![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
Memory Overclocking & high vram games
- Thread starter jamesgalb
- Start date
Armenius
Extremely [H]
- Joined
- Jan 28, 2014
- Messages
- 42,694
Increasing the memory speed increases available bandwidth, so it would generally help in situations with large textures or antialiasing. The benefit would be less apparent where the bus width is narrower like on the 128-bit GTX 960, and vice-versa like on the 512-bit 290X. Total amount of memory has no bearing on bandwidth, but of course it does affect how fast the framebuffer can be pushed through.
For a GTX 960 with reference clocks:
1750 MHz * 4 bits per clock cycle * 128 bits / 8 bits per byte = 112 GB/s, or one full framebuffer of 4GB 28 times per second
For a R9 290X with reference clocks:
1250 MHz * 4 bits per clock cycle * 512 bits / 8 bits per byte = 320 GB/s, or one full framebuffer of 8GB 40 times per second
A typical memory overclock for the 960 is +250 MHz. Over the 128-bit bus this is an extra 16 GB/s of bandwidth, or an increase of 14.3% over the reference 1750 MHz. This means an extra four 4GB framebuffers every second.
A typical memory overclock for an after market 290X is also +250 MHz. Over the 512-bit bus this is an extra 64 GB/s of bandwidth, or an increase of 20% over the reference 1250 MHz. This means an extra eight 8GB framebuffers per second.
Theoretically the above could translate directly to frames per second. So yes, to your question: both of these cards would make better use of their memory if it were clocked faster. How much better is largely dependent on the bus width. This is why people often say a card like the GTX 960 with its narrow 128-bit bus and 4GB of VRAM is largely pointless.
For a GTX 960 with reference clocks:
1750 MHz * 4 bits per clock cycle * 128 bits / 8 bits per byte = 112 GB/s, or one full framebuffer of 4GB 28 times per second
For a R9 290X with reference clocks:
1250 MHz * 4 bits per clock cycle * 512 bits / 8 bits per byte = 320 GB/s, or one full framebuffer of 8GB 40 times per second
A typical memory overclock for the 960 is +250 MHz. Over the 128-bit bus this is an extra 16 GB/s of bandwidth, or an increase of 14.3% over the reference 1750 MHz. This means an extra four 4GB framebuffers every second.
A typical memory overclock for an after market 290X is also +250 MHz. Over the 512-bit bus this is an extra 64 GB/s of bandwidth, or an increase of 20% over the reference 1250 MHz. This means an extra eight 8GB framebuffers per second.
Theoretically the above could translate directly to frames per second. So yes, to your question: both of these cards would make better use of their memory if it were clocked faster. How much better is largely dependent on the bus width. This is why people often say a card like the GTX 960 with its narrow 128-bit bus and 4GB of VRAM is largely pointless.
This is why I was asking
Wondering if a 4GB GTX 960 would make more sense with a good memory overclock.Upon looking at the 4GB selections for the GTX 960s, they all have the same small memory overclock, going up 10MHz to 7010MHz... While they range from 100-200MHz overclock on the GPU...
I'm very surprised the marketing departments would miss a chance at having 'superclocked memory' if they are specifically using the extra 2gb to sell the card, never mind that it would seemingly make sense on the engineering side.
I'd be interested in some 960 4GB reviews where they memory speeds were played with.
Araxie
Supreme [H]ardness
- Joined
- Feb 11, 2013
- Messages
- 6,463
the GTX 960 4GB for that price its a worthless card... with a bit more you can find a much better GTX 970 which crush the 960.. the extra 2GB will not make the 960 any faster, it would only help in memory limited scenarios to avoid stuttering.. anyway Brent said the other day he will be thinking in make a review with a 4GB 960..
There is a $100 difference in the cards, and as games catch up to at least the consoles they will regularly use 4GB of vram and avoiding stutter is always desired, something the 970 happens to have its own issues with... the 960 also has HEVC support, which adds to its value as a hybrid htpc/1080gaming build.
If he does the review, ask him to play with memory clock speeds. Only makes sense!
more expensive ~$50-$80, uses more power, runs hotter, no HEVC, typically larger... Not the ideal small HTPC fit.
