics slot
• One 6-pin and one 8-pin PCI Express supplementary power connector
• Minimum 600 W or greater power supply
• 300 MB of available hard drive space
• 8 GB system memory (16 GB or higher recommended)
• Microsoft Windows 8 and 8.1, Windows 7 or Windows Vista

R9 Fury requirements:
• PCI Express-based PC with at least one ×16 lane graphics slot available on the motherboard. For dual-card installation, it is recommended that there be enough room on the motherboard to provide at least one empty slot between the two graphics cards.
• System power supply with two 8-pin PCIe® power connectors. For dual-card installation, a second pair of 8-pin power connectors is required. AMD does not recommend the use of power adapters or dongles. Visit http://www.amd.com/r9 for a list of compatible power supplies.
• System case with at least one location for a screw-mounted, 120 mm × 120 mm chassis fan. The case should also allow appropriate air flow and cooling of all system components. For dual-card installation, two mounting locations are required.
• A minimum of 8 GB of system memory. For dual-card installation, a minimum of 16 GB of system memory is recommended.
• Windows® 8.1, Windows® 7, or Linux® operating system (64-bit operating system is highly recommended).
• At least one display that is compatible with a DisplayPort or HDMI connection.

 

[ShintaiDK]

Higher VRAM actually leads to lower system RAM usage.

^^ This.

One of the examples was Rise of the Tomb raider tested with 4, 6 and 8GB cards. The game used roughly 14GB in a combination of VRAM and system memory with DX12.

The 4GB cards needed ~10GB system memory and the 8GB(Only 7GB could be used it seems) card needed ~7GB system memory. While 6GB cards needed 8GB system memory.

With DX11 its different depending on what the driver does.

 

[Magic Carpet]

Agree with the above posters ^.

16gb is the recommended amount of ram these days for modern games, if you don't want your pagefile to be working as hard. Personally, I am playing Tomb Raider with 32 gigs of ram with my swap file disabled*, and I have had no issues. Everything is damn fast, respawns are typically 3-5 seconds.

*Using HDD at the moment, improves performance a bit that way. Not recommended for everybody

EDIT: With 16gb video cards, I suspect, Nvidia will start recommending 32 gigs of RAM. Might as well max out your ram today. X99 owners rejoice ( 8 x 16 gb)

 

[Tuna-Fish]

This is one of those rumors that has spread like wildfire on the 'net despite being completely wrong.

That's because it used to be true, a long time ago. In some past DX version (8 or 7, not sure) when you context switched everything in VRAM got wiped, and because of this a copy of everything was also kept in main ram. This hasn't been true for a long time now.

 

[Magic Carpet]

That's because it used to be true, a long time ago. In some past DX version (8 or 7, not sure) when you context switched everything in VRAM got wiped, and because of this a copy of everything was also kept in main ram. This hasn't been true for a long time now.

They made some changes in WDDM 1.1.

Windows 7 saves one copy of the memory allocation per application window by getting rid of the system memory copy entirely. Thus, for a GDI application window visible on the desktop, the memory consumed is cut in half.

Not sure how that effected games, though. Probably, irrelevant in this context.

 

[Bacon1]

^^ This.

One of the examples was Rise of the Tomb raider tested with 4, 6 and 8GB cards. The game used roughly 14GB in a combination of VRAM and system memory with DX12.

The 4GB cards needed ~10GB system memory and the 8GB(Only 7GB could be used it seems) card needed ~7GB system memory. While 6GB cards needed 8GB system memory.

With DX11 its different depending on what the driver does.

Source?

DX12 memory usage seems broken though from my own testing, it just filled up RAM ASAP. Here is DX11 testing:

http://www.hardocp.com/article/2016/02/29/rise_tomb_raider_graphics_features_performance/13

 

[Termie]

All right, enough debate.

I'm going to record RAM and VRAM usage in three popular games on an R9 270X 2GB, R9 290 4GB, and R9 390X 8GB using 8GB and 16GB of system RAM and publish the results. Hope that will settle this... however it turns out. I'll throw in a GTX 980 4GB for good measure, in case it acts differently than the AMD cards.

Stay tuned. Any critiques of my test method, better speak up fast, because I'm already working on this.

 

[Magic Carpet]

Excellent news, Termie. Godspeed, can't wait for your results. I am equally as intrigued :thumbsup:

 

[tential]

All right, enough debate.

