I tried a Mac for video editing...

[Toque]

I'd like to see you build another one Dave and then use them as a Rendering farm cluster:

https://www.sas.upenn.edu/computing/mms/parttime/cluster

http://cg.tutsplus.com/tutorials/autodesk-maya/an-introduction-to-backburner-and-render-farms/

http://en.wikipedia.org/wiki/Dynebolic



I know.....I'm no help, but it seems like you've reached the limits for a single computer.

I kinda find it funny that Dave has meantioned he doesnt care about using gpu for rendering power and asked for no comments on tha subject, but yet everyone still meantions it I havent bothered to look into to, but my off tha bad thoughts from what ive seen around are its not really worth it unless your into professional and have just tha right hardware, can someone correct me if im wrong with tha benchmarks to back it up?

[Toque]

bla blah

Dude, Dave's in Australia.  Check their prices and be careful not to have a heart attack. He had to pick something cheap and available locally.

That PSU is a standardized OEM CWT design (CWT likes to use that dark green tape on transformers), modified at Corsair's request to save more money by using Aishi and Capxon capacitors and using lower AWG wires and so on. 
Most of the components are UNDER the PCB, what you see above is just the through hole stuff.

Ya, I had a freind I used to talk to on tha IRC for many years, if I remmber correctly, something like a $500 electronic/computer item here was $700-750 there back in those days.. But after all, not as much a factor when thats tha price there, they must be somewhat use to it, is it still tha same idea? Oh, large amount of surface mount underneath is there? Still a top notch design for that price point/series..

[mariush]

A good video card matters if you edit videos which involve multiple video sequences in the "timeline" that you have to preview a lot, videos overlayed on top of another, doing color corrections (for example, if someone recorded a video in a room full of wood furniture and the lightning made the people's faces orange from the reflection in the wood you want to adjust the colors), applying logos, transitions and special effects, basically doing.

Dave only imports segments he shot with the camera, trims them where needed, at some points maybe adds some text on picture or speeds up the video, all this stuff doesn't involve the video card or using video card doesn't improve things... it's just basic source video decoding and right away back to output format encoding ...

Recommending a 400$ card to improve the encoding speed by 5-10% (from  2 minutes 20 seconds to 2 minutes 10 seconds) is just stupid. You have to know where it's no longer good value for money.
 

[Toque]

A good video card matters if you edit videos which involve multiple video sequences in the "timeline" that you have to preview a lot, videos overlayed on top of another, doing color corrections (for example, if someone recorded a video in a room full of wood furniture and the lightning made the people's faces orange from the reflection in the wood you want to adjust the colors), applying logos, transitions and special effects, basically doing.

Dave only imports segments he shot with the camera, trims them where needed, at some points maybe adds some text on picture or speeds up the video, all this stuff doesn't involve the video card or using video card doesn't improve things... it's just basic source video decoding and right away back to output format encoding ...

Recommending a 400$ card to improve the encoding speed by 5-10% (from  2 minutes 20 seconds to 2 minutes 10 seconds) is just stupid. You have to know where it's no longer good value for money.
 

Ah, thanks for clearing that up, so more so for closer to/or realtime overlay type stuff, not as healthy for overall encoding speed, makes sense..

[Toque]

So, I bought one of those cooler master T4 cpu coolers, as these new coolermaster seem tobe one of tha best for tha money, got it for $17 as I was paying $5 shipping to get a Netgear GS108 for $24 anyways, thought I would throw it in for a future build, like a AMD A4-5300, impressive cpu/gpu combo for tha moola, that heatsink should easly keep her cool, tha fans that come with tha lower end AMD cpu are horrible, they are coolermaster spec fans I think but at tha best price point possible to box with tha cpus, I have a A10-5800 here that as soon as u moved tha mouse tha fan would scream, horrible, I switched that to this coolermaster for $30 as its in a desktop case, http://www.newegg.ca/Product/Product.aspx?Item=N82E16835103177  I put a A4-5300 together for my sister, and used tha A10-5800 cpu cooler with it, that doubled tha amount of heatsink and set tha MSI smart fan stuff and that worked out okay..

[T4P]

I guess I'm a little late to the party, but I wanted to throw in my 2c... 

I have an i7 920 and I run it with the stock cooler.  I have no problems with noise or overheating whatsoever, even after running for several hours at full tilt.  My RAID array makes more noise than the processor fan. 

