View Single Post
  #3  
Old 12-04-2010, 07:02 PM
atomt atomt is offline
Registered User
atomt's PC Specs
 
Join Date: Dec 2010
Posts: 2
POST ID # = 478995
atomt Reputation: 10
Confused

Quote:
Originally Posted by trackrat View Post
FWIW my testing shows almost no system performance gains with change in latency from 11-11-11-30 2T to 8-8-8-24 1T with DDR3 RAM. If your software can show some significant advantage great but the lower latencies on DDR3 seem to only show possible benefit in benchmark results.

The CMX12GX3M3A2000C9 should provide excellent performance and stability @ 1600 MHz. as long as your CPU is happy running RAM @ 1600 MHz. This will be a significant OC as the i7-920 officially only supports 1066 MHz. The increased frequency (50%), will show more system performance gain than lower latency IME.
Hey - I like my not very useful and absolutely non-realistic synthetic benchmarks thank you very much. Then again already at 1066Mhz 3 channels is more than bandwidth than a poor current generation quad core i7 is able to consume with most real world workloads.. So why bother with high bandwidth RAM at all right.

1600Mhz is fine on my D0 i7-920, its been running at that speed 24/7 for a year or so now. Only problem is that my data sets no longer fit in 6GB (3x2GB), and running more than one stick per channel at 1600Mhz quickly gets kinda shakey. The workload is not very sensitive to latency however (other than when benchmarking of course), but going significantly up in latency just rubs my inner hardware geek the wrong way. Going from 7 to 9 would add a <irony>whopping</irony> 2.5ns of latency at "1600"Mhz Even though the added capacity alone would speed up the workload significantly even with that much added latency my inner geek simply would never accept it.

Speaking strictly in nanosecond latency 8-8-8 is a less tight timing at 1600 (10ns) than 9-9-9 is at 2000 (9ns), so it should probably work. In (over simplified) theory.. 7 could work, perhaps, but thats 8.75ns.. At 1540 possibly.. Hrms, a proper review would put those theories to the test ;)

Anyway, the price is good. The upgrade would be very much worthwhile even with cas 9 @ 1600. Not sure why I'm holding back really.
Reply With Quote