It's a given that the
GPGPU (or General-Purpose Graphics Processing Unit) has a long, long ways to go before it can make a dent in the mainstream market, but given that ATI was talking up
Stream nearly
three whole years ago, we'd say a battle royale between it and its biggest rival was definitely in order. As such, the benchmarking gurus over at
PC Perspective saw fit to pit ATI's Stream and NVIDIA's
CUDA technologies against one another in a knock-down-drag-out for the ages, essentially looking to see which system took the most strain away from the CPU during video encoding and which produced more visually appealing results. We won't bother getting into the nitty-gritty (that's what the read link is for), but we will say this: in testing, ATI's contraption managed to relieve the most stress from the CPU, though NVIDIA's alternative seemed to pump out the highest quality materials. In other words, you can't win for losin'.
posted Aug 10th 2009 8:57AM
Nvidia FTW
this is all very confusing. the gpu was made to offload work from the cpu.
whats next? another card to offload work from the gpu?
oh wait...i think someone made that already
A T I FOR THE LULZZ
NVIDIA WTF
What I want to know is, why can't they shake hands, sit down together, and put together a joint, open implementation of GPU processing architecture and instructions so that we can get OSes and software built around one strong solution instead of two lackluster implementations. Open standards always result in better technology.
it's a bit sad the only thing they base the summary on is transcoding test... which can be influenced from different params..
@Nickesh... that's what OpenCL is all about and it seems that A LOT of manufacturers are supporting it (AMD/ATI, nVidia, IBM, ARM, Sun, Intel, Nokia, Texas Instruments, Ericsson, Samsung, Sony... http://www.khronos.org/about/ )
Though after I found out the members of the Khronos group, I was expecting OpenGL 3.0 to take over the world and assassinate DirectX. That didn't come close to happening.
I hope OpenCL makes progress, I think AMD has a lot riding on that.
Yeah, unfortunately as much as I love ATI and think they have a far superior product, they BADLY need to work on their encoding drivers. I've seen video encoding tests on three different major websites, and all three reported bad-to-terrible quality using ATI cards, with Nvidia cards just barely worse than the CPU encoding.
ATI does need to get the encoding issues worked out in the short term, But In the future I don't think it will matter if everything is written to OpenCL /DirectX11 compute shader spec. Cards from either company will have the same instructions.
Also, while this may change with Nvidia's rumored next-gen GPU with "MIMD" architecture, ATI's 48xx series blows away Nvidia's GT2xx in terms of double precision floating point. While this may not be important for consumer GPGPU applications, it should definitely help ATI make inroads into the scientific/engineering market assuming they can put together a marketing team. Nvidia's "CUDA" marketing has definitely been far larger and better organized, particularly for professional applications.
ATI FTW
Well, At least It's a lose lose? Effing Nvidia... )-;
I wager twenty Quatloos on the newcomer...
ATI FTW
Nvidia ft....err looks like i got a faulty chipset
lol me too
F nvidia
Seconded on the faulty nVidia card. ATI > nVidia
ATI FTW
I agree with the second conclusion. When your presenting a project, you want the best image quality. So 99% of the time you will be using nvidia's chip. The time difference for the videos were small compared to the better output produced by Nvidia. This seems like a clear win.
ATI FTW!
Shouldn't that be, "you can't lose for losin'?" since either of the technologies work out, but just in different ways?
Nvidia for quality, so ATI loses - but...
ATI for speed, so Nvidia loses - but...
to each his own, if you want speed then ATI ftw, and if you want quality then Nvidia ftw irrespective of who "lost".
what? i think it's the other way around. have you seen how many faulty nVidia chips there are compared to Ati's?
They say in the review that ATI also used 20% less CPU processing.
Nobody won or lost because this compares software, not hardware. This is like comparing halo with killzone to decide which console is faster. Run these benchmarks when they make an encoder using openCl and the resulting images are identical.
I hate fanboys, nobody "won" you twits. One had a good advantage that the other didnt have. Quality vs. Speed. If it were me, quality.
