Double check my memory benchmarks?

P

pdp76

Weaksauce
Joined
Jan 31, 2006
Messages
113
I'm not 100% sure, but my memory benchmarks seems kind of slow, would you mind commenting on it real quick? The reason I ask is because my Windows Experience index is 5.9 for memory. Normally, I wouldn't be too concerned because I know you can't trust the WEI too much, but another computer with just about the exact same specs as me got a WEI index for 7.3, which I thought was a pretty big significance. The only thing different is that he was running Corsair and I'm running OCZ memory.

Here are my specs
- Core i3-530 (not OCed)
- GA-H55M-UD2H mobo
- OCZ Gold 4GB Dual Channel DDR3 running at 1333MHz
- Running Intel HD Graphics with 128MB shared memory

Here is what CPUZ reports for my memory
Type: DDR3
Channels#: Dual
Size: 4096MB
DC Mode: Symmetric
DRAM Frequency: 666.7MHz
FSB : DRAM: 4:20
CAS# Latency (CL): 8.0 clocks
RAS# to CAS# Dealy (tRCD): 7 clocks
RAS# Precharge (tRP): 7 clocks
Cycle Time (tRAS): 16 clocks
Row Refresh Cycle Time (tRFC): 60 clocks
Command Rate (CR): 1T

And here are the benchmark results from CrystalMark2004
Read : 8965.72 MB/s ( 8965)
Write : 7088.09 MB/s ( 7088)
Read/Write : 6457.53 MB/s ( 6457)
Cache : 65152.06 MB/s ( 6515)
 
Does your friend have the same amount running in dual channel at 1333MHz? At first glance the only thing I see the "might " chip into your benchmark is the sharing of the ram with the graphics.
 
Yep, he's also has 4GB and using the IGP. Another thing is, his mobo might be different I'm not sure. He's not my friend, he's just some random guy with a blog who pstehd his system specs.
 
I'd say go into your BIOS and tweak some settings until you find some your RAM/System like. I really don't know how Windows rates things. For example, my old GTX 280 got a 6.8, and my new 5870 gets a 6.0. As long as your system runs fine, I'd say not to worry about it too much.
 
How about the numbers from CrystaMark I posted? Do they look reasonable for ddr3 running at 1333mhz?
 
I personally haven't run that benchmark so I couldn't tell you.
 
Proxy said:
I personally haven't run that benchmark so I couldn't tell you.
Click to expand...

The numbers CrystalMark gives are just plain MB/s figures, it's not some cryptic number they come up with.

I just read that DDR3 1333 has a theoretical max bandwidth of 10666MB (666MHz * 2 * 64bits / 8bitperbyte). Factor in Dual Channel, so multiply that by 2 again and you get 21333MB/s as a theoretical max. That is WAY higher than any of my numbers I posted above. Even if you take 25% off the theoretical value for whatever overhead, you still have 16000MB/s. Did I do some calculations wrong?
 
drescherjm said:
Theoretical bandwidth is not normally achieved with ram. This type of calculation does not take into account latency and internal organization of cells which in the real world drastically reduce you from the theoretical bandwidth.


http://en.wikipedia.org/wiki/Memory_latency#Effect_on_memory_access_speed
Click to expand...
Yes I understand there are other factors that will cause real world bandwidth to be signifiantly less than theretical bandwidth. But assuming my theoretical calculations are correct (which I'm still not 100% sure of) my real world measurments are only around 1/3 of the theoretical bandwidth. Something seems wrong, unless someone else can point out other flaws in my reasoning. Maybe I'm just being paranoid :p
 
This depends on what the benchmark actually is testing. I mean sequential transfers where burst transfers can be used most of the time or a combination of sequential and random or just random (where you take the latency hit on most transfers).
 
That's a great find, thanks for much for putting up with my paranoia :) I really appreciate it! Maybe others who are as paranoid as me will find this thread someday.
 
You must log in or register to reply here.
Back
Top