SSD or HDD for video editing

[bmadd89]

I am looking at getting an upgrade tomy current setup as i am not happy with its performance. System is as follows.

I7-980 @ 4Ghz
Asus P6T SE
Corsair 24GB 1600Mhz
Crucial M4 128Gb
4 x 600GB VR in raid 0.
Asus GTX 570 DirectCU
Seasonic 660W 80+gold
2x24" HP ZR2440w

M4 only has Win7m MSE, CS5.5 and CS6 Production Premium installed. Pagefile is reduced to 8GB and the hibernation file has been deleted to save space. Scratch disk is set to use C drive. AHCI is set to work in the BIOS and the raid array is using the in built x58 controller. I currently work on projects that have between 200 and 280GB of footage. Footage size can range from a few MB up to 4 or 5 GB. I get 10 days to complete the project and then that footage is archieved and i get a new set of footage between 200GB and 280GB. Footage is shot at 1920x1080p25 and imported directly into CS 5.5 (and soon to be CS6). This is all done in Premiere Pro.

Issues araise when i am scrubbing the timeline. This is with ZERO effects added to the footage so it is the orignal files. Scrubbing is better than it was when i only had 2x600GB VR but is still not close to fluid, real time scrubbing. CPU usage is still in check as it is below 60% (12 theards) all of the time unless im exporting the finished project (not a concern in this thread). Ram might be maxing out the 21GB assigned just to Premiere Pro. After a fair usage it does cache it all which is great but it still doesn't scrub fluidly so it might be RAM limited with my x58 platform?

I am looking for a solution to cure my scubbing issues as it is a nightmare when i have to wait 3 or 4 second for the footage to stop and start rewinding, scrubbing, anything that requires access the footage. I was looking at getting 2x 256/240GB (or more if it will do the trick or bigger ones?) SSD's in Raid 0 and using them as the storage for footage but as soon as i Raid them they loose the ability to Trim. And putting the amount of footage i do regularly would degrade performace pretty fast if garbage collection is not doing its job well, right?

Then i thought the the sandforce based drives have good garbage collection but their weakness is incompressable data. Like 1080p footage and high quality audio samples such as what i need it for. So then i was looking at the new 1TB VR that have recently be released but i would have to buy 3 to get marginal gains over my current setup and 4 for a respectable gain from what i can read.


Once the footage is on the storage device it is mainly being reads, correct? So if that is the case would the new 1TB VR be a better option as they provide over 200MB/s seq reads and writes.

For those who think im mad and should be happy with the set up i ask if you could shave seconds of a task that you do 5 or 6 times a minute for 50 hours a week, wouldn't you look for a way to remove that wasted time as it is a MASSIVE amount of time wasted waiting for it to respond. Please also understand it is my money that im spending and my time is worth more then i will spend on storage here.

Money was a concern to start with but i am really starting to get impatient and frustrated and am wanting to just throw money at it to get it to work. Clearly you can see i am will to spend upwards of $700 for the SSD's and over $1k for the 1TB VR.

Some benchmarks with footage on the Raid 0 and of the current state of the SSD.
4x600GB

M4

Recommendations or suggestions? And as always thanks.

 

[groberts101]

Before you spend 1 penny on that upgrade.. set up a small ramdisk based scratchdisk and compare the workflow to that of caching alone. Most are surprised that it is faster despite still implementing the same DRAM.

Also keep in mind that if you are running all your drives off the ICH?.. you are bandwidth hogging when transferring/reading data between the C-drive and your storage. You need to separate the volumes from being on the same sata chip and trust that the DMI can handle the overall throughput just fine. Because it can since mine sustains over 700MB's between the faster R0's for most but the absolute largest files which tend to slow to about 300-400MB/s after a minute or so.

You should also be running at least an 256GB OS volume there too. Adding another 128GB would be the quickest and cheapest way to improve the overall experience. You will need to implement idle recovery time to allow GC to clean up the mess without trim though. Which is not a huge deal if you don't mind the additional 10 bucks a month of electricity and parts don't "wear out" from running 24/7. Well.. maybe a few fans years down the road, is all.

