0
I would like to measure my RAM disk performance.
Normal hard drive testing tools can't help because RAM disk is a virtual drive, not a physical one.
I tried creating a large file an copying it across two RAM disks, but it just copies over instantly, I can't measure it.
I need something like a tool that would write random data to a file at given path and display the write speed (or time and size).
lolmaus - Andrey Mikhaylov
Posted 2018-05-14T12:52:32.083
Reputation: 383
Both GUI and CLI tool will do. But what I want to test is RAM disk file read/write speed, not raw memory speed. – lolmaus - Andrey Mikhaylov – 2018-05-14T16:43:30.910
I updated the answer. Get CrystalDiskMark. Note that performance difference between actual RAM speed and RAM disk test should be very low for a good RAM-drive software. – Overmind – 2018-05-15T04:49:36.093
Thank you! CrystalDiskMark's "select folder" feature is exactly what I need. – lolmaus - Andrey Mikhaylov – 2018-05-15T11:25:57.020
"Note that performance difference between actual RAM speed and RAM disk test should be very low for a good RAM-drive software" -- the difference for me is drastic. Linear read/write is 3992/3214 MB/s with CrystalDiskMark and 41573/40166 MB/s with AIDA64. – lolmaus - Andrey Mikhaylov – 2018-05-15T11:27:28.740
For comparison here are my CrystalDiskMark tests for world's fastest consumer grade SSD and HDD available at the time: Intel Optane 900P SSD is 2708/2275 MB/s and Seagate Barracuda Pro ST6000DM004 is 287/176 MB/s. – lolmaus - Andrey Mikhaylov – 2018-05-15T11:30:45.510
What do you use to create the RAM drive ? Note also that if you add encryption to it, performance changes significantly. – Overmind – 2018-05-15T11:36:09.540
I'm using IMDisk, no encryption. Here's a article that benchmarks RAM disk software. Some are better than others, but all of them are one order of magnitude slower than raw memory access. https://www.raymond.cc/blog/12-ram-disk-software-benchmarked-for-fastest-read-and-write-speed/
– lolmaus - Andrey Mikhaylov – 2018-05-15T11:38:59.233There are 200% differences between different programs, which means the software used matters a lot. – Overmind – 2018-05-16T06:33:17.167
Yes, but any RAM disk program is thousands of percents slower than raw RAM access. You claimed that "performance difference between actual RAM speed and RAM disk test should be very low for a good RAM-drive software". Maybe all the apps tested weren't good? Can you recommend a good one? :) – lolmaus - Andrey Mikhaylov – 2018-05-16T09:56:55.150
The testing procedure is much more complicated than it may seem. CrystalDMk is designed to show you a real-world practical speed. But in theory, you can use a RAM disk way faster than the tested programs in your link do. There are many things that matter, including the operating system specific limitations and where do you use the benchmark data from. Also, the function used to copy the data matters. Practically, to achieve a maximum speed you need to have an OS allowing you to freely use DMA and/or a program able to use it disregarding OS protections. Then, I'd make 2 disks and do a test. – Overmind – 2018-05-16T11:22:52.533
A practical similarity to this from a performance perspective using video cards for hashing. Yes, you significantly beat even CPU clusters with only one card but this requires proper software running on proper OS. – Overmind – 2018-05-16T11:23:59.280
CrystalDiskMark tests linear read/write speed. I don't he how any practical disk usage can get any faster than that. – lolmaus - Andrey Mikhaylov – 2018-05-16T15:12:42.583
And I do understand that "many things matter, including the operating system specific limitations" -- that's exactly the reason why I decided to benchmark file read/write speed and not RAM access speed. – lolmaus - Andrey Mikhaylov – 2018-05-16T15:13:43.257
Disk writes and memory rights are controlled first of all by the OS. It matters a lot. – Overmind – 2018-05-18T06:30:41.167