Eva Glass is an anagram of Las Vegas - Adamson Rust
|
|
|
AMD talks up the GPU
THE GPU'S DEATH is greatly exaggerated according to AMD.
Writing in his blog , the chip manufacturer's graphics products technical marketing director, Godfrey Cheng said that if you look at the discrete GPU market, it has just had its best quarter ever. The second quarter 2010 was also very good for AMD, taking over the discrete market leadership from Nvidia.
Apparently to work out if GPU's are going to die out, Cheng said that you need to look at the quantum level. We guess he means that the GPU is both potentially dead and alive at the same time, you really have to observe it and hope that potential cats do not get in the way.
He said that despite having the power budget for nearly unbounded performance, one of the bottlenecks for discrete GPUs is PCI Express, the interface to the system. In the case of graphics workloads, normally PCI Express does not present a constant bottleneck, however, everyone and their dog has heard of parallel computing. This is the case where there is a lot of traffic between the discrete GPU, CPU and main memory, and PCI Express becomes a liability in terms of bandwidth and latency.
"Many people do not realize this, but the "life" of a CPU is dreadful and boring. It exists primarily waiting for data. The use of discrete GPUs for parallel computing through PCI Express will not improve the 'quality of life' of the CPU. While some discrete GPUs will offer graphics and pure FLOP performance over an [Accelerated Processing Unit], the performance will be limited, in some cases, by the PCI Express interconnect," he said.
AMD Fusion APUs have not only been designed to offer great graphics performance, they also have been designed to offer great parallel compute performance, he said.
The fact that the CPU core resides next to the GPU core connected by a bus of mere nanometers, helps diminish the bandwidth and latency issues presented to parallel computing on a PCIE bus.
The design plan for successive generation of AMD APUs includes architectural innovation, as well as tighter and faster interconnects between the CPU cores and the GPU cores.
One goal is to advance the parallel compute capabilities without sacrificing x86 and graphics performance.
Cheng's theory is that the "best days of the GPU are ahead of it" and if anyone thinks that the GPU will die the answer is "hell no".
The future of GPU obviously lies in the continuous increase of parallel compute performance and integrating the best parts for the job.
He said, "One of the main reasons people want a discrete GPU is to improve gaming performance. This segment shows no signs of disappearing anytime soon and AMD intends to continue to provide enthusiasts with leading discrete graphics solutions." µ
Share this:
Share
Seeing is believing. GPU memory speed and memory access architecture is very different. But it opens up a lot of prospects if 'they' find a way around.
But I don't see this landing anywhere near the highend because of these constraints....
Oh yes, I see myself buying CPU+GPU in a socket *and* additional dedicated GPU cards until the sweet end of the current iteration of tech. ( PCI-e and all )
It would be to good; upgrading vidcard memory with DIMMS...
I love the GPU-inside-CPU talks, because they ignore a bloody important fact: whether you GPGPU or you are just gaming, MEMORY BANDWIDTH (sorry for the caps) is really one of the most important parameters. GPUs are fast to process data, they need that 128/256/512-bit but to their GDDR5 otherwise they get starved. Guess if there are enough pins on the CPU package for that? Share the main (super slow compared to GDDR5) RAM, and you're dead too (CPU vs GPU fight for the controller, yay!).
Fusion is great for low-end cheap stuff, it's good for math coprocessor on small datasets, but letting a GPU loose on a big set of data transferred over the arguably slower PCIE is the only way to get the advertised teraflops.
Go ahead, prove me wrong.
Current computers are centralised around the main processor. Future computers will have data and it’s movement the central focus. Processing will be moved to be embedded in memory modules that are networked together. OS structures are parts of a larger database. File system becomes a distributed database with indexes, records inside records, security, relationships and abstraction. File sharing, concurrency, versioning, cloning and archiving are merged into the master DBMS.
IO is the slowest part of the current system and the ALU/FPU of CPU’s are becoming a smaller portion of the total area. The largest part of modern CPU’s are the caches, branch prediction/history & memory/system IO. The speed of the CPU becomes less significant as the memory, disk IO, network IO latency and throughput have not scaled as quickly as CPU performance.