And before you get too caught up on the folks who will likely claim that "R0 does nothing for performance in the real world"?.. look around at other Adobe users.

My workstation has 3 raidcards.. with 1 of them having 8x 500GB HDD in R0 for my main storage volume. Another card with 4x 1TB HDD in R0 for redundant storage of data(no one said R0 redundancy was cheap). Then the most often used card is filled with 2x Vertex3 240GB in R0 for dedicated scratch duty.. 1 x Vertex 4 256GB for OS redundancy and testing.. and various single SSD's/HDD's on the remaining ports. I also have redundant HDD R0 on my Marvell ports.. and use all 6 ports of my ICH10R to run 6x Vertex 2's as my main OS volume. I then use Fancycache by Romex to cache any volumes on my machine that I access redundant data to/from(which is where caching software, including Windows, works best).

Ramdisks are faster.. but you need larger amounts to really make them sing for your type of workflow. Consumer based boards obviously limit you there but WS boards will up the price dramatically and you'll need to reassess just how much your time really is worth.

So, why would I spend/do all that? Because I too have a dedicated vid/gfx workstation which uses Adobe CS4 MS and most of those storage bottlenecks are removed with such a config. My files rarely go over 100 gigs though with most being under 10 gigs.

My rec's to you without taking out a second mortgage:

Get a faster/larger C-drive(which usually requires R0 for such an extreme usage).

Install a raidcard(Highpoint 2720 and IBM1015 off ebay(same as LSI9240-8i) are the best buys) and stick the 4x VR's on the card. You could also use the raidcard for your C-drive array and keep the VR's on the ICH.

Give Ramdisk and caching software a try.

Bout all I can think of with limited time right now.

 

[bmadd89]

I don't know much (anything really) about external raid cards but with the ones you are suggesting, why are they better? Should i start with something like an Adaptec Raid 1430SA or are they garbage? The main reason i ask is i dont need 8 ports and i dont need redundency built in. if i got something similar to the above i could put the 600GB VR on it and give them 500MB each. Get another 128GB M4 and raid it on the ICH10R and possibly get another 128GB for a scatch disk drive?

I don't really have enough ram to properly implement a scratch disk do i? How much would i have to allocate to benifit properly?

 

[Phynaz]

What does perfmon say while you are scrubbing?

 

[groberts101]

On the reread.. I see that I may have barraged you with too much info in one post there. lol

Let's make it simpler. You are effectively bottlenecked due to running your VR raid AND the SSD off the same chip.

Furthermore, your main SSD volume is too small and slow for that intended application. 256GB would likely be the minimum standard.

Card choices should be limited to sata 3.0 spec's. The IBM1015 is a wicked good deal if you don't mind flashing its firmware/bios to an LSI 9240. I paid $70 bucks for mine off ebay. Next best bang for the buck is the $150 dollar Highpoint 2720(plus you'll need sas to sata breakout cables). From there you jump into the $200 range with very little gain over the cheaper versions mentioned above.

As for ram quantity requirements?.. just use what you can and test with smaller files. That should paint a true enough picture for you to assess its impact on the workflow.

You could also look at a PCI-E based SSD to reduce the bandwidth hogging which is likely occurring when running all volumes off that same chip/ICH.

 

[bmadd89]

This is what Perfmon says. I think i added the right disks (total volume bytes) and put it to the correct scall so it is showing 1Mbytes /second.

Thanks for all the info so far groberts101 and i got what you were trying to say. Thats why i thought if i just limited the 2 M4's on the built in raid and added the 4 raptors to the raid card it would certainly eliminate said bottleneck as the raptors would get 500MB over a 4x pci-e v2 bus.

I looked last night and i couldnt get the Highpoint card in australia but i look tonight and it is there. The IBM is no were to be seen and im not entirly keen on ordering it from the states.

So i will buy the HighPoint card but like you mentioned i will need SAS to SATA cables. o've never dealt with this before. Can you give some advice please.

