• 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.

Memory Overclocking Gains On Hawaii With A 512 Bit Bus

LtMatt

LtMatt

LtMatt

Caporegime
Joined
12 Jul 2007
Posts
43,698
Location
United Kingdom
System Specs

[email protected]
16GB DDR3 2133Mhz
2x290 Pro Crossfire
Win 8.1 64Bit
1440P Resolution

Games - Ultra-Highest Preset & AA/SSAA Settings Used
Dirt Showdown - MSAA x8
Metro2033 - MSAA x4
Tomb Raider - SSAA x2

I ran three gaming benchmarks to see what performance boost a 30% overclock on the memory would give a couple of AMD 290 Hawaii cards. Is it possible to increase the performance of the 512 bit bus over stock, or is it already delivering more than enough performance? At stock the bus delivers 320gb of bandwidth. Overclocked it delivers 416gb of bandwidth.



1250mhz Testing - Settings

CNRZD4t.jpg



Results

4AMwjpW.jpg



3ta7L7a.jpg



o368TkY.jpg



1625mhz Testing - Settings

q0XnI0j.jpg



Results

W1HsTCU.jpg



UHMuXw6.jpg



ooRRWWK.jpg



Conclusion

Dirt Showdown Minimum: 0.6% increase from overclocking the memory
Dirt Showdown Average: 2% increase from overclocking the memory

Metro2033 Minimum: 10% increase from overclocking the memory
Metro2033 Average: 2.3% increase from overclocking the memory

Tomb Raider Minimum: No improvement 0%.
Tomb Raider Average: 0.8% increase from overclocking the memory.


Overclocking the memory has very little effect on Hawaii as not many games are capable of saturating the bandwidth that the memory and 512 bit bus offers, at least at 1080P/1440P. I would expect memory overclocking to have more of an effect at Eyefinity or 4k. Outside of synthetic benchmarks overclocking the memory is almost pointless for Hawaii users who should instead stick to overclocking the core for the biggest gains. According to my testing memory stability and overclock potential is strongly tied to core voltage with Hawaii. As you can see overclocking the memory to 1625 required me to increase the voltage from -0.050mv to +0.100mv to ensure stability and no black screens. I can probably lower the voltage a tad, around +0.075v or slightly higher should be enough for me if i wish to keep that ridiculously high memory overclock/bus bandwidth.

I might add/test some more games in due course. I also welcome testing from other Hawaii 290 users to see if they find something similar.
 
Last edited:
Hi there

Like you say to use that bandwidth you need to conduct testing on eyefinity or 4k resolution ideally to see if the benenfits are any better.

Good testing though.

The only reason we overclock Hawaii on memory is it gives improvements in benchmarking and there might be some game titles which tax the memory more.
 
Gibbo said:
Hi there

Like you say to use that bandwidth you need to conduct testing on eyefinity or 4k resolution ideally to see if the benenfits are any better.

Good testing though.

The only reason we overclock Hawaii on memory is it gives improvements in benchmarking and there might be some game titles which tax the memory more.
Click to expand...

Yes i would expect the results to be different and for there to be bigger gains from memory overclocking at Eyefinity/4k. Would be great if someone with that setup could confirm and contribute some results.
 
Last edited:
I noticed this when testing my clock speeds on BF4 when I first got the 290x but I didnt know enough abourt the technical details to pin it down on the memory bus size. I did drop my memory clocks and only when for it on the core after that.

I use a catleap B2 monitor, 1440p @ 120hz on ultra doesnt seems to bother the big memory bus!

Now i understand why Nvidia 600 series get big performance gains when clocking up the memory because of the smaller mem bus size.
 
Thanks for this Matt, very interesting indeed. A 30% increase on the memory I would consider to be quite a considerable overclock however it seems to show little improvement on FPS which doesn't justify it. Although as you've mentioned and so has Gibbo it probably will have more of an improvement on larger resolutions such as 1600p and above so it would be justified here but this is 2 card territory anyways lol.
So when I get a 290 pro ill probably pass on overclocking the memory and just push the core now from seeing this as I only game on a 1080p monitor at the moment so thanks again :)
 
r7slayer said:
Thanks for this Matt, very interesting indeed. A 30% increase on the memory I would consider to be quite a considerable overclock however it seems to show little improvement on FPS which doesn't justify it. Although as you've mentioned and so has Gibbo it probably will have more of an improvement on larger resolutions such as 1600p and above so it would be justified here but this is 2 card territory anyways lol.
So when I get a 290 pro ill probably pass on overclocking the memory and just push the core now from seeing this as I only game on a 1080p monitor at the moment so thanks again :)
Click to expand...

1600P is only slightly more than 1440P (kinda like 1080P vs 1200P) so i believe you would need Eyefinity/Surround - Multi Monitor for memory overclocking to play any significant part on performance improvement on Hawaii with its 512 bit bus.
 
Good testing Matt. However remember memory clocks are what fill the BUS. If the memory isn't fast enough it's unlikely to saturate the bus, it's not all down to the game. You need at least 2Ghz on the memory to make real use of the 512bit bus, but there is no real way to prove that yet.
 
Hmm

Well i did those sort of tests 2 months ago and my results are in 290x thread.
Where i shown that passing 1500mhz on memory got NO POINT at all. Tests ware in 3D so card renders 2x 1080p scenes at once. And where with overclocked 7950 any mhz on memory gave higher scores 290x got more than enough to handle 3D gaming :)



zpaf
do 1500 vs 1625 and we will see then :)
 
zpaf said:
Nice thread LtMatt.

Depend on game engine to see gains.
Here is my lovely game AvP ...


Click to expand...

Thanks for the contribution zpaf, a 7.7% increase there on the average fps

I wonder if you could test some more games as you're using a resolution of 5760x1080 and will no doubt see more gains from memory overclocking than your typical 290 user.
 
Lord_ZED said:
zpaf
do 1500 vs 1625 and we will see then :)
Click to expand...

Here is without any touch on voltage.



LtMatt said:
Thanks for the contribution zpaf, a 7.7% increase there on the average fps

I wonder if you could test some more games as you're using a resolution of 5760x1080 and will no doubt see more gains from memory overclocking than your typical 290 user.
Click to expand...

Another test with Grid II 8xmsaa ultra.
1200/1250 vs 1200/1625


Here we have big gains on minimum.

I don't want to make this thread a gpu war :D but ...
Very important to tell you that friend of mine with gtx780ti@1200/1925 cannot touch these minimum on specific game.
 
LtMatt said:
That's some nice gains. Guess the 256bit bus is limiting you a bit there. I will run some AVP tests myself as it seems to benefit from overclocking the memory somewhat.
Click to expand...

If it had a 384Bit bus it would be with in a hairs breath of a 7950 clock for clock, the SP count is the same difference from the 7950 to the 7970. as it is its probably about 10 to 15% slower.

Its a pretty quick GPU for £176 bought 8 months ago.
 
humbug said:
If it had a 384Bit bus it would be with in a hairs breath of a 7950 clock for clock, the SP count is the same difference from the 7950 to the 7970. as it is its probably about 10 to 15% slower.

Its a pretty quick GPU for £176 bought 8 months ago.
Click to expand...

Got a link to the benchmark? The one i just downloaded didn't have any options it just ran the benchmark and was using low res and low details.
 
You must log in or register to reply here.
Back
Top Bottom