Stock frequency?  I'm using a L5520  with a MASSIVE heatsink

Most of you are using tiny heatsinks compared to mine ... i'm planning to jump in WC'ing as well. Don't know why i'm doing that on a chip that runs 3.6GHz with top 50W power consumption 
Man,

No, not really a good value.

I've reconfigured the i7 3770K build:
http://www.scorptec.com.au/product/45658
http://www.scorptec.com.au/product/45270
http://www.scorptec.com.au/product/49376
http://www.scorptec.com.au/product/44675
http://www.scorptec.com.au/product/45304
http://www.scorptec.com.au/product/38787

$962

It looks good. I'd only change the power supply. Thermaltake is not really known for consistency, they use power supplies from different OEM manufacturers in the same series, or from revision to another, it's a mess.
That power supply is most likely a budget FSP based design from a few years ago. It's probably technologically outdated by now but it would work fine, it will be stable and work with your system, but for how long that's unknown. It's worth spending 10-25$ more for a reliable power supply, you'll be able to reuse the power supply on other systems later on.

You're also not including additional cpu cooler.

The Intel stock coolers are kind of anemic, they don't handle temperature variations well, they ramp up their fan speed quite fast the moment the cpu starts doing something and it can be annoying. In addition, the design is prone to getting full of dust real fast.

AMD stock coolers are more relaxed in this sense and they cool better and use heat pipes (intel stock coolers are just copper die in center and aluminum fins) but they're still somewhat noisy.

I don't agree with the guy that said to go for 4x4 GB memory modules. You don't get more bandwidth with 4 modules, just more heat and if you're overclocking, a tiny bit more risk of not being able to overclock as much.
The ones I recommended and the ones you linked to are LOW PROFILE, so they're work with any third party cooler, have low latency, they're very good. They're a bit more expensive but it's worth it.

Simple solution : BUY SEASONIC. EVERYTHING IS EPIC.
I actually use 4x4 ya know ... they don't waste much heat to begin with (<1W per DIMM)
AMD stock coolers use high RPM 70mm AVC fans that's what  Intel uses 92mm fans but the heatsink itself is anemic

Dave, just be careful, QS produces low quality videos but OpenCL (AMD) and CUDA (Nvidia) outputs decent quality videos
If you want a GPU to render for you choose a HD7850, it's got lots of computing horsepower being GCN and if you only have CUDA support it's time to switch to a OpenCL supported program, the speed difference from a GTX650 Ti (Priced similarly) and a 7850 is dramatic

[peter.mitchell]

If you want a GPU to render for you choose a HD7850, it's got lots of computing horsepower being GCN and if you only have CUDA support it's time to switch to a OpenCL supported program, the speed difference from a GTX650 Ti (Priced similarly) and a 7850 is dramatic

If you're not concerned too much about power usage, http://en.wikipedia.org/wiki/Comparison_of_AMD_graphics_processing_units#Evergreen_.28HD_5xxx.29_Series the 5870 is better than the 7850 by a fair margin in rated maximum performance, however, if the GPU isn't under full load than the difference may be negligable.

[T4P]

If you want a GPU to render for you choose a HD7850, it's got lots of computing horsepower being GCN and if you only have CUDA support it's time to switch to a OpenCL supported program, the speed difference from a GTX650 Ti (Priced similarly) and a 7850 is dramatic

If you're not concerned too much about power usage, http://en.wikipedia.org/wiki/Comparison_of_AMD_graphics_processing_units#Evergreen_.28HD_5xxx.29_Series the 5870 is better than the 7850 by a fair margin in rated maximum performance, however, if the GPU isn't under full load than the difference may be negligable.

The only thing is that GCN is alot better because it's very efficient at using all the compute horsepower, VLIW5 always had 1 idling section no matter what you were pushing it at therefore it's true compute performance is 4/5

[4to20Milliamps]

I was curious about whether or not my video card (nvidia gtx430, 96 cuda cores, 2gigs ram) would use the cuda cores and speed up rendering.....definitely not.

I used media coder, then handbrake with x264<2 processors only,  and cuda<96 more?

Same results, cuda was a bit slower by a few seconds, so either Nvidia is blowing smoke up everyone's ass or the software isn't using it properly.

[T4P]

I was curious about whether or not my video card (nvidia gtx430, 96 cuda cores, 2gigs ram) would use the cuda cores and speed up rendering.....definitely not.

