Threaded Mode | Linear Mode

ZiNgA BuRgA · 18/02/2009 12:33 AM

Vegetano1 Wrote: ehhh so basicly your saying,.. it is a "wee don't have the application today issue",. >!?>? because its difficult but not imposible,. so these program will appear,. ;)

No, many programs will still be serial. Majority of problems are executed serially.

Vegetano1 Wrote:ATI,.. seems to be 1 step behind nvidia,.. or maybe its nvidia exclusive,. why should ATI have to suport a CUDA if nvidia invented CUDA,.

ATI has their own GPGPU thing, FireStream, it's just not as publicised out as CUDA.

Assassinator · 18/02/2009 12:53 AM

Vegetano1 Wrote:- nvidia claiming video processing 20x faster with gpu then cpu is not true,..
- programs using graphics abbility's like with CUDA is't worth it,.!? (eg in CS4)
http://www.nvidia.com/content/graphicspl...index.html

I never said NVidia's claim of 20x faster encoding was false, it most likely is true. I'm just saying it's probably not a very fair comparison, so I question how much it's actually worth. But the main problem is that there isn't any big encoders out there that currently support CUDA. So even if the GPU can fairly encode 20x faster, it doesn't really mean much because nothing can use it.

Vegetano1 Wrote:- future cpu's will take over gpu's job's>!? no more graphics cards,.>!?

The GPU will never get completely taken over by the CPU (at least not the kind of CPU as wee know it today). Graphics processing needs heavy parallel processing power. The CPU sucked at that, and that's why the GPU was created separate from the CPU in the first place.

Actually, Intel and AMD are trying to make sort of a CPU and GPU combined in one. Wee'll see how that turns out... But that's not "CPU taking over GPU", it's rather "CPU and GPU become one".

ZiNgA BuRgA Wrote:Anyway whot i do understand,. if there is a posibility that CUDA is worth it and encoding with GPU is faster,.. then buying a 295gtx(or any other faster x.x card) is worth it,. because you might not see the difference in game fps speeds (dif between 60 fps and 120 fps,. altough i doubt highest settings in crysis with 295gtx will give you 120 fps) but you will notice difference working with CUDA and GPU encoding times,.

I'm not really saying CUDA isn't worth it or anything. The main argument I'm trying to pull off is essentially, as quoted previously, "I have nothing against the card (x295), I just have something against the price. I mean sure the card is awesome, but the price is definitely not awesome." And the price pretty much kills all the awesomeness out of the card.

No, the highest settings in Crysis won't give you 60fps with a normal card, you probably can't even pull 30fps. But that's pretty much just Crysis, and like 95% of other games you probably can easily pull off 60fps with something like a 9800. I personally think it isn't really worth it paying the loads of extra money just to make a few games like Crysis run at a bit higher settings. But if you think it's worth it, go ahead and buy it.

feinicks Wrote:this is a good watch..

Source: Google Video

watch the 38 min version:

http://www.charlierose.com/view/interview/10060

Heh, my sister or someone else thought something retarded like "hey, wee have faster internet now, so I'm going to download as much as I can", and wee got ourselves capped. (Australian internet is poo poo).

Currently like dial-up speed. Pretty gay. Not going to even attempt to stream a 38minute long video.

Vegetano1 · 18/02/2009 02:56 AM

If you only bought the game Crysis and its the only game you play then you prob be better of buying 4x 295gtx graphics cards,.quad Sli ;p
Programmer's are't interested in making GPU video converter's because it got no potential.
CUDA, the ATI version is called firestream, did both nvidia & ATI got the same idee?>?
Gpu parallel processing is fast,. so why not come up with a parallel cpu,. >> amd's cpu&gpu in 1
I don't think the price of the 295gtx is awesome too,. i hope the price will be half of whot it is now over 6 month.
If i can convert a video to PSP format 20x faster then normal,. i be very happy. :) and it seems the program already excists? looking at the flash movie,.?

the i7 cpu seems like a good idee to buy but the 295gtx is to expensive for now,.. i thought it was worth the video proccesing speed & CUDA suported program's but there is not many software suport,. yet? (faster CUDA graphics card=faster video converting speeds)
maybe the "295gtx power" serves the Physx & 3d view right,.. with higher settings and all i mean,..

meanwhile prices are still the same,. Hihi

Assassinator · 18/02/2009 07:03 AM

Vegetano1 Wrote:Gpu parallel processing is fast,. so why not come up with a parallel cpu,.

