Quad Channel RAM Disk Benchmark with Crstal Diskmark

bigdogchris · Nov 24, 2014

Does anyone have a 4 dimm Quad channel system that has a RAM disk you could benchmark with Crystal Diskmark? 50MB is fine.

If you could copy and paste that would be great (to see IOPS).

the snake · Nov 25, 2014

ram @ 1600 cas 9/9/9/24

-----------------------------------------------------------------------
CrystalDiskMark 3.0.3 x64 (C) 2007-2013 hiyohiyo
Crystal Dew World : http://crystalmark.info/
-----------------------------------------------------------------------
* MB/s = 1,000,000 byte/s [SATA/300 = 300,000,000 byte/s]

Sequential Read : 5239.387 MB/s
Sequential Write : 7179.686 MB/s
Random Read 512KB : 4628.692 MB/s
Random Write 512KB : 6124.843 MB/s
Random Read 4KB (QD=1) : 225.328 MB/s [ 55011.7 IOPS]
Random Write 4KB (QD=1) : 208.201 MB/s [ 50830.4 IOPS]
Random Read 4KB (QD=32) : 391.163 MB/s [ 95498.8 IOPS]
Random Write 4KB (QD=32) : 352.906 MB/s [ 86158.7 IOPS]

Test : 1000 MB [R: 42.6% (3484.5/8189.0 MB)] (x2)
Date : 2014/11/25 6:18:45
OS : Windows 7 Professional SP1 [6.1 Build 7601] (x64)

the snake · Nov 25, 2014

for some reason it will not do a read on 512B

Filter · Dec 2, 2014

disregard.

753951 · Dec 2, 2014

the snake said:
for some reason it will not do a read on 512B

What RAM drive software is that?

the snake · Dec 3, 2014

753951 said:
What RAM drive software is that?

http://memory.dataram.com/products-and-services/software/ramdisk

qkslvr221 · Dec 3, 2014

Sure.

PrimoCache, 850s RAID 0, 5820k @ 4.7, Corsair DDR4 2666 @ 14-14-14-42

bigdogchris · Dec 3, 2014

Considering quad channel should have at least double the bandwidth of dual channel, I'm surprised the sequential speeds of a QC ramdisk are the same as a DC ramdisk. Unless there are some file system limitations in place.

753951 · Dec 3, 2014

About 2 years ago, when dataram RAM disk software was still free I did some comparison testing and RAM disk from WinRamTech (formerly QSoft) was seriously trumping dataram's implementation. To the tune of some 30%.
Granted, time has passed and things may be reversed now, but at least, it's something to think about.

bigdogchris · Dec 4, 2014

DDR2 Dual Channel

Why is quad channel so much slower?

-----------------------------------------------------------------------
CrystalDiskMark 3.0.3 x64 (C) 2007-2013 hiyohiyo
Crystal Dew World : http://crystalmark.info/
-----------------------------------------------------------------------
* MB/s = 1,000,000 byte/s [SATA/300 = 300,000,000 byte/s]

Sequential Read : 8536.339 MB/s
Sequential Write : 9575.242 MB/s
Random Read 512KB : 9605.250 MB/s
Random Write 512KB : 10803.722 MB/s
Random Read 4KB (QD=1) : 1526.043 MB/s [372569.1 IOPS]
Random Write 4KB (QD=1) : 1261.915 MB/s [308084.8 IOPS]
Random Read 4KB (QD=32) : 1389.996 MB/s [339354.6 IOPS]
Random Write 4KB (QD=32) : 1198.440 MB/s [292587.9 IOPS]

Test : 1000 MB [R: 1.1% (45.2/4088.0 MB)] (x5)
Date : 2014/12/04 1:03:42
OS : Windows 8.1 Pro [6.3 Build 9600] (x64)

753951 · Dec 5, 2014

I mentioned in previous post that not all RAM disk drivers are same. Since I'm in the process of building new PC and doing all sorts of testing I decided to compare, again, DataRam and WinRamTech RAM disks.

i7-5820K @ 4GHz, G.Skill 2133 DDR4

First DataRam

Code:

