Sign in with
Sign up | Sign in
Your question

I7 920 to i7 980x upgrade with mem bandwidth issue

Last response: in CPUs
Share
December 28, 2010 1:26:19 PM

Hello,
I recently upgraded my 920 to 980x and memory bandwidth decreased causing drop in graphics performance. All settings was kept the same from my 920. The 920 was clocked @ 3.6ghz and the same goes for the 980x that is clocked also now @ 3.6ghz. Bios was flashed to newest on my Asus Rampage Extreme 2 board. Question is, is that normal? What i have noticed is that my copy bandwidth and latency decreased but the rest was indeed faster. In cpu benches i got total of 50% performance increase except for the mentioned above. Weird ?!? is the bandwidth the culprit for getting lower frames for graphics benches? My Asus 5870 has catalyst 10.12 installed. Any advice would be appreciated-thanx in advance.
a c 131 à CPUs
December 29, 2010 12:45:00 AM

The i7 920 has a lower stock clock. So when you overclocked, you overclocked your ram more to achieve the 920 overclock vs when you achieved the 980X overclock. Try reducing your CPU multiplier on the 980X and increasing your FSB to the same speeds as it was for the 920.

As for the cause of lower framerate, I don't understand how that could be the issue.
a b à CPUs
December 29, 2010 1:52:19 AM

What power supply do you have? Might not be enough power??

Dunno, stumped.
Related resources
a b à CPUs
December 29, 2010 6:37:37 AM

Your gaming performance should not have changed. How much did the copy bandwidth and latency change (oh, and if the latency decreased, as stated in your original post, it's better)? If it's just a small change, I wouldn't worry about it - realistically, the memory performance shouldn't be much different from what it was on the 920, although if your uncore was faster on the 920, that could make a bit of a difference.

Regardless, I wouldn't expect it to affect your graphics performance much (either positively or negatively). What kind of a drop did you see?
December 29, 2010 11:01:34 AM

I have corsair dominator ram which is clocked at 1600mhz 8-8-8-24 1.65v. As mentioned settings is the same from the previous cpu. The only thing i have changed was the cpu voltage. 920 was 1.2v and changed to 1.24v for the 980x that is stock. I don't volt mod, i rather see what the components can achieve by doing so and then take a few steps back for stability. I have all settings on manual so that the board doesn't auto select too high voltages which is unnecessary. I didn't push the 980x yet to see what it really can do but one thing for sure is that the 920 did 4ghz with no volt modding.

CPU Ratio 18
BCLK 200
CPU 3600Mhz & Dram 1603Mhz
O and i am using a Antec Signature 850w psu which is more than capable. The 980x uses less power too. About 20w less on average according to my zalman zm-mfc2 controller.

On everest the mem scores are as follow

920 980x
mem read
18141 18401
mem write
14922 15618
mem copy
20087 17600
mem latency
47.6 50.4

I just have a hunch that the weaker mem copy and latency can have a small role in reduced graphics score on vantage. All though the cpu was 50% faster in vantage the gpu score was still less.

Vantage performance default
920@3.6 vs 980x@3.6
22057 33257
gpu
18593 17800

the small reduction in gpu score caused an average of 2 fps
Ok and i have not yet tested other benches that can justify changes

I think i will counter the changes with better memory like 2000mhz 8-8-8-24 clocked to 1800mhz 7-7-7-20 or something like that i suppose.

I have been doing a lot of researching on this issue and it seems to be evident with no real good explanation and settings/tweaks that can counter this issue.
December 29, 2010 11:11:14 AM

2 fps? That's a small enough difference to be thrown out in most benchmark comparisons...

You shouldn't have upgraded the proc for gaming performance... You should've upgraded your GPU...

Once you get into i7 territory there is absolutely no real-world difference in gaming performance between the different models. The only reason I'd recommend someone upgrade from a 920 to 980X is for rendering and folding performance, nothing more. Games simply do not have the computing requirements to take full advantage of these processors, not even half, not even a third. If you'd noticed when you were playing games, your 920 should have been sitting somewhere between 10% and 20% usage in most modern games. Your GPU, however, is probably a completely different story.

Take the 980X back and get a pair of 580s.
December 29, 2010 12:02:15 PM

Haha i will probably upgrade my gpu later. I do a bit of video editing- vegas studio 10 & CS4. I just couldn't understand how the 980x could have a negative influence on the above mentioned.I got the 980x at a good price and seemed worthy to max out my board with it. In cpu threaded applications the 980x is a monster.
December 29, 2010 12:07:06 PM

sonicmarx said:
Haha i will probably upgrade my gpu later. I do a bit of video editing- vegas studio 10 & CS4. I just couldn't understand how the 980x could have a negative influence on the above mentioned.I got the 980x at a good price and seemed worthy to max out my board with it. In cpu threaded applications the 980x is a monster.


Well if you're using it for media production then I'd say it's alright then. BUT, the 2 fps difference that you're seeing is not anything to be alarming. That, again, is a small enough difference to be attributed to background processes, etc. If it were, say, 5-10+, I'd say the proc is doing something dirty... But 2 is not enough difference to indicate problems IMO.
a b à CPUs
December 29, 2010 9:57:24 PM

Honestly, I wouldn't worry about it too much. As stated above, it's not too surprising that it didn't change things very much, and IMHO, buying some DDR3-2000 or 1800 would just be a waste of money. Your RAM is already more than fast enough so that it won't be the bottleneck for pretty much anything.
a b à CPUs
December 30, 2010 12:05:42 AM

Psychoteddy said:
Well if you're using it for media production then I'd say it's alright then. BUT, the 2 fps difference that you're seeing is not anything to be alarming. That, again, is a small enough difference to be attributed to background processes, etc. If it were, say, 5-10+, I'd say the proc is doing something dirty... But 2 is not enough difference to indicate problems IMO.


Well you should always leave a 3-5% error margin when benchmarking.
!