I used media coder, then handbrake with x264<2 processors only,  and cuda<96 more?

Same results, cuda was a bit slower by a few seconds, so either Nvidia is blowing smoke up everyone's ass or the software isn't using it properly.

I'll say i'll point to your card. What's your CPU? I won't be entirely surprised if your CPU is more powerful than your GPU 
And yes nvidia might also be blowing smoke up everyone's ass
But also GT430 with it's tiny 268.8 Gflops is not possibly what anyone uses for rendering ... and also not optimized. A

[4to20Milliamps]

I was curious about whether or not my video card (nvidia gtx430, 96 cuda cores, 2gigs ram) would use the cuda cores and speed up rendering.....definitely not.

I used media coder, then handbrake with x264<2 processors only,  and cuda<96 more?

Same results, cuda was a bit slower by a few seconds, so either Nvidia is blowing smoke up everyone's ass or the software isn't using it properly.

I'll say i'll point to your card. What's your CPU? I won't be entirely surprised if your CPU is more powerful than your GPU 
And yes nvidia might also be blowing smoke up everyone's ass
But also GT430 with it's tiny 268.8 Gflops is not possibly what anyone uses for rendering ... and also not optimized. A

You're joking right?

It doesn't matter, if I use the cuda cores.......they should help out with rendering.

The fact is I'm not the only person that doesn't see much of an improvement, and certainly not enough to justify spending 700 bucks on a video card that doesn't do what it claims to do.

I'm the patient type so I would just wait for the video to encode, but I can see where this could be a real pain with lots of videos.

[mariush]

You're joking right?

It doesn't matter, if I use the cuda cores.......they should help out with rendering.

The fact is I'm not the only person that doesn't see much of an improvement, and certainly not enough to justify spending 700 bucks on a video card that doesn't do what it claims to do.

Actually it does matter.  The gpu clock frequency, the number of cuda cores, the memory throughput, all matter. Data has to be uploaded to the video card, then code runs in parallel on cores, then data has to be retrieved from video card...
If the memory bandwidth is too small or the cuda cores are too weak, the total upload to card + download from card + process time on cores can be more than just processing everything on cpu.

And a lot of stuff that's done in Cuda/OpenCL is actually using dumber algorithms that just take a brute force approach to something but are more suitable for working with lots of cuda cores (splitting image in lots of small chunks and parallelizing stuff), so they actually do more work than the version of algorithm used on cpu.  If the cores are weak, of course it will take more time.

[4to20Milliamps]

I certainly know how cuda works......what I'm saying is it doesn't work.....at all.

If I encode video with just my processor, then I try it with cuda, there is no appreciable difference.

Why would I buy a high dollar video card when I can just spread the task out to several processors? I'll just buy a top of the line xeon and a 40 dollar video card.

[hans]

you need to add some extra voltage if you boost up to that clockrate, and btw that board vrm are not meant to be used for overclocking, even with heatsinks attached to other boards some peeps manage to fry the mosfets or the driver chips

Based on tha fact that tha i5-2500 cpu is a 95w part, I would imagine all LGA1155 boards have a vrm design of at least 100w, i7-3770k is a 77w part, should be a little overclocking headroom on all boards, and a 4.6ghz overclock on that cpu would work out right around tha 100w mark i think.. Now, if u start overvolt/current other chips/parts on tha boards u could be in trouble, if tha either tha chips cant handle it, or tha board cant provide tha power your asking for them, a more expensive board would take those things into account.. If u were going for a extreme overclock, more then 4.6ghz, like 5.2+ghz on phase change or something like that, u would then want a highend board forsure I know thats not allways tha case, like tha AMD problem where tha entry level boards didnt have enuf vrm to handle tha high end cpus, but intel runs a little tighter spec and im sure all 1155 boards are 100w or more.. http://news.cnet.com/8301-10784_3-9931070-7.html

Now my i7-920 overlcock on LGA1366 is a different story, I never had no problems with tha P6T-Deluxe board, and I took tha 130w chip to around 170w.. Im sure some boards where only designed with 150w in mind, and some people fried things, like pads on there chips, when some of tha sockets werent design for more then 150w current draw and/or didnt connect properly to provide more then 130w draw

According to this, if doing extreme, it looks like its still possible to fry your chip/socket even on LGA1155 http://www.techpowerup.com/138604/Socket-Pin-Burnout-Returns-to-Haunt-LGA1155-.html?cp=3

