ATI Stream: Finally, CUDA Has Competition

[blackened144]

[citation][nom]scrumworks[/nom]You sure know how to waste your time and Watts.[/citation]
Right.. because lending some CPU cycles to help figure out how proteins fold and finding cures to diseases that have plagued mankind for eternity is just totally useless.

 

[Spanky Deluxe]

[citation][nom]williamvw[/nom]4. Yes, I agree that, ultimately, OpenCL and DirectX 11 will lay the entire Stream/CUDA issue to rest. But that's someday. For now, this article's purpose was to take a look at today's technology.[/citation]

Oh I'm not complaining at all! These articles have been fantastic. These two articles and Fedy Abi-Chahla's original CUDA article are truly superb and go a long way to explain the nuances of GPGPU APIs. I look forward to the equivalent OpenCL articles that will likely arise in September/October.

 

[cyberlink]

[citation][nom]spanky deluxe[/nom]Stream and CUDA are likely to go the way of the dodo soon though. OpenCL's where its at. Unfortunately its a tad hard to get programming with it right now since you need to be a registered developer on nVidia's Early Access Program or you have to be a registered developer with Apple's developer program with access to pre-release copies of Snow Leopard.Virtually no one will bother using CUDA or Steam after OpenCL's out - why limit yourself to one hardware base after all? It'd be like writing Windows software that only ran on AMD processors and not Intel. Developers will not bother writing for both when they can just use one language that can run on both hardware platforms.[/citation]
As you can see, Cyberlink isn't limiting our software to one hardware base.

While OpenCL or DirectX Compute Shader promise to make life easier for developers, there are advantages and disadvantages to using these APIs instead of the graphics vendor's hardware platform specific APIs. The main issue today is whether an API is ready or not. In any case, it's a little bit misleading to say that OpenCL will make CUDA and Stream obsolete. OpenCL is just an abstraction layer API. CUDA and Stream, and other platform-specific GPGPU technologies to be named later are still the core technologies that will power OpenCL or DirectX Compute Shader.

 

[dingumf]

DAMN YOU PEOPLE ARE LATE.

THIS IS LIKE LAST YEAR'S NEWS FOR GOD's SAKE

 

[cangelini]

Actually, the 9.5 hotfix, which enabled this story, was released late May 2009. So yes, we're about two weeks late, but there was an absolute ton of back-and-forth with ATI and CyberLink in order to generate the information presented here. Thanks for the feedback, though!

Chris

 

[anamaniac]

Interesting article.

I would love more applications to take advantage of the increased processing power of the GPU.
However, I guess it's like the IBM PPC cell processor. Except even more challenging and even more rewarding.
Keep up the excellent articles,

 

[universalremonster]

It just so happens that I'm starting on ripping and encoding my entire dvd collection for my iPod touch, so I gave the trial version a try. I thought I would share with you my findings on 2 different systems, a C2D and a Core i7, using a 6.6GB VOB file. I am encoding it with the default Apple Devices settings for iPod touch, which is MPEG-4 @ 640x320 and AAC for audio. My main system specs are as follows:

Core i7 920 @ 3.33 Mhz w/Zalman 9700 cooler
6GB Corsair Platinum DDR3-1600 @ 1331 7-7-7-20
Asus P6T v2 Deluxe mobo
Sapphire Radeon 4870 1GB @ 810/1000 w/ Arctic Cooling Accelero Twin Turbo GPU Cooler
Antec Quattro 850w PSU
2x WD Cavier 160GB SATA II HD's Raid0
2x SATA Samsung DVD Burners
Antec 900 Case

With this setup WITH ATI Stream enabled it encoded in 16m27s. Core activity on core#2 was 68%-70% all 7 seven other threads hovered anywhere between 20%-25% (except for 1 that was 25%-35%) and GPU Activity as reported by gpu-z was between 10%-17% with a peak of 20%. The odd thing is it wasn't constant and every 4 or 5 seconds it would drop to 0% for 1 second throughout the entire process.

With ATI Stream disabled my time was 15m16s with core activity across all 8 threads evenly loaded and would fluctuate between 80-92% and all threads would intermittenly hit 95% or higher for one click on the activity graph.