-----------------------------------------------------------------------
CrystalDiskMark 3.0.3 x64 (C) 2007-2013 hiyohiyo
                           Crystal Dew World : [URL]http://crystalmark.info/[/URL]
-----------------------------------------------------------------------
* MB/s = 1,000,000 byte/s [SATA/300 = 300,000,000 byte/s]
           Sequential Read :  7016.748 MB/s
          Sequential Write : 10852.684 MB/s
         Random Read 512KB :  6860.988 MB/s
        Random Write 512KB : 10310.968 MB/s
    Random Read 4KB (QD=1) :  1090.375 MB/s [266204.8 IOPS]
   Random Write 4KB (QD=1) :   942.783 MB/s [230171.7 IOPS]
   Random Read 4KB (QD=32) :  1099.550 MB/s [268444.9 IOPS]
  Random Write 4KB (QD=32) :   948.875 MB/s [231658.9 IOPS]
   Test : 1000 MB [D: 0.0% (0.0/4084.0 MB)] (x5)
  Date : 2014/12/06 20:41:16
    OS : Windows 8.1 Enterprise [6.3 Build 9600] (x64)

And now WinRamTech

Code:

-----------------------------------------------------------------------
CrystalDiskMark 3.0.3 x64 (C) 2007-2013 hiyohiyo
                           Crystal Dew World : [URL]http://crystalmark.info/[/URL]
-----------------------------------------------------------------------
* MB/s = 1,000,000 byte/s [SATA/300 = 300,000,000 byte/s]
           Sequential Read :  7407.388 MB/s
          Sequential Write : 11478.855 MB/s
         Random Read 512KB :  7218.039 MB/s
        Random Write 512KB : 11622.896 MB/s
    Random Read 4KB (QD=1) :  2270.713 MB/s [554373.3 IOPS]
   Random Write 4KB (QD=1) :  1918.684 MB/s [468428.6 IOPS]
   Random Read 4KB (QD=32) :  2172.251 MB/s [530334.8 IOPS]
  Random Write 4KB (QD=32) :  1901.191 MB/s [464157.9 IOPS]
   Test : 1000 MB [D: 0.0% (0.0/4064.0 MB)] (x5)
  Date : 2014/12/06 20:58:02
    OS : Windows 8.1 Enterprise [6.3 Build 9600] (x64)

That's with RAM disk of 4GB in size. As they grow bigger, DataRam lags behind even more.

qkslvr221 · Dec 6, 2014

I'm running the bench on a 540GB RAID volume. PrimoCache is a sort of midway solution that automatically RAMdisks frequently accessed blocks, and I far prefer it for daily use to regular RAMdisks. I went ahead and ran some benches on a 4GB drive for comparisons sake.

4GB Primo Ramdisk Direct/IO

4GB WinRamTech NTFS

Code:

-----------------------------------------------------------------------
CrystalDiskMark 3.0.3 x64 (C) 2007-2013 hiyohiyo
                           Crystal Dew World : http://crystalmark.info/
-----------------------------------------------------------------------
* MB/s = 1,000,000 byte/s [SATA/300 = 300,000,000 byte/s]

           Sequential Read : 10431.862 MB/s
          Sequential Write : 12466.557 MB/s
         Random Read 512KB : 10039.520 MB/s
        Random Write 512KB : 11986.221 MB/s
    Random Read 4KB (QD=1) :  1791.995 MB/s [437498.7 IOPS]
   Random Write 4KB (QD=1) :  1310.720 MB/s [320000.0 IOPS]
   Random Read 4KB (QD=32) :  1695.725 MB/s [413995.3 IOPS]
  Random Write 4KB (QD=32) :  1287.518 MB/s [314335.3 IOPS]

  Test : 1000 MB [Z: 0.6% (24.8/4096.0 MB)] (x5)
  Date : 2014/12/06 0:54:35
    OS : Windows 8.1  [6.3 Build 9600] (x64)

Difference isn't that huge, but for raw speed the pure RAMdisks are definitely faster. This WinRamTech is pretty well-made and compact software, too.

Tsumi · Dec 6, 2014

This is what I got on my system.

i7 3820 @ 4.2 ghz, 2gb RAMDisk using MSI Control Center, RAM is 1333 mhz CAS 7-7-7-21. Windows 8.1.

I find it interesting compared to some of the results above, my RAMDisk has slower sequentials and 512K, but 4K and 4K QD32 are faster. Perhaps lower CAS results in the higher 4K performance, while raw mhz improve sequentials.

qkslvr221 · Dec 6, 2014

Tsumi said:
I find it interesting compared to some of the results above, my RAMDisk has slower sequentials and 512K, but 4K and 4K QD32 are faster. Perhaps lower CAS results in the higher 4K performance, while raw mhz improve sequentials.