I used to keep a REALLY tight ship on hardware, its kinda my thing, but these days I miss all kinds of things, as I dont pay attention to it as much or work with it as much.. However I had x58 then I skipped x79, I am thinking of possibly building a new system when this stuff comes out soon, even thou I dont need it, thinking a 4930 on z99 with dual or triple GTX770 and my first water cooled computer not a loop cooler.. I dont even game much, but I like to have tha latest hardware, not like I even needed to go from a i7-920/GTX570 system to i7-3770k/GTX680 either to be honest

The Ivy Bridge CPU's run hotter than Sandy Bridge when overclocked, due to the smaller die which has less surface area to evenly spread the heat.
So in that sense, core temperatures increase faster at the same overclock..

However, with good ventilation I don't see how a small overclock can hurt. I wouldn't be jumping at raising voltages straight away. In my experience, my i5 runs 10-15% faster without raising any voltages, nor having anything extreme core temperatures on full load. But it does yield 10% performance improvement. It basically saved me buying the i7 in single core performance..

It's true though that the low-end boards, like the Asus LX version only boasts with "4+1+1" phase design. However the Sandy Bridge consume quite a bit more power , so I do believe there is a little headroom.

[EEVblog]

Dave, just be careful, QS produces low quality videos but OpenCL (AMD) and CUDA (Nvidia) outputs decent quality videos
If you want a GPU to render for you choose a HD7850, it's got lots of computing horsepower being GCN and if you only have CUDA support it's time to switch to a OpenCL supported program, the speed difference from a GTX650 Ti (Priced similarly) and a 7850 is dramatic

My current workflow is XDCAM output from Sony MovieStudio and Handbrake, neither of which support OpenCL, CUDA, or QSV for encoding acceleration.
So any arguments over GPU acceleration are not relevant at present.

[EEVblog]

If I encode video with just my processor, then I try it with cuda, there is no appreciable difference.

That has been my experience with the Sony encoders that support both CUDA and OpenCL as well. In some cases it is actually slower to enable GPU acceleration.

[ecat]

If I encode video with just my processor, then I try it with cuda, there is no appreciable difference.

Then I guess either MediaCoder's support for cuda is poor or cuda has poor support for the required algorithms or you simply have a bloody fast cpu

A quick test of MediaCoder - taking care to avoid accidentally installing RealMedia or whatever rubbish comes bundled in the download - gives me CPU only 1:20 on Daves test video with options set to high (1.2Mbps is as good as it gets) and 0:49 using Intel gpu. So again, there is a point where gpu acceleration is worthwhile, unfortunately I cannot test cuda

That has been my experience with the Sony encoders that support both CUDA and OpenCL as well. In some cases it is actually slower to enable GPU acceleration.

Just ran DaveTest through XDCAM HQ 1440x1080-24p, 35 Mbps VBR , 2:05.

If you would be so kind as to provide DaveTest numbers for XDCAM, Sony AVC/MVC with and without iGPU (high quality), and MainConcept AVC/ACC with and without OpenCL, then we can add the points to the graph, ponder our bellybuttons and get a good nights sleep

[Toque]

I decided to give it a try, I downloaded handbrake, dropped tha file in, set it to CF=22, took 40 seconds here.. Now that was tha EEVblogTestRender2Min22.MTS file, no other settings changed, no script, and I am missing a AVC step in there to have tha right source file?

[ptricks]

Same results, cuda was a bit slower by a few seconds, so either Nvidia is blowing smoke up everyone's ass or the software isn't using it properly.

Cuda can be faster, but for it to be faster it has to have everything in the pipeline just right. I do 3d animation as a hobby and have used the program 3d studio max since its dos version and they have added gpu rendering in the last couple versions. The rendering using the gpu can be faster but it requires compatible textures, compatible options, etc. GPU computing can be faster  but only under strict conditions where the software and hardware were well thought out, it is far from 'plug and play' computing.

[T4P]

I certainly know how cuda works......what I'm saying is it doesn't work.....at all.

If I encode video with just my processor, then I try it with cuda, there is no appreciable difference.

Why would I buy a high dollar video card when I can just spread the task out to several processors? I'll just buy a top of the line xeon and a 40 dollar video card.