Because most tasks (that the CPU does) can't be processed in parallel, at least not mass parallel.

Lets use Zinga's analogy. For a job like lets say wrapping 10000 presents, then yeah, obviously having 240 people working on it at the same time is MUCH faster than having 4 people. This is the kind of thing where GPU > CPU.

But lets say you want to find the square root of 909571093571092571095710971015. Having 240 people working on it won't be any faster than having 1 person work on it, because that calculation simply can't be split up properly so that 240 people can work on it at the same time. This is where CPU > GPU.

And so both the CPU and GPU will have their own uses. Having the CPU do lots of parallel is not very useful.

Vegetano1 Wrote:the i7 cpu seems like a good idee to buy but the 295gtx is to expensive for now,.. i thought it was worth the video proccesing speed & CUDA suported program's but there is not many software suport,. yet? (faster CUDA graphics card=faster video converting speeds)
maybe the "295gtx power" serves the Physx & 3d view right,.. with higher settings and all i mean,..

Oh, I just looked at your prices... It seems not right. The Corei7 920 shouldn't cost even close to 645 euro. Maybe you mean the 940 instead? If you really mean the 920, then you're getting ripped off. The 920 should only cost like 1/2 of that much.

Actually, I think Corei7 (mainly the 940. The 920 is expensive, but not that bad) is also way too expensive. Not just the chip, but all the stuff that support Corei7 are also extremely expensive. Like a Corei7 motherboard costs like twice as much as a normal one.

If you're mainly a gamer (I assume you are?), the CPU won't even give you that much of a performance boost. Gaming is mostly GPU dependent. So if you have to get one of them, you would probably rather pay for the 295GTX than the Corei7.

I would wait on both. But hey, my opinion isn't the 10 commandments, so if you're happy with paying for the GTX295 and Corei7, then ignore me and go for it!

Oh, and if you plan on buying a Corei7 940, I strongly suggest you buy a 920 instead. It's like only a little bit weaker (if I remember correctly, only 0.3 GHz weaker), and is MUCH cheaper, like 1/2 the price. The 965 is a "testament to ripoffness", don't even look at the 965, it doesn't exist.

Vegetano1 Wrote:If i can convert a video to PSP format 20x faster then normal,. i be very happy. :) and it seems the program already excists? looking at the flash movie,.?

I'll go watch that flash video tomorrow at school then. (Can't do it at home).

Also, there's another side to encoding, that this discussion is pretty much completely ignoring. The level of compression in encoding is probably equally, or even more important than speed. Take file compression as an example. Winzip (.zip) format is way faster than 7zip (.7z) and winrar (.rar) in both compressing and extracting. But 7zip and winrar compresses better, so in the end, I still choose 7zip over winzip, even though it's slower.

So in video encoding's case, comparing different codecs, XviD encodes faster than H264, but H264 compresses better. Within the same codec, H.264 using the highest settings (for x264, 16b and ref, adaptive b, highest subme and merange) can encode like 50 times slower than using the lowest settings, but compresses a lot better.

So the essence of what I'm saying is that currently, x264 is currently the best encoder for encoding into H.264. And in order to defeat it, you will need to beat it in more than speed alone. That's what I meant by nVidia's claim of 20x faster being not fair. I mean winzip is probably more than 5x faster than 7zip. But does that necessarily mean winzip > 7zip? Not really (= no)...

by the way... in case you're getting confused:
H.264 = the video format
x264 = an extremely popular encoder that encodes H.264 video. Frontends like XviD4PSP and MeGUI all use x264.

Vegetano1 · 18/02/2009 07:19 AM

- that's a very good example! so maybe some processes video encoding can't be done by GPU?