So kinda interesting that Stream was a little slower than CPU only.

My 2nd system I have is a Dell T3400 Workstation that I got for next to nothing so I picked it up to give an HTPC a try for the fun of it. It is just the stock tower but has a different gpu in it than what shipped with it. Specs are as follows:
Core 2 Duo E6850 @ 3Ghz
Biostar Radeon 4650 @ 600/900
Intel X48 family mobo (Dell brand?)
4GB ECC-DDR2 @ 800Mhz
1x 160GB WD Cavier HD - system drive
1x 500GB WD Cavier HD
1x Generic DVD Burner
Generic Dell 375w PSU

With this setup and ATI Stream enabled the time to encode the file was 47m23s with 88% CPU activity and 12%-20% GPU activity with a peak of 22%. With Stream disabled the time was 1h20m23s at 100% CPU activity.

So on this setup there was a HUGE difference in encoding times with Stream enabled. If I didnt have everything all tucked away nicely I would try my 4870 inside the Dell to see how much a difference that would make but I'm just too lazy =P Anyways, just wanted to share some C2D to Ci7 results for those interested.

 

[dingumf]

ATI stream was released a quite a while back.

This is old news to me.

 

[Guest]

Excellent article.

Just wondering what a system with 7 dual GPU cards will be able to do in near future. I mean besides frying potatoes...

 

[Guest]

"These didn’t require regular driver updates as they operated in very stable environments, both for hardware (FireStream) and software."

Thank you ATI for your all time horrible drivers. I wonder what are for them "stable" environments. Had a 9800, x800, using for example DXVA was hard and sometimes impossible in standard scenarios and so much frustrating problems. And not even talking about Linux. No troll intended, no ATI hater, but this phrase got me on the roof.

 

[randomizer]

[citation][nom]dingumf[/nom]ATI stream was released a quite a while back.This is old news to me.[/citation]
If they tested it when it was new you'd say they were paid by NVIDIA to write the article because they'd basically only be testing CUDA. It's a shame people mindlessly bash articles when they don't have a clue what they are on about.

 

[cruiseoveride]

OpenCL FTW

 

[quantumrand]

I can't see how your score adds up. I've read through it a few times now, and it seems like you just randomly give points to nVidia. Whats the deal?

 

[amnotanoobie]

[citation][nom]wasteoftime[/nom]How come Folding is not something you guys ever cover in your CUDA comparisons. The main reason I left an ati card and bought an nvidia was the huge increase in my ppd.[/citation]

Unless the F@H team rewrite the cores for ATi cards, I would suggest you stick with your nvidia option. It's not that I'm bashing ati, but I've been folding with my 3870 and the problem is that it uses one cpu. With an nvidia option, or if the cores for ATi cards didn't hog a cpu, I could actually run 2 CPU clients and 1 GPU client. With the current state, I could only run 1 CPU client and 1 GPU client.

 

[scrumworks]

[citation][nom]blackened144[/nom]Right.. because lending some CPU cycles to help figure out how proteins fold and finding cures to diseases that have plagued mankind for eternity is just totally useless.[/citation]

Yes, yes, curing the diseases. Doctors and scientist can do that. BTW most diseases are self inflicted with unhealthy diet and poor lifestyle. So all those wasted billions dollars/euros caused by things like obesity and smoking could have been directed to somewhat more fruitful things.

What we really should be thinking is climate change. Excessive energy consumption leads to increased pollution levels and destroys atmosphere as we all know. Temperature change could eventually kill every living thing on earth. So think about that when you leave your 1000W PC crunching numbers 24/7.

 

[randomizer]

[citation][nom]scrumworks[/nom]What we really should be thinking is climate change. Excessive energy consumption leads to increased pollution levels and destroys atmosphere as we all know. Temperature change could eventually kill every living thing on earth.[/citation]
What press release did you read that from?