I have a Nehalem Xeon and it's not what it seems ... my GPU is WAYYY faster than my Xeon ever is (even OC'd)

The Ivy Bridge CPU's run hotter than Sandy Bridge when overclocked, due to the smaller die which has less surface area to evenly spread the heat.
So in that sense, core temperatures increase faster at the same overclock..

However, with good ventilation I don't see how a small overclock can hurt. I wouldn't be jumping at raising voltages straight away. In my experience, my i5 runs 10-15% faster without raising any voltages, nor having anything extreme core temperatures on full load. But it does yield 10% performance improvement. It basically saved me buying the i7 in single core performance..

It's true though that the low-end boards, like the Asus LX version only boasts with "4+1+1" phase design. However the Sandy Bridge consume quite a bit more power , so I do believe there is a little headroom.

Nah, IVB is not soldered to the IHS but rather using X23-7783D thermal paste. NOT GOOD AT ALL.

[M. András]

I was curious about whether or not my video card (nvidia gtx430, 96 cuda cores, 2gigs ram) would use the cuda cores and speed up rendering.....definitely not.

I used media coder, then handbrake with x264<2 processors only,  and cuda<96 more?

Same results, cuda was a bit slower by a few seconds, so either Nvidia is blowing smoke up everyone's ass or the software isn't using it properly.

I'll say i'll point to your card. What's your CPU? I won't be entirely surprised if your CPU is more powerful than your GPU 
And yes nvidia might also be blowing smoke up everyone's ass
But also GT430 with it's tiny 268.8 Gflops is not possibly what anyone uses for rendering ... and also not optimized. A

thats a lot of computing power it its used correctly in software, i measured my 8350 on stock settings around 120Gflops, the 6970gpu is 2.7Tflops.
im wondering what can be the new i7 processors computing power  in the same league.
i used sisoftware ansdra 2013 for this, it had a few intel processor on the results but not so far away from this amd

[4to20Milliamps]

Don't get me wrong, the nvidia card is an awesome card for what I bought it for which certainly wasn't for a high end video workstation, the difference is night and day from the stock motherboard graphics, you definitely want to have a good graphics card.... no question about it.

The computer I'm on is not a video computer at all, two core 2 gig processor and 3 gigs of ram, I can encode Dave's video in 1:44 which is slightly better than real time even with this old clunker, the coreI7 should do it in seconds considering the processing power,  so the bottleneck has to be in the file transfer and not from the hard drives it almost has to be the way the software is handling it.

I had cyberlink media espresso from when I bought my bluray drive, it's "optimized" for cuda and it is about 30% faster than the other encoders, but there's no way to change the setttings or even see what's set different, it does output pretty good video though.

The rendering farm from that university said you could encode an hour of video in 14 minutes, what exactly could the difference be? especially if you are able to use the cuda cores in a similar manner. That coreI7 and the high end graphics card should match that time at least....I would think.

[T4P]

OpenCL. The way forward with the massive crunching power of GCN, but be reminded that  Nvidia nerfs the cuda computing power on the desktop parts so if you want to see the true potential you would have to buy a quadro. AMD doesn't do that shit 

A easy way is to see how many hash/s your CPU can do while mining bitcoins
My L5520 does 3MH/s with -v on (Does boost from 2.2 to 3 ...) in GUIMiner and my GPU ... wait for it, 342 MH/s on my 7850 OC'd to 1170/1320 Can't do anymore RAM or i would BSOD 
Mind you if you that doesn't seem like much some of you know it's just as powerful as the 6950 at stock ... and at OC'd it is unbelievable
And as for anyone who plans to buy Kepler 1 (6xx) or 2 (700 series) for video editing, i'll tell you not. It's terrible due to the sheer lack of single compute or double compute
I don't talk about gaming performance because this is freaking eevblog not OCN 

[M. András]

completly off topic but if you mentioned how does that bitcion mining work? always runs the stuffs at full load etc?

[T4P]

completly off topic but if you mentioned how does that bitcion mining work? always runs the stuffs at full load etc?

Yes, it does. Just remember that to get decent mining speeds your CPU must be at least free for GPU mining but no vice versa
Decent hashrates can be gotten using latest hardware but if you want ultimate hashspeed use fedora or another linux distro
but also want to try hashing with a intel iGPU? forget it ... HD4000 still isn't anywhere near HD6550D and only netting 17-18 vs 60 for a i7-3770 @ 4.4GHz on Win7