Like in previous years and posts, he is so full of himself he ignores the simple fact, PC gaming is dead and shrinking and with the launch of the next consoles, it will continue its slow march to obscurity...
Godfrey is justifying his paycheque, sad, maybe he made that stupid video they released on youtube with thwe SWAT team finding a Fermi... GO make better drivers bozo's... NV has you there...
We do need and want more powerful GPUs. That's not to say that we are asking for them to consume more power. The dye shrink is helping abit. It will come to a point where they are working on more efficient designs once they get all the functionality they need on the chip.
Regarding the PCI express issue... It would be cool if AMD creates a new interfaces that will get rid of this issue, or at least deminish it. They don't have to wait for intel, they didn'e for 32/64-bit hybrid processors. Nor did they for the memory controller in the processor package, and they are not for their new APU design. So if descrete GPUs need a better interface, why not give them one that is better. Or is something else already on its way...
<quote
He said that despite having the power budget for nearly unbounded performance, one of the bottlenecks for discrete GPUs is PCI Express, the interface to the system.
<unqupte
And he lied as a parrot!
As of now, neither AMD, no nVidia GPUs may utilise more thaninternal PCI-X bus thoughput (less than 1.2GB/s). Yes, AMD has some problems with CPU-PCIe bandwidth, but not Intel (up to 12.6GB/s one way on a single Nehalem processor - more than enough to support 10GPUs!).
So, the ebmed GPU is just an attempt to push the "big brother" off the market (at least for a while).
I think the embed GPU is not bad for a ultra-budget market, but waste of buiers money (same as SLI "sertification") on all others.
I have a feeling that quite soon it may be possible that the APU will end up like the FPU did for the 386 etc. They won't be a requirement at first but if done right the level of performance brought to the table will make them a must have. I would probably be more concerned about the adoption rate.
As for PowerVR, they can suck a nut until they come up with some decent Linux support. Lets not forget the evil that GMA500 has wrought on the Linux user.
From my understanding Intel's GPU is actually pretty good so to counter AMD is supposedly going to put in a rather beefy GPU this time around so we should see a half decent GPU built in soon enough.
Either way there are several major benefits to having a GPU on hand.
1- Video Encoding and Decoding.
2- Anti-Virus scanning the GPU is much better than a CPU here and can scan much faster.
3- We will finally see Physics in games if the onboard/onchip GPU can be leveraged for physics when you add a dedicated video card tot he mix. Only when its mainstream on the majority of PC's will be mainstream in games.
IOW, integrated graphics, even if of the on-die GPGPU variety, will continue to, relatively speaking, suck until the end of time
tell us something we don't already know
you(commenters) are all forgetting GPGPU. which is what this integration is all about.
with this AMD will be able to win quit a lot of benchmarks that have traditionally been intel's strong suite. chief among them rendering and encoding. and not by a few percent but often by 2 or more factors.
and for gaming with a discrete GPU in addition to this it will be like having a extra CPU dedicated just to physics and AI.
Aryan @ they are integrating a modern main-stream 480stream processor GPU, not a 80stream-processor IGM.
We dont need more and more powerful GPUs.
Thats the wrong way to go. It's not clever to have to keep upping the power requirements every generation.
What we need is better and more efficient rendering techniques.
Everything at the moment is designed around brute-force computing. We need more finesse and efficiency.
We need to look at techniques such as those developed by PowerVR (tile rendering).
Just throwing more power and bandwidth at the problem just shows a lack of vision and imagination.
Talking about serious Highend GPU applications for introduction of the bottom-of-the-barrel IGP.
Wake me up for a surprise the day that a GPU the size of a now, and then, regular highend system is integrated with the tiny CPU die....
Not gonna happen this decade. Heat, heat, heat. They would then need to create a highend CPU with a powerconsumption of ~500W
There has to be some point of diminishing returns, and integrating graphics next to the CPU probably is it. I suspect that the next standard after HDTV at 2MSs is about the best a human eyesight can appreciate, and on a 30" monitor.
At this point, discrete cards would only be needed for additional high definition screens or to to boost graphic performance for some future highly detailed game or CAD program.