I wish ATI would really work on improving their linux drivers, so I could build an extremely cheap and capable boxee system with their CPU + Motherboard + GPU combos. The integrated 4XXX in their mobo's playback 1080p blu-rays without hesitation and their newly released $40 dollar semprons can be unlocked to be dual core's and OC'ed to crazy high levels without much thermal increase. Talking a sub $300 dollar system built on newegg that could run pretty much everything except the most demanding of applications. Guess I'll just use windows 7 (stripped and ripped) and throw a tuner in there. I really wish amd/ati would use their experience producing all-in-wonder cards and build a motherboard that had a tuner and gpu integrated for under $150 in micro atx format. How amazing would that be for an HTPC?
I second the notion that the previous All-in-Wonder line should have yielded some ridiculous HTPC options by now. Wtf is wrong with them?
Weird. This is exactly the opposite result I expected. ATI has always been known for better colors and overall image quality while Nvidia was the framerate king. I guess when you turn the vidcard into a CPU all bets are off.
Things changed some years ago - when ATI started seriously competing with the nVidia.
Results were not surprising: framerate went up - quality down.
For gamers framerate trumps quality.
If I'm doing video conversion I don't want my CPU to have some slack. I paid for a quad, I want it pumping out 100%, and if I can get my GPU to join in too then all the better. Work those chips, no pain no gain!
So Nvidia for me if it get the best results.
These GPGPUs are for more than just consumer applications (video/audo encoding, PhysX, etc.)... what really matters is whether or not you get these successfully installed in a high performance computer for scientific/financial applications.
nvidia and ati don't need these type of features to sell more in the consumer marketplace... they need them to tap into an entirely different marketplace, HPCs.
Agreed, the GPGPU movement is more than graphical outputs for home users, it's about the computational power these cards afford for accelerating general mathematics (hence the first "G"). Computations that used to be possible only on supercomputers or highly-specialized hardware is now "easy" on GPUs. The catch is that us programmers need to be able to access that compute functionality -- in which case, I'm a big fan of NVidia for their C-based API, GPU libraries, support of research into GPUs as computational platforms, and educational efforts.
The bottom line: even if NVidia *might* be (slightly) slower at certain operations, the computations are going to get done much sooner than on my CPU. ATI isn't going to get anything done if I can't program it easily, so any (perceived) speed benefits are moot.
Didn't AMD/ATI give up on stream and refocus all it's development and tweaking on OpenCL?
turn my headphones up!
What u guys didn`t say, is that Extreme Tech did almost the same thing a few weeks a go... http://www.extremetech.com/article2/0,2845,2348550,00.asp
Its actually one of their last articles...
I also noticed that PCPer left out the best GPGPU video encoder for CUDA - Elemental Technology's Badaboom, which should in theory be the main competitor for the ATi AVIVO Encoder.
I'm using Badaboom myself, and I love the thing.
ATI's stuff is very half-baked and requires the latest drivers, no separate download, but unfortunately their latest drivers are usually total crap and to hit a somewhat stable release takes many months, so yeah I guess OpenCL is what to wait for?
ATI likes to tell the press about bright futures but the PR guys and the developers and the management are 3 entities that have no internal communication really, and their commitment to anything they boast about is approximately 8 days.
Nvidia AND ATI FTL.
They both suck in their own ways, and both are great in their own ways.
I'd like to see another company personally.
Don't count out ATI on anything. Even though they are not pumping the hard numbers, they are developing their GPUs properly.
What does this mean?
ATI was early to adopt to the new GPU technologies based on their work with Microsoft and the XBox 360. (Just like NVidia did with the original XBox that gave NVidia their Geforce 5xxx to 7xxx architecture - as most of it was based on Microsoft engineering.)
As the world finally moves from XP and towards Win7, where DirectX 11 can FINALLY be fully turned on. (DirectX10 was to have most of DX11 features, but NVidia threw a fit because their 8xxx series cards would not have qualified, just as they still don't even manage DX10.1)
ATI is set to lead this generation with DX11 GPUs, and this is important for two reasons.