- no its the price of an upgrade packages,. motherboard/3gig ram and i7 cpu,. ;p

Intel Core™ i7 Quad Processor i7-920 4,8GT/sec, 8MB, Boxed

Intel DX58SO, X58, Socket-1366, ATX, DDR3, QPI, GbLAN, 3xPCI-Ex(2.0)x16

Corsair TWIN3X 1333MHz DDR3, 3GB, Kit w/3x 1GB XMS3 modules, CL9-9-9-24

- in the flash clip i can see the converter uses H.264 but i guess its not using the x264,.?
dunno if you can see this image,. but if you look closely you see H.264
[Image: ej79tk.jpg]

Assassinator · 18/02/2009 07:48 AM

Vegetano1 Wrote:- that's a very good example! so maybe some processes video encoding can't be done by GPU?

I think some parts of encoding can be made parallel, and some parts can't.

So you CAN make it parallel, but it's probably very hard, and it may not work that well, because the parts that can be made parallel will probably be bottle necked by the parts that can't.

Vegetano1 Wrote:- no its the price of an upgrade packages,. motherboard/3gig ram and i7 cpu,. ;p

Intel Core™ i7 Quad Processor i7-920 4,8GT/sec, 8MB, Boxed

Intel DX58SO, X58, Socket-1366, ATX, DDR3, QPI, GbLAN, 3xPCI-Ex(2.0)x16

Corsair TWIN3X 1333MHz DDR3, 3GB, Kit w/3x 1GB XMS3 modules, CL9-9-9-24

Ah, ok.

Vegetano1 Wrote:- in the flash clip i can see the converter uses H.264 but i guess its not using the x264,.?
dunno if you can see this image,. but if you look closely you see H.264

Don't know, it's most likely not, but something that's very important (in terms of speed) is the settings.

This

Code:

x264 --crf 10.0 --ref 16 --mixed-refs --no-fast-pskip --bframes 16 --b-adapt 2 --b-pyramid --weightb --direct auto --deblock -1:0 --subme 9 --trellis 2 --psy-rd 1.0:1.0 --partitions all  --8x8dct --me tesa --merange 64 --threads auto --thread-input --aq-strength 0.5 --progress --no-dct-decimate --no-psnr --no-ssim --output "output" "input"

is like 50 times slower than this

Code:

x264 --crf 40.0 --nf --no-cabac --subme 1 --no-chroma-me --partitions none --me dia --merange 4 --threads auto --thread-input --aq-mode 0 --progress --no-psnr --no-ssim --output "output" "input"

So you see, comparing speed directly without knowing what settings and everything is not very accurate...

Sorry I haven't watched the vid yet, (have to wait for tomorrow for that), so I can't make comments directly based on it.

Ok, going to sleep. 1:43AM.

feinicks · 18/02/2009 07:54 AM

Assassinator Wrote:
Vegetano1 Wrote:Gpu parallel processing is fast,. so why not come up with a parallel cpu,.

Because most tasks (that the CPU does) can't be processed in parallel, at least not mass parallel.

Lets use Zinga's analogy. For a job like lets say wrapping 10000 presents, then yeah, obviously having 240 people working on it at the same time is MUCH faster than having 4 people. This is the kind of thing where GPU > CPU.

But lets say you want to find the square root of 909571093571092571095710971015. Having 240 people working on it won't be any faster than having 1 person work on it, because that calculation simply can't be split up properly so that 240 people can work on it at the same time. This is where CPU > GPU.

Yes, but you can have 240 people solve 240 different problems increasing your efficiency by 240%. As I mentioned earlier, the computational capability of a GPU exceeds that of a CPU due to various factors; far more transistors, stream processing etc. Also, current day GPUs may not be programmable to that great extent, but it has started. Even CPUs weren't as flexible in their formative years, and they were intended to do what they do even now, even back then. GPU has been developed to boost graphical performance only. This has made it strictly adopt parallel process. Now, with focus shifting towards more practical application of the GPU's capabilities, its s slowly being imparted with that flexibility.

Assassinator Wrote:And so both the CPU and GPU will have their own uses. Having the CPU do lots of parallel is not very useful.

Even in the ideal computer, the microprocessor chip will have separate parts for CPU and GPU. They will never truly become one, an hence the more accurate observation would be to note which takes more importance in terms of general application processing, in the coming years. Maybe the CPU will be responsible for the basic non-intensive tasks like word processing, but rest everything (not only encoding and gaming, but more intensive computing as well) will be directed to the GPU.

