SSD Speed Optimization
I am wanting to speed up my C1 7/8/9 exports on a Win 10 PC (or Mac Pro 2012) with solid state drives. I am considering using two SSDs per machine.
Would reading and writing from the same SSD cause loss of speed like it does on HDs with spinning platters?
Should I have one read SSD and a target save SSD to get the absolute fastest export speeds?
This is assuming no bottlenecks with the RAM or the CPU.
Would reading and writing from the same SSD cause loss of speed like it does on HDs with spinning platters?
Should I have one read SSD and a target save SSD to get the absolute fastest export speeds?
This is assuming no bottlenecks with the RAM or the CPU.
0
-
I have begun running tests to determine the best setup. Sadly, the PC I am using seems to be mostly processor bound and I am only getting 20 seconds faster times going SSD to SSD on SATA III (7 minutes) versus HD to HD on SATA II (7 min 20 sec).
It is an i7-2600K 3.4Ghz processor. I am going to switch to an i7-4820K 3.7Ghz machine in the hopes that the newer chip with better a better PCI bus will take advantage of the SSD on SATA III.0 -
Cliff,
Which SSD are you using and what is the claimed speed for read and write?
It's been a while since I looked into the speed issues but from what I recall of what was available then even the fastest SSD disks were not likely to sustain transfer speeds over extended processes and file sizes came into play alongside everything else.
I have an older (at least 4 yr old tech now) Samsung SATA 3 SSD installed on the main bus as the C drive (PM830 512GB) and a more recent Samsung 840 EVO 1TB drive in an mSATA format drive on a spare comms card slot which is only SATA 2- so nominally constrained to half speed.
Both drives have about the same read and write speeds specs - something like 540MB/s read and 520MB/s write afaik.
I tend ot do all of my C1 editing on the mSATA 1TB drive.Personally I don't really notice any great speed difference between the two or, indeed, an external 4Tb USB3 drive - although that can be a bit less "instant" in response.
When copying to the external drive on USB3 the speed seems to cap out at about 30% of the nominal USB2 comms speed. However I can have around 3 large file transfer running in parallel as about the same speeds for each of them. So I think the internal transfer speeds from the SSDs are constrained compared to notional performance specifiactions for this machine and the typical operations I might undertake.
I hope you may find this interesting.
Grant0 -
CliffCheney,
What are your raw- and processed file sizes and your batch count?[quote="CliffCheney" wrote:
...
Would reading and writing from the same SSD cause loss of speed like it does on HDs with spinning platters?
...
Reading and writing simultaneously always causes a bottleneck. Let it be HDD or SSD, a single one or two of them. That's because both actions share the same I/O-subsystem (in part or in a whole).
But with SSD, there is no delay because of the HDD's head moving or disk spinning. SSD accesses all blocks at the same speed, random as well as sequential.[quote="CliffCheney" wrote:
...
Should I have one read SSD and a target save SSD to get the absolute fastest export speeds?
...
CO reads a file, processes it, and writes the output file to disk. Then it takes the next file, and so on. That's a sequential process. I don't expect a significant difference between using a single internal SSD or two separate ones.
From my experience, reading (for example) a 20Mbyte RAW from internal SSD and writing a processed 60Mbyte tiff to internal SSD does not take a noticeable amount of time. Processing itself is the time consuming job here.
External USB disks in combination with processing large batches are a different story. But for small batches or single images, I don't care.
On my 2014 laptop (2.8GHz i7),
- the on-board SSD reads and writes a 24GByte file with 972MByte/s.
Buffered write response with a 20MByte file can be much faster. I saw up to 1.37GByte/s.
- a standard external USB3 HDD reads and writes a 24GByte file with 115MByte/s.
Buffered write response with a 20MByte file can be slightly faster. I saw up to 125MByte/s.
Finally, I think that improving processing power is the best way to go. I don't know the exact hardware specs of your system, but I don't expect that your i/o subsystem is the bottleneck in processing.
Regards,
Hans0 -
Thanks for your responses.
My controlled tests use the same 25 raw files from an IQ180 back. The SSDs are 250GB Samsung EVO 850 SATA III. I made sure to use the SATA III connections on the PCs.
The slowness does seem to be limited by the processor. I timed the export going:
D: drive to itself
D: to an X: drive SSD
X: SSD to Y: SSD
X: to X:
All of these scenarios on two PCs were about the same for the machine.The i7-4820 3.7Ghz (4 core) 16 GB ram was twice as fast as the i7-2600 3.4Ghz (4 cores) 16 Gb ram no matter what drive I was reading or writing. The Mac Pro 12-Core dual Xeon 2.4Ghz 12 GB ram times going only to the internal drive were about the same as the 3.4Ghz i7-2600 machine.
It was telling that the tests going from the D drives on SATA II 3Gbs to the SATA III 6Gbs SSDs did not differ more than a few seconds from the SSD SATA III to SSD SATA III tests. This points to the processor or ram speed as the bottleneck.
I may still run the SSD test on the Mac Pro and also do the tests on a set of Canon 6D files since they are only 20 megapixel and may not tax the processor as much as the 80 megapixel IQ180 files.
I was really hoping to get a reasonable speed boost. These results we quite disappointing.0 -
rom my experience, reading (for example) a 20Mbyte RAW from internal SSD and writing a processed 60Mbyte tiff to internal SSD does not take a noticeable amount of time. Processing itself is the time consuming job here.
How might I get numbers on how the discs are performing? I am testing on Windows 7 and Mac OS X 10.10.0 -
[quote="CliffCheney" wrote:
...
How might I get numbers on how the discs are performing? I am testing on Windows 7 and Mac OS X 10.10.
There are a number of good freeware speed test utilities available.
Windows users, don't hate me for this answer. Please. 😉
On Windows, I don't know.
On Mac, you can use the command line to do some basic testing. The 'dd' command can do the job.
dd if=/dev/zero of=~/testfile bs=1000000000 count=1
'dd' will print create a file of 1GB in a single run and print out a write speed. But it ignores existing write buffers. It might be more useful to do
time dd if=/dev/zero of=~/testfile bs=1000000000 count=1 && sync
'sync' forces the system to write buffered data and waits for successful execution.
'time' measures the time used to execute the command sequence of 'dd' and 'sync'.
Then calculate the speed from 'time' command and file size manually.
(The difference between 'time' and 'dd' is below 2% on my system.)
It's the same on Linux (a LIve-CD will do).
Regards,
Hans0 -
No Idea about Win10 but Win 7 offers some visibility via the Resource Monitor.
I can also still run Gadgets and have one called Drives Meter.
In terms of multiple session analysis and total elapsed times - which may be more useful than nominally "instant" speed reports - the C1 log file(s) may be a good source of timing information.
HTH.
Grant0 -
A good tool to test the performance of SSDs in general is "AS SSD Benchmark":
http://www.alex-is.de/PHP/fusion/downlo ... nload_id=90 -
[quote="hobie" wrote:
A good tool to test the performance of SSDs in general is "AS SSD Benchmark":
http://www.alex-is.de/PHP/fusion/downlo ... nload_id=9
Thanks for that link.
Very interesting and useful.
Grant0 -
The tests on the Mac Pro 12-core were about the same as the slower i7 PC.
So, I will probably just use the SSDs as OS drives since they don't seem to help C1 at all.0
Post is closed for comments.
Comments
10 comments