But I am also interested in more r9 290x 8GB reviews in high vram games, especially with memory overclocking (interested for the main rig)... I think that is a huge sleeper value with DX12 and 4k/VR incoming during an era where 720/1080 console ports will already use 4GB of vram.
Last edited:
defaultluser
[H]F Junkie
- Joined
- Jan 14, 2006
- Messages
- 14,398
If you're curious about the benefits of 4GB to the GTX 960, GURU3D did a test of the 4GB card last week. In all their tests the average framerate was unchanged between the two, even running at 4k resolution (which was unplayable).
They also did 3 game runs to analyze the potential stutter difference between the two cards. Of the three games they tested at 1440p, only one game showed any signs of stutter:
http://www.guru3d.com/articles_pages/gigabyte_geforce_gtx_960_g1_gaming_4gb_review,13.html
And 30fps is really not what I would call playable. I'm really starting to think these 4GB cards serve no purpose, and the GTX 960 2GB really is a truly good value card.
As for memory overclocking effects, I thought that most of these cards already shipped near their limits on memory clocks? If there's only a few hundred Mhz room left, you''re not gong to see massive performance improvements pushing the memory to the very limits.
They also did 3 game runs to analyze the potential stutter difference between the two cards. Of the three games they tested at 1440p, only one game showed any signs of stutter:
http://www.guru3d.com/articles_pages/gigabyte_geforce_gtx_960_g1_gaming_4gb_review,13.html
And 30fps is really not what I would call playable. I'm really starting to think these 4GB cards serve no purpose, and the GTX 960 2GB really is a truly good value card.
As for memory overclocking effects, I thought that most of these cards already shipped near their limits on memory clocks? If there's only a few hundred Mhz room left, you''re not gong to see massive performance improvements pushing the memory to the very limits.
Last edited:
The test they did was at Gigabyte's stock memory overclock, which for this card was a stock 7000MHz memory (or 7010 after other tweaks, I think).
http://www.guru3d.com/articles_pages/gigabyte_geforce_gtx_960_g1_gaming_4gb_review,25.html
Guru3D seems to get the same obtainable memory overlock that techreport and others get, around 8000MHz or a 1000MHz boost (14% increase)
http://techreport.com/review/27806/five-geforce-gtx-960-cards-overclocked/2
Would tuning down the GPU overclock allow for even more memory overclocking?
CaptNumbNutz
Fully [H]
- Joined
- Apr 11, 2007
- Messages
- 25,080
I regularly use more than 2GB of VRAM on my GTX760 4GB card while playing BF4 on a mix of high and ultra settings. Granted, the most I've ever seen used at one time was 2.3GB (1.8-1.9 was just BF4), but that was more than 2GB regardless.
The primary reason I've seen the usage go above 2GB was because I have 2 monitors. I run my games on my primary 1080p monitor, while I keep Teamspeak and EVGA Precision X open on my second monitor. I've tried using only one monitor and the usage hovered damn close to 2GB but did not go over. With the 2nd monitor it used another couple hundred megabytes.
Maybe there was a driver issue, maybe it was the game, maybe its just Windows stretching its legs unnecessarily, but I am thankful my card has 4GB just for a little breathing room. With my setup I hold a consistent 80+ fps and never see massive frame rate drops.
Single monitor and 1080p, 2GB is barely cutting it for mid level cards like the 760 and 960. The real issue here is the pricing structure. When I bought my 760 4GB at Microcenter, it was $300 which also got me the FTW version with the even higher overclock. The 760 2GB was $250 and the 770 was nearly $400 making the 760 4GB card a clear choice.
durquavian
Gawd
- Joined
- Dec 13, 2014
- Messages
- 757
But wouldn't the driver dictate what goes in or rather gets stored in Vram? So 2Gb may be all that is required and 4Gb only allows more to be loaded and stored, which might help minimums. However many tests show that 4Gb in these small bit buffer cards adds no extra performance. I guess any improvement would only be within error % so likely happening oonly at best <5% of the time.