I'm going to record RAM and VRAM usage in three popular games on an R9 270X 2GB, R9 290 4GB, and R9 390X 8GB using 8GB and 16GB of system RAM and publish the results. Hope that will settle this... however it turns out. I'll throw in a GTX 980 4GB for good measure, in case it acts differently than the AMD cards.

Stay tuned. Any critiques of my test method, better speak up fast, because I'm already working on this.

Dude you've already tested this.
This is how I knew this information in the first place.

It's hilarious to see someone quote the box rather than listen to a person who actually tests cards.

Was going to come in and post what you said and quote your articles but well looks like you're here!

I think your tomb raider review shows this explicitly when you look at ran usage between the 980ti 970 and Titan x I believe. It always uses the same amount of total ram just different allocations of vram and system ram based on your pc configuration. That's just 1 game though.

I'm sure you know more though.

 

[ShintaiDK]

Source?

http://techbuyersguru.com/first-look-dx12-performance-rise-tomb-raider?page=1

 

[Termie]

Dude you've already tested this.
This is how I knew this information in the first place.

It's hilarious to see someone quote the box rather than listen to a person who actually tests cards.

Was going to come in and post what you said and quote your articles but well looks like you're here!

I think your tomb raider review shows this explicitly when you look at ran usage between the 980ti 970 and Titan x I believe. It always uses the same amount of total ram just different allocations of vram and system ram based on your pc configuration. That's just 1 game though.

I'm sure you know more though.

http://techbuyersguru.com/first-look-dx12-performance-rise-tomb-raider?page=1

I'm testing under DX11 this time to make sure that wasn't a DX12 artifact.

BF4, Witcher 3, RoTR.

Will take a while, lots of component swapping! And thanks, Magic Carpet, for the well wishes!

 

[tential]

I'm testing under DX11 this time to make sure that wasn't a DX12 artifact.

BF4, Witcher 3, RoTR.

Will take a while, lots of component swapping! And thanks, Magic Carpet, for the well wishes!

I see, well as always I look forward to more testing and benching. The more we know the better.

 

[CakeMonster]

Awesome! Thanks for doing this.

 

[SlowSpyder]

All right, enough debate.

I'm going to record RAM and VRAM usage in three popular games on an R9 270X 2GB, R9 290 4GB, and R9 390X 8GB using 8GB and 16GB of system RAM and publish the results. Hope that will settle this... however it turns out. I'll throw in a GTX 980 4GB for good measure, in case it acts differently than the AMD cards.

Stay tuned. Any critiques of my test method, better speak up fast, because I'm already working on this.

You should test with 4GB of system ram, too (if you have it).

 

[Termie]

You should test with 4GB of system ram, too (if you have it).

Well, I certainly have plenty of 4GB sticks, but that would add 50% to my testing time, and it's not a configuration a lot of people are running today. In fact, DDR4 doesn't even come in anything less than 4GB sticks, so to run with 4GB requires using single-channel mode, which has a detrimental impact on performance. Given that 4GB of RAM costs about $25, it's also a cheap upgrade.

That being said, I'm adding a 980 Ti to the mix, so the testing looks like this:

CPU: 6700K@4.4
RAM: 2x4GB and 2x8GB
Video Cards: R9 270X 2GB, R9 290 4GB, GTX 980 4GB, GTX 980 Ti 6GB, R9 390X 8GB,

 

[SlowSpyder]

No worries. I thought it might be interesting to see how a really low amount of ram (by today's standards) might affect performance. But that's not really what this thread was about anyway. Thanks for doing what you're doing!

 

[Adored]

4GB is really pushing it in some games now, stuff like Fallout 4 runs much worse even on an SSD. I got a bunch of people to test it and it's apparently not the single channel doing it (and single vs dual channel rarely shows any major impact in most games), it's just the amount that is the issue.

What's actually important is of course the impact on gameplay, not how much RAM is being "used". I think there are only a handful of people in the world who truly understand what's going on with Windows memory management tbh.

And obviously DX12 is being an issue right now as reported. I used your RoTTR benchmarks in a recent video actually Termie. I would be careful about making any definitive statements based on the early DX12 titles however as the devs will likely find ways to fix the issues, like Oxide apparently have.

 

[Bacon1]

http://slickdeals.net/f/8665078-10-...or-98-99-ac-more-free-shipping-newegg-com?v=1

8GB of ram for $27, go get some more lol

 

[escrow4]

Its surprising how much fatter games have gotten with new consoles. In 2012 you had a title that would barely use 2GB RAM and take up around 10GB HDD space. Now games take up 40-50GB and swallow 4GB RAM . . . . . for the same tiny 5hr single player campaign.