Does an SSD improve render times?

[GoodEnough]

Bottleneck is mostly the CPU, but can SSD ever improve render times ?
Faster read/write than HDD

[johns0]

Unless you can render faster than a regular hdd at around 120 megabytes a sec then nope.

[johnmeyer]

The time you can save by using a faster disk drive is easy to calculate. Here's how.

Take a file that you rendered, and copy it to another location on the hard disk drive. Make sure you COPY it, rather than move it. Measure the time it takes to do that copy.

Repeat the test, but this time do it on an SSD drive. The difference in those two times is how much time you will save by using an SSD.

If you don't have an SSD, you can approximate the answer by looking at various sites which measure SSD and hard drive writing speeds. Some sites claim that the ratio is 10:1. If that is really true, then to a first approximation, the SSD write time is zero (compared to the hard drive) and therefore the time to copy the file to another folder on the hard drive is the maximum time you can expect to save. For "typical" 2-10 GB files, you should be able to do the copy, even on a hard drive, in only a couple of minutes. That is therefore the maximum amount of time you can save.

Bottom line: an SSD is not going to save you much time.

[jagabo]

The time you can save by using a faster disk drive is easy to calculate. Here's how.

Take a file that you rendered, and copy it to another location on the hard disk drive. Make sure you COPY it, rather than move it. Measure the time it takes to do that copy.

Repeat the test, but this time do it on an SSD drive. The difference in those two times is how much time you will save by using an SSD.

That's usually not accurate. With multithreading the reads and writes can overlap with the encoding so the time it takes to read/write may be completely inconsequential. It starts to matter when the source or output video is uncompressed or losslessly compressed, or when encoding is very fast.

[johnmeyer]

The time you can save by using a faster disk drive is easy to calculate. Here's how.<snip>

That's usually not accurate. With multithreading the reads and writes can overlap with the encoding so the time it takes to read/write may be completely inconsequential. It starts to matter when the source or output video is uncompressed or losslessly compressed, or when encoding is very fast.

It's accurate, but I agree that it is a worst-case calculation. And I also agree that for h.264 source files, which are highly compressed, the time difference is inconsequential.

Also, I tried to get information I could trust on the current write speeds for SSDs. In the early days of SSD, their write times weren't much better than hard drives (might even have been worse), but they were spectacularly faster for reading. I know they've gotten faster, but for writing, they may not be massively faster. However, I haven't looked at this for years and could very definitely be wrong about that.

[johns0]

Depending on the ssd the writes speeds are very fast up to a point were after 100+gb of files will slow the ssd to 100 megabytes per second.

[abolibibelot]

Depending on the ssd the writes speeds are very fast up to a point were after 100+gb of files will slow the ssd to 100 megabytes per second.

That's due to “throttling”, right ? It would seem like some models are particularly prone to this issue, others not at all, from what I could gather a few years ago.

Also there's the issue of fragmentation ; it shouldn't affect a SSD, but on a highly fragmented HDD the actual read/write rate can be significantly reduced. Yesterday I copied a bunch of video files between two 3TB HDD, both 5400RPM models, normally this should be done reasonably fast at around 120MB/s, as said above (7200RPM models with larger capacity platters can have a read rate beyond 200MB/s), but those files happened to be highly fragmented (most likely because some of them were downloaded simultaneously by a utility which doesn't preallocate the file's expected full size), I'm talking 15008 to 17505 fragments each (as analyzed by Defraggler), so the actual copy rate was a measly 20MB/s. (I purposefully transferred the most fragmented first, as it would be a very intensive task to defragment them in-place. I still have many highly fragmented files on that drive, which I don't like, but again, doing a full defragmentation would be hazardous, since most of those files will have to be transferred anyway I'll wait until I have enough storage space to transfer them all.)

And the read/write rate on a HDD is significantly lower at the end of the physical storage space (corresponding to the innermost “cylinders”) than at the beginning (corresponding to the outermost “cylinders”). If the max rate is 120MB/s at the beginning, it can reach ~70MB/s at the end. So a HDD which is 80% full (or if using a dedicated partition that corresponds to the last 20%) and with a high fragmentation of free space will have a dramatically reduced performance compared with the same HDD freshly formatted.

[johns0]

Not throttling,just running out of cache.