 

[morpheas768]

[citation][nom]scrumworks[/nom]Yes, yes, curing the diseases. Doctors and scientist can do that. BTW most diseases are self inflicted with unhealthy diet and poor lifestyle. So all those wasted billions dollars/euros caused by things like obesity and smoking could have been directed to somewhat more fruitful things.[/citation]
If you think that Alzheimer and Parkinson diseases are self inflicted then you are badly mistaken.
[citation][nom]scrumworks[/nom]What we really should be thinking is climate change. Excessive energy consumption leads to increased pollution levels and destroys atmosphere as we all know. Temperature change could eventually kill every living thing on earth. So think about that when you leave your 1000W PC crunching numbers 24/7.[/citation]
However I agree with your opinion about all this energy goes wasted and
temperature of the earth climbing. We should be a little more conservative about these things.

 

[z999]

what's happening w/ xvid? that's the only thing I need from this kind of software. either x.264@mkv(unsupported)>xvid@avi(again unsupported) or xvid@avi>3gp@mp4 (or whatever the encoder is, which again-is unsupported).
I understand that xvid development is up to the developers of xvid but it's open source, they could've done it too. and again the 3gp format for mobile phone support.
it's no use to me till these options are out... so calculating buy current development...... wake me up in 2011-2012...

 

[computergeek1231]

LOL @ scumworks. Climate change is not going to kill everyone despite what sensationalized reports claim. Yes, we should monitor climate change and reduce etc. but what really will kill us is excessive emissions. In LA people have higher rates of lung cancer compared to similar demographics because of all the crap in the air. If you're worried about a 400W full load PC ( Look up gaming machine power consumption at load, it isn't anywhere near 1000 watts unless you're running quad SLI etc. ) think again. Your car is around 149,000 watts ( 200HP ) and while you don't use that all the time, running a car for an hour on the freeway is almost as much as my PC running 24/7 for a year ( and it doesn't, only when it's already on ). F@H is a great project that tries to address immediate concerns. Climate change is projecting a whopping 2 inches or so of water increase in the next 20 years.

 

[quantumrand]

I have to admit, I've been waiting for ATI's answer to CUDA for a while, and I'm not very impressed. nVidia still has PhysX support which keeps it well ahead of ATI. The sad thing is that CUDA is an open platform, so there's really nothing stopping ATI from adding CUDA support.

It's sad that ATI has the more powerful cards at the cheaper price, but if I want to play around with PhysX or anything GPGPU related (other than transcoding), I'm stuck choosing nVidia.

 

[randomizer]

PhysX support is nothing but marketing. No good games use it, and the ones that do use it don't do much with it.

 

[quantumrand]

[citation][nom]randomizer[/nom]PhysX support is nothing but marketing. No good games use it, and the ones that do use it don't do much with it.[/citation]

Doesn't change the fact that essentially all GPGPU code is done with CUDA. ATI was a bit too late, and CUDA really did hit a homerun...I've always preferred ATI, but I see CUDA as a very strong selling point of an nVidia card...well worth the extra $10-$20 they charge for the similarly performing card.

 

[Narg]

Overkill, pure overkill...

A lot of time and money spent on some programming ideas that probably will be overtaken by 3rd party solutions in the near future. Cuda and Stream are a waist of time, and end up only raising the prices of good hardware.

 

[mospeada]

Expresso maybe using low performance codec for CPU transcoding so that it looks like the GPU transcoding have some benefits. Its a dirty marketing tactic. See http://www.pcgeek.my

 

[williamvw]

[citation][nom]scrumworks[/nom]Yes, yes, curing the diseases. Doctors and scientist can do that. BTW most diseases are self inflicted with unhealthy diet and poor lifestyle. So all those wasted billions dollars/euros caused by things like obesity and smoking could have been directed to somewhat more fruitful things. What we really should be thinking is climate change. Excessive energy consumption leads to increased pollution levels and destroys atmosphere as we all know. Temperature change could eventually kill every living thing on earth. So think about that when you leave your 1000W PC crunching numbers 24/7.[/citation]

How about this? I agree that when the Sun goes nova and the Earth is turned into a blackened biscuit, yes, temperature change will probably kill every living thing on the planet. But before that, why don't we keep rewarding vendors that design hardware ever more able to solve real-world problems through accelerated performance while simultaneously reducing power consumption? Or is it just more entertaining to see everything in black and white hyperbole?