 

[groberts101]

you will want to stay away from the kit and stick with the model designated as "SGL" which does not include the SAS to SAS cable that does you no good with your current drives. No use in paying for cables you don't need.

http://www.amazon.com/HighPoint-Rock.../dp/B0050SLTPC

Then you'll need forward sf-8087 to sata breakout cables like this.

http://www.amazon.com/gp/product/B00...7NFDK7NS4YJ3VW

Once you get the card you will want to make sure the card has the latest bios. All the necessary downloads are here.

http://www.highpoint-tech.com/USA_new/CS-PCI-E_2_0_x8_Configuration.html

Otherwise, it's all pretty straight forward from there in that you can build the raid in the cards ROM at boot?.. or use Highpoints Windows based webGUI utility. Since it's an entry level card it will not allow you many config options but you will surely see a major jump up from that ICH config. Probably at least 100+MB/s gains to be had there.

Then get the M4's in raid on the ICH10R or you could play with the remaining ports on the 2720 if you want(which of course requires 2 cables for use of all the card's ports). I've toyed around with 6x vertex 2's achieving 1.5GB/s on the 2720 and I still prefer the much fatter lowend grunt that the ICH has to offer over the tall sequential's of the 2720 when it comes to an OS volume. Therefore, I prefer to use cards for their generally designed and intended HDD storage duty.

Then after all that, I would seriously give Fancycache by Romex a try. You could set up the cache in front of the storage array and make use of it for small'ish files(by your standards, anyways..lol). Fancycache also gives you the option of making a hybrid drive(HDD/raided HDD volume + SSD for cache). That is something that I would quickly test to get a better overview where it may be best to spend your cash because you may very well find that a hybrid drive solution may be cheapest in the long run.

As a quick and dirty test, you could install fancycache to your current setup and use part of your SSD's available space(Fancycache will assign its own folder to the SSD for caching purposes) and it's completely reversible without causing the boot volume any issues. You'll obviously consume larger amounts of fresh blocks on the SSD when using larger files.. but you'll get a very real look at what people who swear by hybrid arrangements are taliking about. They do work well, and especially with redundantly used data(because the file needs to be cached first. Fancycache is also advanced enough to allow 2 levels of cache. The fastest obviously being ram with SSD being second in line as the ram runs out. It's kind of like allowing your pagefile to grow as required so you never slow down.

I'll stop there so I don't overload you again.. and good luck on the hunt.

 

[bmadd89]

So AMazon can't ship either of them to Australia so i will have to get them elsewhere. Do you mind having a look at the links below and double checking they are right please.

We have this one
http://www.techbuy.com.au/p/179076/...ATTACHED_SCSI_(SAS)_C/HighPoint/RR2720SGL.asp

And the cables.
http://www.techbuy.com.au/p/76805/CABLES_DRIVE_SAS/3Ware/CAB-8087OCF-10M.asp

I will try the fancycache when i get home tonight. Thanks for all the help.

 

[Hulk]

Get a 256GB SSD when the next $200 sale pops up. Use this for your video work drive. When it's time to render, do that to a mechanical disk to save a little wear on the SSD. Renders are almost always CPU limited anyway.
Will the SSD help with timeline preview? Hard to say but you have a lot invested in your video editing rig and performance is important so I think it would be worth it to give it a try. The fact that your CPU usage isn't floored makes it seems like a good SSD would help your issue.

You see while a mechanical drive has the throughput capability to handle all of your streams, the heads have to move all over the place to grab the data for each stream, which kills read performance. It's quite likely the SSD won't have that problem. Of course as I wrote above, video editing rigs can defy logic at times so there are no guarantees. But I think your logic is sound.

 

[groberts101]

So AMazon can't ship either of them to Australia so i will have to get them elsewhere. Do you mind having a look at the links below and double checking they are right please.

We have this one
http://www.techbuy.com.au/p/179076/...ATTACHED_SCSI_(SAS)_C/HighPoint/RR2720SGL.asp

And the cables.
http://www.techbuy.com.au/p/76805/CABLES_DRIVE_SAS/3Ware/CAB-8087OCF-10M.asp

I will try the fancycache when i get home tonight. Thanks for all the help.

holy crap!.. your prices are high there. That card has increased in price since the popularity with SSD has gone up lately though, so there's that to consider as well.

The 2720SGL is correct.

