Slashdot Mirror
AMD Breaks 1GHz GPU Barrier With Radeon HD 4890
MojoKid writes "AMD announced today that they can lay claim to the
world's first 1GHz graphics processor with their ATI Radeon HD 4890 GPU. There's been no formal announcement made about what partners will be selling the 1GHz variant, but AMD does note that Asus, Club 3D, Diamond Multimedia, Force3D, GECUBE,
Gigabyte, HIS, MSI, Palit Multimedia, PowerColor, SAPPHIRE, XFX and others are all aligning to release higher performance cards." The new card, says AMD, delivers 1.6 TeraFLOPs of compute power.
AMD Breaks 1GHz GPU Barrier
I was diligently working at XYZ Corp a few buildings down when Incident One happened in their lab. At first, I was just sitting in my cubicle when suddenly we felt a severe shuddering of space & time around us. Then a few seconds later everyone heard a loud "Ka-BOOM" and everyone stood up to see what was going on outside. The buildings directly adjacent to the AMD lab had all their windows blown out and every car alarm within a square mile was going off. Some scientists with their hair blown straight back and carbon scoring randomly on their faces and white lab coats were seen to climb out of the rubble of AMD's R&D building. They immediately began dusting themselves off, high-fiving each other and patting each other on the back laughing and ecstatic. Then they headed towards the liqueur store down the street to pick up some champagne. Shortly after it was discovered that 1Ghz is the frequency at which æther vibrates when it is at rest so once you pass it, you leave a wake of æther behind your time cone. Roger Penrose and Stephen Hawking are due to give a speech at "GPU Ground Zero" this week, I hope to make it.
If I were working marketing for AMD, I would be pointing out how switching from base ten to base eleven, twelve, thirteen, etc provides a theoretically unlimited amount of newsworthy advertisements in broken barriers. "We just need to make it to 2,357,947,691 hertz and we'll be the first to claim we've broken the 1 Ghz (base11) barrier! Where the hell was the report that we broke base9 last year?!"
My work here is dung.
one will finally have a graphics card capable of playing Duke Nukem Forever.
Oh wait...
We will bankrupt ourselves in the vain search for absolute security. -- Dwight D. Eisenhower
Digital Broke that with the DEC Alpha (Was it DEC at that time?). Wasn't popular but it was a desktop CPU for high end workstations.
Why is it harder to raise the clock frequenceies on GPUs than CPUs? Is more code in use at the same time per unit area, or?
Sorry. It was Compaq who owned the Alpha at that time. It was still DEC who designed it though.
As you may have seen from the sales of netbooks and low-power computers, the future is... wait for it... low-power devices!
Where are the 5W GPUs? Does the nVidia 9400M require more than 5W?
No mention of power consumption or heat dissipation. My PC is already a radiator and in the summer fights with my AC.
I am interested in the computing power, 1.6 terraflops is no small number even if it is single precision.
According to intel it's about 0.04.
last time i checked, a graphics card will get about 100x more Flops than a similarly priced CPU, give or take an order of magnitude (hey, im an astrophycist, order of magnitude is good enough)
I can finally get a 5.0 on the Vista Experience Index!
www.purevolume.com/martyd
if you're imagining insane numbers of cards for silly TFlops, the 4TFlop nvidea Tesla has a 1U rack form, so you can shove as many of them as you like in a rack
The 1242 MHz speed is the frequency of vertex shaders, not the core speed. Also, 1 GHz is the core speed without overclocking.
How about the fact that it runs each instruction on 800 pieces of data at once? This isn't a 1 GHz one, two, four, or even 16-way chip. It's processing up to 800 pieces of data at once, and its clock for doing that ticks every billionth of a second. You're absolutely right, the clock speed by itself means nothing. The clock speed times the amount of work done per clock does mean something. If you raise either without lowering the other, you raise the overall amount of work the chip can do.
And it's still slower than a GTX 285 OC edition. Ghz != Preformance. And Nvidia, stop renaming your cards damn it!
AMD does note that Asus, Club 3D, Diamond Multimedia, Force3D, GECUBE, Gigabyte, HIS, MSI, Palit Multimedia, PowerColor, SAPPHIRE, XFX and others are all aligning to release higher performance cards."
Wait, let me get this straight. Graphics card manufacturers are actually attempting to make their graphics cards perform better? Why was I not informed of this before???
You keep using that word. I do not think it means what you think it means.
Also, 1 GHz is the core speed without overclocking.
False. It's overclocked alright, it just doesn't have to be overclocked by users or the third party manufacturers to run at 1 ghz. From their press release:
Nine years after launching the world's first 1 GHz CPU, AMD is again first to break the gigahertz barrier with the factory overclocked, air-cooled ATI Radeon(TM) HD 4890 GPU -
I'm pretty sure that word doesn't mean what you think it means. "Overclocked" means "our reliability people don't think this is smart, but it might work for you." In this case, you get a part that may or may not die before you expect it to, it might not last much beyond the warranty, it might have non-standard cooling to enable an operating window that Reliability can't assume (say they model frequency shifting at 85C and they have a heat sink that puts it at 55C; feel free to substitute any other numbers).
Any conditions where the company's Reliability department didn't endorse frequency over the lifetime of the product for 3 sigma worth of sellable parts would be "overclocked".
Modern GPUs including every single Nvidia GPU since the G80 series has had a full integer instruction set capable of doing integer arithmetic and bit operations.
CPUs aren't designed to be good at everything, they're designed to be exceedingly good at executing bad code, which is the vast majority of code written by poor programmers or in high level languages.
You can write code for a CPU without worrying specifically about the cache line size, cache coherency, register usage, memory access address patterns and alignment or memory latency on branches or pipeline stalls and the difference in performance compared to optimized code will significant but not unbearable.
GPUs devote significantly less (or in some cases no) die space to things like branch prediction and automatically managed caches. Poorly written GPU code is sometimes almost 2 orders of magnitude slower than well written GPU code, but well written GPU code has much higher potential than what is achievable on modern CPUs. See: CUDA.
Actually as a PC repairman I can tell you the "trick" with AMD, and it is this- always buy a generation or two behind. I have sold many ATI and Nvidia cards as well as AMD PCs with ATI chipsets, and as long as you stay a generation or two behind you're good to go. My dual core Kuma with 780V chipset is solid as a rock
So what I tell my customers is this: If you want to spend top dollar and be on the bleeding edge, go Nvidia. Their drivers will be rock solid even for the card they just released. With AMD/ATI always get a generation or two behind and NEVER upgrade the drivers! Unlike Nvidia whose drivers are pretty painless to upgrade, upgrading to the latest Catalyst drivers usually end up bring nothing but instability and headaches. Now I don't know if this "trick" work with Linux, as I'm a Windows only shop. But I have found in Windows if you follow this rule you'll be good to go and save a few bucks as well. The "bang for the buck" ratio is very good on AMD/ATI which is why I just built my first AMD PC since the old Barton Core. You just have to be careful not to get too close to the bleeding edge with ATI, as out of the box their new drivers always suck.
ACs don't waste your time replying, your posts are never seen by me.
Everyone thought it would be 999MHz this year, 999.9 MHz the next year, 999.99999 MHz a few years later. It looked uncrossable! Well done, AMD!
"Believe me!" -- Donald Trump
There's a difference in clientele and expectations. Most people want a part to last for 5 years and be qualified for 85C. That's one speed sort. Some may provide better cooling and keep it under 55C and only expect 3 years. Depending on the technology, that's a different sort entirely. I'm not disagreeing that it's a marketing thing, but this is not the same QA standards.