Help! Tell me your HD/RAID max average frame rate with this test program, try ZIP

[Dan Hudgins]

I made a small utility to test a PCs HD,SSD,CF,SD,USB memory stick, drive's average speed for saving frames.

Please help me get an idea of what speed you computer can operate at by running this utility on your PC computers, it is a simple DOS program, and should run under ME or XP Home, maybe some versions of Vista. Its a 32bit program, so probably won't run under FreeDOS, I could make another version for that maybe later.

Anyway you make a folder (8 letters or less) and give the full path like,

C:\FOLDER\

Unzip HDSCK002.ZIP into it,

rename the file DANHDSCK.BIN to DANHDSCK.EXE, I renamed it to help it get past some virus filters that don't like EXE files.

You enter the values asked for i.e.

C:\FOLDER\
2592
1102
1.5
1440

And let it run, it should make 1440 "frames" in less than a minute, but may take longer, it then gives a report of your HD/drive speed.

To remove the frames you need to use the DOS DELETE command like this,

DELETE C:\FOLDER\00??????.XXX

Or you can use the Windows delete marked block command, but you then need to empty the waste basket also.

Please tell me in a reply or PM/email what frame rate you get on your system from the settings give above. The program works up to 6K, but I am interested in 2592x1102x1.5@24fps for now, to see what people can get on their RAID arrays, SSD, and normal HD.

Its best to enter 14400 frames if you have the free HD space, since the frame rate tends to fall off as the frame numbers go up, 14400 is 10 minutes of recording, so that would be the longest shot for the most part.

Thanks for any help. I need to find out what drive types can record better than 24fps, I only get about 1.5fps on my system, and 4fps on my Brother's, at 1440 frames, so let me know what you get and what type of drive you are using and system specs and number of frames made, it will be a big help, thanks!

EDIT: NEW WGUI-v0.04 now put in ZIP file for testing under VISTA and WIN-7 etc! let me know if it works, rename DANHDSCW.BIN to DANHDSCW.EXE to run it, see about.txt in the ZIP file.

EDIT: APRIL 2010. I removed the ZIP file from here, if you are interested in it you can PM me, I am thinking of making a revised version and it could be too large to fit in the ZIP file space here, so I may post it on my WEB site. Thanks for the results they were very useful and constructive.

 

[Paul Leeming]

Doesn't work under Windows 7 x64 sorry. Look forward to a version that does.

HTH

Paul

 

[Dan Hudgins]

New windows gui version!

New windows gui version!

Doesn't work under Windows 7 x64 sorry. Look forward to a version that does.

HTH

Paul

I ported the DOS version to Windows GUI today, it seems to work under XP Home but not under W95 or W98 (ME?), so use the DOS version for OS under XP Home, both DOS and Windows GUI versions are in the ZIP file above, see the file about.txt in the ZIP for more information. Be sure to rename the BIN files EXE before you run them.

 

[Paul Leeming]

Thanks Dan, your new version with GUI worked fine under Windows 7 x64.

Here are the results for my fastest drive (2x Samsung SLC based SSDs in RAID0):

If I'm reading the end result correctly, I wrote 1 minute's worth of footage in a shade over 30 seconds, which is almost 2x realtime? If that's true, I think maybe your end %@24fps value is incorrectly calculated: it should be (60/32)%, not (32/60)%, right?

HTH for your testing!

Paul

 

[Paul Leeming]

Here's my other SSD array, 2x Patriot Warp v2 MLC based SSDs, again in RAID0:

So either way, one of them is faster than real time, the other isn't it appears. Which one is the best one?

HTH

Paul

 

[Dan Hudgins]

Which is the best one? lower Ts value.

Which is the best one? lower Ts value.

Which one is the best one?
Paul

Thanks for the great reply!

The drive with the lower Ts value (Total Seconds) is the fastest one, you can confirm that with a stop watch, but since I am using the Windows timer in the GUI version of the HD/SSD test program it should read about right(?). %@24fps is the percent of the TARGET time, so should be less than 100% to work.

The % in the lower right is the percent of the time it takes to the time required for shooting the frames, so only values LESS than 100% will work. In other words if it you enter 1440 frames which would record for 60 seconds at 24fps the test program should take LESS than 60 seconds to save 1440 frames in order to give the computer time to do other things like read the LAN or USB port to get the image data. Its the time part used for saving the frames, in a real camera you would add the time parts for all tasks to try to come out less than 100%, ie 30% to save frames + 20% to download frame data = 50% total time used (so the computer can use the other 50% for the viewfinder and other tasks).

Its great that you posted the screen so I can see all the readings.

Since one of your drives is fast enough that is also good news!

What controller (card or motherboard?) are you using for the "2x Samsung SLC based SSDs in RAID0" pair of drives?