The 1 meter cables are a bit long if the drives are not externally located though. I have 1 meter cables that I used to run an external 8x HDD enclosure and they are looped and tucked away in my case now that I have all 8 of those drives located internally now. Might want to go with half meter cables to avoid the rat's nest later on.

Other than that.. looks like you're on you way towards a faster system. Many people underestimate the time savings possible when you have storage speeds matched to a "fast as hell" raided SSD volume. Especially with larger transfer volumes. Gives the impression that you upgraded your ram AND cpu all at once. lol

 

[bmadd89]

Welcome to Australia IT prices. That is the cheaper side of life really. Off topic for a moment. Inno3d GTX680 is $703, 3770k is $383, 3930k is $616. M4 128gb is $177, sapphire 7970 is $537. considering our dollar is almost equal with you its a bit annoying.

Back on topic. I have just order the card and the 1M cable so that i have come to move later on in a bigger case which i might need. I will post back some results probably tuesday depending on when they get it shipped to me.

Many thanks.

 

[bmadd89]

Well it hasn't made a difference i am still gettiing the same transfer rates as i was using the onbard raid. I have ordered 3 more M4 128GB and am picking them up this afternoon. I will use 2 of the VR in raid 0 for the OS and 2 in Raid 0 for the scratch disk and have a 4x128GB M4 raid on the pci-e raid. That should give me 1GB read and write speeds and enough storage for 1 project. Will post back some stats when i install them later tonight.

 

[Makaveli]

Interesting thread looking forward to seeing your updates.

 

[Phynaz]

Well it hasn't made a difference i am still gettiing the same transfer rates as i was using the onbard raid. I have ordered 3 more M4 128GB and am picking them up this afternoon. I will use 2 of the VR in raid 0 for the OS and 2 in Raid 0 for the scratch disk and have a 4x128GB M4 raid on the pci-e raid. That should give me 1GB read and write speeds and enough storage for 1 project. Will post back some stats when i install them later tonight.

You jumped the gun. Look at your Perfmon graph. You're not I/O limited, you're CPU limited.

 

[Makaveli]

You jumped the gun. Look at your Perfmon graph. You're not I/O limited, you're CPU limited.

Question how is he cpu limited if the graphs show the processor never hitting 100% usage.

While I/O hits it.

 

[BrightCandle]

Question how is he cpu limited if the graphs show the processor never hitting 100% usage.

While I/O hits it.

The IO is only hitting about 100MB/s, its fair to assume from the previous benchmarks that the peak should be 3 to 4 times that. The task may not scale to multiple CPUs well.

I have not seen enough data to determine the current bottleneck, but I would hazard a guess its more likely CPU than HDD based on the data on display.

 

[Phynaz]

Question how is he cpu limited if the graphs show the processor never hitting 100% usage.

While I/O hits it.

CPU limiting happens before 100% utilization. Especially on a threaded CPU.

He hits 80 - 90 percent. That would be CPU limited in a perfect world. Since I'm willing to bet Premier isn't perfectly multi-threaded, he's peaking on CPU performance.

 

[Makaveli]

The IO is only hitting about 100MB/s, its fair to assume from the previous benchmarks that the peak should be 3 to 4 times that. The task may not scale to multiple CPUs well.

I have not seen enough data to determine the current bottleneck, but I would hazard a guess its more likely CPU than HDD based on the data on display.

CPU limiting happens before 100% utilization. Especially on a threaded CPU.

He hits 80 - 90 percent. That would be CPU limited in a perfect world. Since I'm willing to bet Premier isn't perfectly multi-threaded, he's peaking on CPU performance.

Thanks for the update both.

If he is indeed cpu limited he is already on a 6 core chip his only option would be SB-E.

I would assumed since he is video editing it would be multithreaded, but you are probably correct phynaz.

