• 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.

384bit vs 512bit memory bus Tomb Raider shootout

If your running Win 8 or 8.1 is a little tricky, it will start in a little box top left, click on it and then Alt + Return Twice to get it to normal full screen, do it as fast as you can. before the credits are over and the bench starts.
 
Just out of curiosity, earlier I ran 7 benches on Firestrike with just altering the memory bus. From the standard 1502Mhz (288.4GB/s) to +700 1852Mhz (355.6GB/s) I received a 3.8% increase in performance. Not very much.

What is the average 290X managing on bandwidth please?

http://www.3dmark.com/3dm/1835847

http://www.3dmark.com/3dm/1835942

http://www.3dmark.com/3dm/1835967

http://www.3dmark.com/3dm/1835986

http://www.3dmark.com/3dm/1835997

http://www.3dmark.com/3dm/1836047

http://www.3dmark.com/3dm/1836068

http://www.3dmark.com/3dm/1836086

all increments of +100
 
clearly you and Kaap must be wrong :D
AMD has 512-bit which is just better, just like my amp goes to 11 making it one louder than whatever you have :D

It is better and that shows itself at multi-monitor resolutions and 4k. It would seem from Kaap's testing it is not a big advantage at 1600p over the 384 bit bus. The r9 290x was marketed as a 4k card which is where amd must feel the 512 bit bus has it's biggest advantage. Nvidia must see it this way as well as with the 780ti they added faster memory to give it more bandwidth over the r9 290x.
 
Last edited:
I don't think it will be long before AMD or NVidia do it if they are serious about 4K.:D

MSI 290X with Samsung IC, I reckon could push 7500-8000Mhz if cherry picked.

All current 290's use Eplida and Hynix, everyone knows Samsung beats them both and is what is used on the EVGA Classy's. ;)

Stick some Samsung on an MSI Lightning 290X and things could be interesting.
 
Really Interesting reading Kaap, thank you ;)

If it's not too dumb a question... how can you tell what type of memory is being used on your gpu? Eplida, Hynix, or Samsung. Will Gpuz show this?
 
If it's not too dumb a question... how can you tell what type of memory is being used on your gpu? Eplida, Hynix, or Samsung. Will Gpuz show this?

There is a tool called HawaiiInfo.

I can't find it anywhere except OCN which requires reg, so I put it here:

http://www53.zippyshare.com/v/38595604/file.html

Also included is a less-useful program called MemoryInfo which doesn't work with multi-GPU.
 
Last edited:
Its actually more to do with what speed the Memory can do, on the 290/X the memory IC's are rated for 1500Mhz, they don't usually go much over that unless you over-volt them, espesialy if they are Elpida. clock them to much past 1500Mhz and the performance gain falls off a cliff.

You can check if memory is error free.
Almost 2% gain with 100MHz up.
Asus r9 290 with Elpida.

enny.jpg

o36c.jpg


edit
Here is with mem at 1625MHz.
After that I have no gains on AvP.

o883.jpg
 
Last edited:
It is better and that shows itself at multi-monitor resolutions and 4k. It would seem from Kaap's testing it is not a big advantage at 1600p over the 384 bit bus. The r9 290x was marketed as a 4k card which is where amd must feel the 512 bit bus has it's biggest advantage. Nvidia must see it this way as well as with the 780ti they added faster memory to give it more bandwidth over the r9 290x.

thing is, at 4K resolution you are talking about needing 3-4 cards of the current gen to get decent playable frame rates... in multi card setups each card has it's own discrete 384 or 512-bit bus, there are some textures and data that need to be shared but if you have plenty of VRAM available then this is minimised, so the memory bandwidth requirements for each card actually go down, not up when you go from single card 1440 to 3 card 4K

780ti's have a big chunk less bandwidth than a 290X out of the box (about 10% less at stock settings) but it doesn't seem to affect frame rates looking at 3 way 4K comparisons - BF4 is the only game that currently comes close to "favouring" the 290X and that is hardly surprising

most Titans and 780's will easily do 7000 on the memory in any case

a single card at 1440 is being pushed quite hard, as can be seen by the sub-30fps minimums and sub-50 averages, and yet in these conditions that memory bandwidth alone is making single digit or even fractions of an fps differences
 
Last edited:
thing is, at 4K resolution you are talking about needing 3-4 cards of the current gen to get decent playable frame rates... in multi card setups each card has it's own discrete 384 or 512-bit bus, there are some textures and data that need to be shared but if you have plenty of VRAM available then this is minimised, so the memory bandwidth requirements for each card actually go down, not up when you go from single card 1440 to 3 card 4K

780ti's have a big chunk less bandwidth than a 290X out of the box (about 10% less at stock settings) but it doesn't seem to affect frame rates looking at 3 way 4K comparisons - BF4 is the only game that currently comes close to "favouring" the 290X and that is hardly surprising

most Titans and 780's will easily do 7000 on the memory in any case

a single card at 1440 is being pushed quite hard, as can be seen by the sub-30fps minimums and sub-50 averages, and yet in these conditions that memory bandwidth alone is making single digit or even fractions of an fps differences

No a gtx780ti comes out of the box with 336 gb/s memory bandwidth compared to 320 gb/s on the r9 290x. You don't need more than a single card to play at 4k as has been shown but you need to compromise setting's. I don't think it works the way you are saying anyhow as at multi-monitor resolutions 7970's were stomping gtx680's even in sli and tri-fire and it was all down to the memory bus at the time.
 
No a gtx780ti comes out of the box with 336 gb/s memory bandwidth compared to 320 gb/s on the r9 290x. You don't need more than a single card to play at 4k as has been shown but you need to compromise setting's. I don't think it works the way you are saying anyhow as at multi-monitor resolutions 7970's were stomping gtx680's even in sli and tri-fire and it was all down to the memory bus at the time.

sorry, I'd miscalculated the stock 290x as 6000 instead of 5

7970's with later drivers were stomping 680's because they were much bigger GPU's, they should have always done so but the early drivers weren't up to much
 
sorry, I'd miscalculated the stock 290x as 6000 instead of 5

7970's with later drivers were stomping 680's because they were much bigger GPU's, they should have always done so but the early drivers weren't up to much

Even with the 780ti having more memory bandwidth out of the box the r9 290x is almost equal to the 780 ti at 4k where as at resolutions below this the ti is clearly faster. The only thing that could be helping the r9 290x is the wider memory bus.
 
Last edited:
Even with the 780ti having more memory bandwidth out of the box the r9 290x is almost equal to the 780 ti at 4k where as at resolutions below this the ti is clearly faster. The only thing that could be helping the r9 290x is the wider memory bus.

4K = huge pixel mass.

Pixel Fill Rate:

780Ti = 43.9
290X = 64.0
 
sorry, I'd miscalculated the stock 290x as 6000 instead of 5

7970's with later drivers were stomping 680's because they were much bigger GPU's, they should have always done so but the early drivers weren't up to much

In BF3, moving from 1920x1080 to 5760x1080: my SLI OC 680s went from 8% faster than CF OC 7950s to 27% slower. Only difference being the resolution. Wasn't VRAM limited on the 680s.

Was the memory bandwidth. Neither set up provided particularly pleasing frame rates on max settings but the 7950s got a lot closer.

But with regards to this bench and game done by Kaap we can see the bandwidth provided by the Titan is more than fine.
 
Back
Top Bottom