You can enter other resolutions, like for Scarlet if it gets a true RAW port,

3072x1536*1.5*24=170MB/s (for 12bit sensor, 1.5bytes per pixe 2:1 ratio)

3072*1728*1.5*24=191.1MB/s (for 12bit sensor, 1.5 bytes per pixel 16:9)

3072*1728*1.75*24=223MB/s (for 14bit sensor, 1.75 bytes per pixel 16:9)

For the REDONE,

2048x1024*1.5*24=76MB/s (for 12bit sensor, 1.5 bytes per pixel 2:1)

2048x1152*1.5*24=85MB/s (for 12bit sensor, 1.5 bytes per pixel 16:9)

4096*2048*1.5*24=302MB/s (for 12bit sensor, 1.5 bytes per pixel 2:1)

4096*2304*1.5*24=340MB/s (for 12bit sensor, 1.5 bytes per pixel 16:9)

It is possable to record true RAW data with less bandwidth, closer to 1 byte per pixel or less, by using delta encoding, that is lossless, its just a way of writing the same data with fewer bits. But this gives you a test to see what is possable for Digital Cinema to get the maximum quality out of any given sensor for true RAW recording. I was interested in the 2592x1102x12bit@24fps since that is about the maximum of the <$50 Aptina MT9P401I12STC sensor, its about the lowest cost 2.5K sensor for a Digital Cinema camera, its a little larger than 1/2 the size of the Scarlet so called 2/3" sensor.

So from your readings the Samsung SLC took 32.6 seconds and the Patriot Warp v2 MLC took 105.1 seconds, so

32.6/105.1 = 31% of the time required for the Samsung,

or the other way,

105.1/32.6 = 322% faster for the Samsung!

It would seem at least from this test, like I said you can confirm the Ts with a stop watch to see if the program and Windows timer are giving the right time values. The MB are figured from the size of the frames being saved.

In a real camera some kind of FIFO would be needed probably to make sure that no frames are dropped when the HD/SSD glitch for a few milliseconds or longer.

The other feedback I got on the DOS HD/SSD tester may help me get it working in DOSBOX later, I did some tests and think I figured out that the problem is that it uses more than 16MB, so that shoud not be too hard to fix later, anyway the GUI version probably gives faster readings under W-7 anyway, its great news that the GUI version runs under W-7, its my first Windows program!

For Scarlet it looks like the Samsung drives are close to being able to record 3K 2:1 ratio at 24fps maybe, so true RAW recording (and using delta encoding) is not that hard any longer, maybe? It looks like both 2:1 and 16:9 2K modes on REDONE could be recorded since they are under the reading needed it seems?

Please let me know about what controler was used to RAID the Samsung drives, also please the model number of the Samsung drives used for the test, thank-you!

 

[Paul Leeming]

I'm using the AMCC/3Ware 9650SE 4PML hardware RAID card, and the Samsung drives are these ones:

The rest of the specs are as per my Monolith thread, but the gist of it is an overclocked i7 CPU to 3.6GHz along with an ASUS P6T6 WS Revolution motherboard with its front side bus overclocked to 1600MHz (Quad 400MHz).

Funny thing is, I realised later that I had other stuff running in the background so it could have been even faster! :laugh:

I'll try the Scarlet 3K specs and report back....

HTH

Paul

 

[Paul Leeming]

Ok, here's my Scarlet 3K tests. It looks like 3K 2:1 12bit can be recorded RAW (just), but 16:9 is just a little slower than real time.

But.... these SSDs are not the latest generation hardware even though they are extremely fast (they are enterprise level drives). And I've only RAIDed them with 2 drives. I think with 4 current generation SSDs RAIDed together as one unit, you could easily get 3K or even 4K RAW throughput.

Scarlet 3K 16:9 Test Result (FAIL):

Scarlet 3K 2:1 Test Result (PASS!):

HTH

Paul

 

[Dan Hudgins]

Great Info!

Great Info!

... I think with 4 current generation SSDs RAIDed together as one unit, you could easily get 3K or even 4K RAW throughput.

Paul

Great job with the info on the SSD and the Scarlet resolution tests. With 11% of the computer time "free" for other things it just might be possable to shoot 3K 2:1 at 24fps into 2xRAID0-SSD now! I think a true RAW port module for Scarlet would need to have a large FIFO to avoid dropped frames, maybe a 4 minute FIFO so that a whole shot would fit, but is seems like SSD are getting fast enough to record all the sensor data for even better results from the great sensors. I put up a suggested pinout for a 25pin connector that could be used for a true RAW port on the Open-Source camera thread at ScarletUser. Newer LAN might also work with delta encoding and other lossless tricks. With a 4 minute FIFO you can run the camera in "Shot-then-Dump" mode where the FIFO fills while shooting and drains between shots over a slower link like a LAN or even USB.

Thanks for all your work in putting up the tests and for the other feedback I have gotten by PM, I have learned a lot so far!