 

[bmadd89]

It seems that everyone has been sort of right in their own regards. I went to get 4 128GB M4s (only had 3 there) Put them on the raid card and set up raid 0. Transfers are now much better (see pic). Now when i am importing files im getting a solid 130MB/s disk activity were before it was speratic and below 100, scrubbing is now fluid except for when i am doing multicam with 3 or more cameras (using perfmon im getting around 180MB disk activityduring scrubing), playing back in the source monitor is now the only real issue that remains and that i believe this is becuase i am cpu limited. If i play through the timeline and then see something i need to change and press the rewind button 3 times for quick rewind the cpu usage jumps up to 80% and sits there for a second or 2 and then the footage starts to rewind. I'm going to try and clock the i7-980 a bit more but it seems i might have to look at getting a 3930k or one of intels new 8 core xeons. But my question is the problem with the xeons is they might have more threads but they don't have the clock speed im getting from the i7-980. So would it really benifit me if i moved from a 6C/12T @ 4.2Ghz to a 8C/16T @ 2.4Ghz? Or if i went down the 3930k what kinda of % increase would i get from the IPC increase and an extra 300Mhz?

All in all it is better then it was when i posted my question but still not perfect and maybe i would have to spend upwards of 10k to get it to were i want (which i dont really want to do)

 

[DominionSeraph]

You are running a clean and tweaked install of Windows with updated chipset drivers installed, right?

 

[BrightCandle]

You already have an i7 980, alas that is about the bees knees for CPU performance, Intel just hasn't gone very far in the last few years with performance. IPC improvements of SB-E amount to at most about 20% and you should be looking at 4.5 Ghz on the SB, which is another 12.5%, giving you a grand total of 35% improvement. That isn't going to cut a 2 second lag down to anything better than 1.5. CPUs just aren't fast enough for what you are doing and I feel your pain as I have code here that takes 10-15 seconds to compile and test and its a real pain in the backside.

Only thing I can think of is to consider going up to 2 sockets. IIRC EVGA has a dual socket SB-E motherboard and you could see if the task you have scales to 12C/24T. Its possible it would scale and the combination of SB-E plus dual cores would get you well under a second (nearly 0.5s) instead of 2s on those seeks, but its not a cheap thing to test and it might not work. I would start by asking Adobe how far it scales and whether dual cores will improve it yet further.

 

[bmadd89]

@DominionSeraph I most certainly am. As far as i know anyway. Do you have any recommendations. Drivers are up to date. First thing i went to.

@BrightCandle Say i bought a 3930k, clocked it to 4.5Ghz and filled the system with ram (64GB). I'm looking at $1300. I should be able to net a bit more of an improvement at of having some 40GB more ram shouldn't I? And if i get some 1866 latency would be down a bit more and there would be bandwidth issues. So maybe more then 40% increase?

Say i put a good water cooling loop in on the 3930k i should be able to push the clock up a bit so a similar 4.7 like you or maybe higher?

 

[BrightCandle]

@DominionSeraph I most certainly am. As far as i know anyway. Do you have any recommendations. Drivers are up to date. First thing i went to.

@BrightCandle Say i bought a 3930k, clocked it to 4.5Ghz and filled the system with ram (64GB). I'm looking at $1300. I should be able to net a bit more of an improvement at of having some 40GB more ram shouldn't I? And if i get some 1866 latency would be down a bit more and there would be bandwidth issues. So maybe more then 40% increase?

Say i put a good water cooling loop in on the 3930k i should be able to push the clock up a bit so a similar 4.7 like you or maybe higher?

Buying more of anything only helps if that is the limitation. Now it might be that actually the CPU is stalled on RAM and hence higher bandwidth RAM directly translates into faster performance, but its pretty unlikely. SB-E has a memory controller that in many ways is actually worse than your 980X, its highly tuned for the cores to all be accessing different chunks of RAM and the single threaded access pattern is somewhat hindered. In your circumstance it might be one of the cases it is actually better, but I suspect its single percentage points only, and more likely it does nothing at all.

If you went to more RAM that would allow the computer to not use the SSD's as much, which may help a lot depending on how much IO wait time you are seeing. But with 90% CPU utilisation assuming all the rest is IO wait you could only possibly gain 10% by removing the wait entirely.

As to 4.7 Ghz, its doable with a custom water loop and a decent chip. But right now I am at 4.5 Ghz and I can't get it stable above that, not without going silly on voltages for a work horse machine. I wouldn't count on getting much higher than 4.5 unless your willing to search for a golden chip to get 200 more Mhz.