1) DX11 finally brings the gaming/DirectX API sets on the PC to the same level as the XBox 360. This will mean better engines that get rid of DX9 and can finally use all the DX10 and DX11 optimizations that make a large performance difference at lower levels.
2) DX11 extends the GP/GPU functionality that was introduced in DX10, making the ancilary CUDA technologies irrelevant, as game designers may be 'branching' to CUDA or Stream, but stay within the DX11/DX10 API sets that open up this technology.
3) GPU multi-tasking - Vista introduced a revolution in how GPU technology is handled, giving the OS control of the GPU and creating a pre-emptive multi-tasking GPU environment, thus letting multiple GPU based applications runs seamlessly at the same time (Look up WDDM GPU Scheduler).
The ATI implementation of the WDM 1.x that enables this is more advanced than the NVidia implementation, and has far less locks giving the OS the control it needs to manage tons of applications sending GPGPU operations in addition to 'graphic' 3D applicaitons running on screen with multiple other 3D applications.
The last point is of great note to anyone that understands what this means. Instead of relying on Application Yielding, the OS scheduler creates a pre-emtive multitasking model for the GPU and also extends this to better SMP on multiple GPU without the need for an OS agnostic technology like SLI.
This is also where the new Snow Leopard is having problems, as they are opening up more GPU operation to applications, but sadly, the OS has no control over properly scheduling these. So an errant application can lock the GPU in addition to the problem of when you have several applications using the 3D GPU functions for calculations and onscreen 3D applications, as they all go back to the 'cooperative multitasking' concepts but this time with the GPU.
-----
I like NVidia products and use them in a majority of systems, mainly because of the high end performance and their lead in the Mobile GPU world. (This laptop I am using is an 2005 model, and it has a Geforce 7850GTX that often still outperforms any NVidia 86xx or 96xx or lower class GPU.)
However, I also understand a bit about the technology of the GPUs and direction ATI has been setting themselves up for, and they have a better technology set with better Unified Shader and better hardware hand off for WDM for Win7, in addition to already behing a generation ahead of NVidia for DX11 technologies.
(ATI like NVidia got a lot of good engineer assistance from Microsoft on the XBox project, and it created the powerhouse NVidia like of GPUs we saw in the early 00s, and ATI should be able to continue this trend, especially when the ATI 'based' GPU in the XBox 360 is already doing DX11 level technology and bringing their desktop GPU offerings up to this feature set won't be hard.)
I'm not sure if anyone noticed, but the quality tests are all faulty. They haven't compared the exact frame in any of the tests, so the differences seen are irrelevant.
Engadget writers need to follow Darren's fine example and use the word "contraption" more often.
Haven't video cards always off-loaded the cpu's work in regards to video processing. My understanding is that CUDA and Stream were for offloading other (not necessarily video) work to the gpu, Hence the first GP in GPGPU. Applying the massively parallel architecture of a gpu to problems that are inherently parallel. I have a 4 GPU Tesla machine at school and we use it for scientific computing (mostly vector and parallel algorithms).
Wouldn't it be more appropriate to compare Stream and CUDA based on their available functions and scaled speed up in regards to the number of individual processors on the units. They are api's after all, it doesn't make sense to make them hardware specific.
An added note is that a colleague of a professor of mine who had been using gpu's for non-graphics purposes long before either cuda or stream said both architectures dramatically detracted from performance.
I'd bet most people in most applications wouldn't be able to see the difference between the performance "boost" and the better quality image.
How can NVidia be faster and ATI better image quality. If you played crysis on medium would the ati card look better??? Wont both be on medium so they both have the same image quality?
This article is stupid. Although CUDA and STREAM take graphics processing strain away from the CPU more, that is not this technologies greatest benefit....
The idea is that they will be used to process more than graphics. A graphics card processes 1080x1920 things in miliseconds all at once. Imagine having a 280-core cpu or more that can process 280 things at once (GTX 280 stream processors number I think). That is what this is. A CPU will always be needed to run the kernels, but a GPU will become the new multi-core eventually.