Assassinator Wrote:
Vegetano1 Wrote:If i can convert a video to PSP format 20x faster then normal,. i be very happy. :) and it seems the program already excists? looking at the flash movie,.?

I'll go watch that flash video tomorrow at school then. (Can't do it at home).

Also, there's another side to encoding, that this discussion is pretty much completely ignoring. The level of compression in encoding is probably equally, or even more important than speed. Take file compression as an example. Winzip (.zip) format is way faster than 7zip (.7z) and winrar (.rar) in both compressing and extracting. But 7zip and winrar compresses better, so in the end, I still choose 7zip over winzip, even though it's slower.

So in video encoding's case, comparing different codecs, XviD encodes faster than H264, but H264 compresses better. Within the same codec, H.264 using the highest settings (for x264, 16b and ref, adaptive b, highest subme and merange) can encode like 50 times slower than using the lowest settings, but compresses a lot better.

So the essence of what I'm saying is that currently, x264 is currently the best encoder for encoding into H.264. And in order to defeat it, you will need to beat it in more than speed alone. That's what I meant by nVidia's claim of 20x faster being not fair. I mean winzip is probably more than 5x faster than 7zip. But does that necessarily mean winzip > 7zip? Not really (= no)...

by the way... in case you're getting confused:
H.264 = the video format
x264 = an extremely popular encoder that encodes H.264 video. Frontends like XviD4PSP and MeGUI all use x264.

However, if you have multiple threads encoding at the same time, then the time required is going to be less. I haven't read Nvidia's claim and will be able to comment better after understanding why they say its 20% faster. However, it has generally been seen that application (like Adobe Premier) which use CUDA or GPU based functionality, have shown tremendous increase in efficiency and giving better times.

Vegetano1 · 18/02/2009 08:04 AM

feinicks Wrote:I haven't read Nvidia's claim and will be able to comment better after understanding why they say its 20% faster.

not 20% but 20x faster ;)

Assassinator · 19/02/2009 02:43 AM

feinicks Wrote:Yes, but you can have 240 people solve 240 different problems increasing your efficiency by 240%.

Ok, go back and read every post Zinga and I wrote.

Which bit of "most things probably can't be effectively made massively parallel" is that hard to understand?

feinicks Wrote:As I mentioned earlier, the computational capability of a GPU exceeds that of a CPU due to various factors; far more transistors, stream processing etc.

How the hell do you compare "computational power" for 2 different things? I mean that's like comparing whether your heater is more powerful or your microwave.... (I mean like obviously my microwave is more powerful, right?)

Also, the number of transistors is NOT directly proportional to performance, which is what you have based your argument on many times. Lets use a real life example. The Core2 E6400 has less than 1/2 the amount of transistors as a Pentium D 900 CPU. Tell me the Pentium D is more twice as strong. This is especially the case if the 2 things in question are different. I mean wow, my microwave has more transistors than my heater, it has to be "stronger" amirite amirite amirite???

feinicks Wrote:Also, current day GPUs may not be programmable to that great extent, but it has started. Even CPUs weren't as flexible in their formative years, and they were intended to do what they do even now, even back then. GPU has been developed to boost graphical performance only. This has made it strictly adopt parallel process. Now, with focus shifting towards more practical application of the GPU's capabilities, its s slowly being imparted with that flexibility.

The GPU is specifically designed to process things like graphics, and that's where it draws its strength from. Ever notice that something specialized at doing a task is generally better than something general at doing that same task? Try to make a GPU become a CPU, you loose that specialization advantage, and you will probably end up on a long road that leads back to the CPU.

ZiNgA BuRgA · 19/02/2009 02:53 AM

Assassinator Wrote:
feinicks Wrote:Yes, but you can have 240 people solve 240 different problems increasing your efficiency by 240%.

Ok, go back and read every post Zinga and I wrote.

Which bit of "most things probably can't be effectively made massively parallel" is that hard to understand?

He's talking about running 240 serial tasks at once.
To be honest, it's one thing I've been interested in - ie running 20+ encodes simultaneously.
Unfortunately, at present, there's other limitations, such as disk seeks, and RAM usage.

It's also rare that you want to run 240 of the same task at once as well.

Threaded Mode \| Linear Mode whot do you guess??
Author	Message