Overclock.net - An Overclocking Community - View Single Post - RAMDisks Roundup and testing

View Single Post
post #8 of (permalink) Old 04-14-2013, 06:25 AM - Thread Starter
Brokenstorm
New to Overclock.net
 
Brokenstorm's Avatar
 
Join Date: Apr 2012
Posts: 216
Rep: 42 (Unique: 26)
I'll only include the results from Anvil, ATTO and CrystalDisk from now on since I think they are the most representative. FDBench hasn't been updated in over 5 years, IOMeter seems like a dead project and PCMark 7 takes forever to run. I'll still mention the other if they had interesting results.
I also reduced the size of the RAMDisk to 4.125GB so that it would be the same for all the tests.


Test 1: Results are nearly identical.


Test 2: The maximum my BIOS would accept was 11-10-10-31-2N, so I went with that.
Latency went up but results stayed very similar.


Test 3: There was slightly more variations in those results, but they were still very much in line with the previous ones if only slightly slower.
Even in these less than ideal conditions memory doesn't appear to be the bottleneck. Testing with SiSoft Sandra, I still had 10GB/s of memory bandwidth. So it seems unlikely that memory bandwidth will ever but an issue.


Test 4: Latency went down ever so slightly but results stayed the same across the board.


Test 5: Well that didn't go has expected. I wasn't able to overclock my RAM, one of the stick kept dropping out. I might try again later with another set but for now I'll have to try to extrapolate the result of the next test to try to get an idea of the effect of higher memory frequency.

If anyone with 2400MHz+ memory would be willing to run this test I would be grateful.


Test 6: Memory was set at 5-5-5-18 @ 800MHz.
Performance went down across the board by around 5% but without the results from higher clocked memory it's hard to draw any conclusion from those results.


Since my 920 has a locked multiplier I wasn't able to overclock my CPU without also overclocking my RAM. As a result I was not able to test for a memory bottleneck.
But given the extreme effort one would have to make to cripple his machine in order to reach a memory bottleneck, I think it's safe to say that nobody needs to worry about it.


Test 7: Other than the CPU utilization predictably rising to 25%, results were nearly identical. Only FDBench gave me unexpected result with 20% faster copy. Hard to tell if this is due to a flaw in the program or something else.


Test 8: For this test I only ran Anvil, ATTO and CrystalDisk as the other either take too long to run or don't give any additional information.
NTFS seems to be slower than both FAT32 and exFAT when the queue depth gets above 4. Other than that the result where similar.

NTFS:


FAT32:


exFAT:


Test 9: This RAMDisk doesn't like NTFS compression at all. In both Anvil and ATTO reads were slashed in half and writes went down by an order of magnitude. Even with compressible data the result were still pretty bad. Only Crystal disk gave me results similar to the previous ones.


Test 10: I didn't have time to do tests with different OS, so instead I did some with my CPU at 3.8GHz.




So now that we can prove that processor speed is the main factor that determines a RAMDisk performance, I think I'm ready to start testing each program.
I'll probably have my CPU a 3.6GHz for these so that the differences in performance (if any) are more noticeable.

I still have the determine what settings I should use (RAMDisk size, file system, etc) for the tests to be fair.
So if you have any ideas about that or if you want me to test other things such as the boot speed when loading an image let me know.

Brokenstorm is offline