I suspected the same thing with 753951 's results, but don't know what timings they're using. Haven't touched the voltage on my set yet, but I've been reading about some folks getting down to CL12 @ 1.4v. Gonna see if I can get mine there later and run the same benches.

753951 · Dec 6, 2014

Looks like it. Hard to find DDR4 under CAS 15, and that's what I have.

qkslvr221 · Dec 6, 2014

753951 said:
Looks like it. Hard to find DDR4 under CAS 15, and that's what I have.

Well this is odd, mine are already at 14

Still going to try getting them lower, but there must be something else at play here too

qkslvr221 · Dec 6, 2014

Well this definitely confirms CL as one factor in 4K RAMdisk bottlenecks, best I was able to get at 2666 is 13-13-13-39 @ 1T. Running 1.35v

Code:

-----------------------------------------------------------------------
CrystalDiskMark 3.0.3 x64 (C) 2007-2013 hiyohiyo
                           Crystal Dew World : http://crystalmark.info/
-----------------------------------------------------------------------
* MB/s = 1,000,000 byte/s [SATA/300 = 300,000,000 byte/s]

           Sequential Read : 10489.256 MB/s
          Sequential Write : 12242.871 MB/s
         Random Read 512KB : 10103.094 MB/s
        Random Write 512KB : 11920.009 MB/s
    Random Read 4KB (QD=1) :  1813.953 MB/s [442859.5 IOPS]
   Random Write 4KB (QD=1) :  1356.837 MB/s [331259.0 IOPS]
   Random Read 4KB (QD=32) :  1758.220 MB/s [429252.9 IOPS]
  Random Write 4KB (QD=32) :  1340.484 MB/s [327266.7 IOPS]

  Test : 1000 MB [Z: 0.6% (24.8/4096.0 MB)] (x5)
  Date : 2014/12/06 12:46:54
    OS : Windows 8.1  [6.3 Build 9600] (x64)

About a 7-8% performance gain in 4k ops with only 1 timing cycle reduction. I may pick up some VRM fans and take them to around 1.5v, but there seems to be another important factor we are missing here.

753951 · Dec 6, 2014

I lowered mine from 15 to 14, and gain was about 3%.

schmuckley · Dec 6, 2014

schmuckley · Dec 6, 2014

man..I find it hard to edit around that picture..
but I wanted to post this for you guys:
http://overclocking.guide/x99-ddr4-memory-overclocking-guide/
A little birdie told me that tertiaries are very important with ddr4.
I use SoftPerfect for RAMdisk.

qkslvr221 · Dec 7, 2014

schmuckley said:
man..I find it hard to edit around that picture..
but I wanted to post this for you guys:
http://overclocking.guide/x99-ddr4-memory-overclocking-guide/
A little birdie told me that tertiaries are very important with ddr4.
I use SoftPerfect for RAMdisk.

Some very unfamiliar settings there, aside from tWCL I don't have any of those IOL settings, or any way to view this "RTL". My board has a huge number of other settings not included in this article (tRWSR, tWRDD, ODT PARK/WR/NOM etc.), so I'm not sure if they are using a different name for one of them.

schmuckley · Dec 8, 2014

qkslvr221 said:
Some very unfamiliar settings there, aside from tWCL I don't have any of those IOL settings, or any way to view this "RTL". My board has a huge number of other settings not included in this article (tRWSR, tWRDD, ODT PARK/WR/NOM etc.), so I'm not sure if they are using a different name for one of them.

RTL=Round Trip Latency
You could DL that tool they use for the 5Ghz 32 screenshots on XS then cross-reference the values to what they are called in your BIOS.
http://www.asrock.com/mb/Intel/X99 OC Formula/?cat=Download&os=Win764

Quad Channel RAM Disk Benchmark with Crstal Diskmark

Fully [H]

[H]ard|Gawd

[H]ard|Gawd

[H]F Junkie

Weaksauce

[H]ard|Gawd

Limp Gawd

Fully [H]

Weaksauce

Fully [H]

Weaksauce

Limp Gawd

[H]F Junkie

Limp Gawd

Weaksauce

Limp Gawd

Limp Gawd

Weaksauce

[H]ard|Gawd

[H]ard|Gawd

Limp Gawd

[H]ard|Gawd