Tenet insanabile multos scribendi cacoethes - Juvenal
|
|
|
Nvidia delays Fermi GPU chip
WORD ON THE STREETS of Taiwan is that Nvidia has pushed back release of its Fermi chip another whole quarter.
Fermi was supposed to come out in November, but the Green Goblin reportedly had trouble getting its design finalised and delayed it until the Consumer Electronics Show next month. Now Digitimes claims the chip will not see the light of day until March 2010. We assume the added delay was required to enable Nvidia's next über-GPU to run Duke Nukem Forever.
However Nvidia's mounting delays give ATI's "Evergreen" GPU family some three quarters of uncontested DX11 leadership, making it start to look like the Green Goblin is losing its grip.
Earlier the company stopped making chipsets for non-Atom x86 platforms and that left it with discrete GPUs and high-performance computing as the only reasons giving it meaning in life.
If Fermi does not happen then all the Green Goblin will be left with will be Tegra, which could help the outfit avoid sliding into obscurity next year. The ARM-based embedded and mobile processor looks like it might end up in Apple's tablet, which should help sales a bit, but Tegra alone isn't likely to sell in enough volume to reverse Nvidia's declining fortunes.
Nvidia's roadmap seems to be to launch a 40nm GDDR5 memory based Fermi GF100 GPU in March 2010, followed by a GF104 version in the second quarter to target the high-end market. However, by then AMD will have maintained a solid lead in the high-end discrete GPU market for six months or more.
The company apparently will set its GeForce GTS250, GT240/220 and 9800GT/9500GT against AMD's Radeon HD 5770/5750, 4870/4850 and 4670/4650 in the performance market.
In the mainstream segment the Green Goblin will peddle its GeForce 210, while AMD is apparently on track to launch its 40nm Radeon HD 5670/5570 "Redwood" and HD 5450 "Cedar" GPUs in January or February.
Both AMD and Nvidia might run into problems meeting their forecast schedules due to their dependence on TSMC. The Taiwanese chip foundry's 40nm process yields have improved, but its production capacity is still not high enough to fully satisfy the demands of both GPU giants. µ
Share this:
Share
Invidia might have a GPU that is too goood for the comp so they might be dumbing it down alot or they might just be in a load of poooo i hope it aint the latter but this putts back my new build or i might just give up like Nvidia .jj7
Can't say I've ever been so depressed.
Our engineers have failed us completely.
Products are old and slow.
Our time to market has become an embarassement.
The worst thing is that now all of the netbook, notebook, workstation and major OEM cycles have been lost - which means a MASSIVE movement toward ATI over the next 12 months.
AMD's shareprice has gone up around 500% in just 12 months - ALL driven by ATI's graphics division.
I remember the golden days where we lorded it over everyone, spending nothing on new development and laughing our way to $1Bn in sales per quarter.
Now our partners are dropping like flies - and we're pinning all our hopes on an Android phone.
It's a sad time to be in Santa Clara :-(
Maybe advertising wouldn't help immediately, but it certainly would help. This perception that nVidia is the only type of graphics card you should buy has been around forever. I remember going to the electronics store to try and get my hands on the new Radeon 9700 Pro, knowing they'd likely only stock 1. The isle was full of people oggling the fancy nVidia boxes, I slipped through and grabbed the ONLY 9700 Pro on the shelf and greedily turned for the register without saying a word. People actually stopped me. They were like, "Whoa! What are you doing man! Don't buy that, it's ATI!" Talk about blind loyalty. If anything, advertising could counteract this perception and maybe people will elect to wait for their opportunity to buy an ATI card instead of buying whatever nVidia product they can get their hands on.
A delay from your competitor means nothing if you can't produce anything but your top card to sell (and not in good quantity). I'm still waiting since October for my Vid card from amazon (a 5850). Who cares if Fermi is late, neither can produce many 40nm cards because of TSMC. AMD has been here before with Intel. They couldn't produce enough CPU's to take advantage of the fact they were the best cpu in town (A64 etc - Until Core2). If you can't produce the chips you can't take advantage of your enemies mishaps.
I agree 100%, except that with availability what it is right now I'm not sure an advertising push would help.
So, Charlie wrote most of this information up in his May article about Fermi on The Inquirer. Strange how tech sites have spent the last several months denying that, and then slowly coming around to agree with it again as it comes true.
Most morons buying new PCs and video card upgrades off the shelves only know nVidia. They don't know anything about the products, just that they want nVidia. I browse the video card retail shelf at Fry's Electronics sometimes and I see this over and over. NVidia this nvidia that.
If AMD were halfway smart, they would *gasp* advertise their new line of 5xxx cards through many forms of media, perhaps even go big and do a superbowl commercial. You can't own the market if the average dildo doesn't know that you exist.
It's okay. Don't worry. They're just having a hard time sourcing the correct woodscrews. A quick trip to B&Q and they'll be back on track.
No, even a Hyper Threading enabled quad-core can't chug the code that fast. Only a CPU with as many threads as Intel's Larrabee can do that, but Larrabee got delayed. Not that we need Physx/CUDA when we have compute shader in DX11 and OpenCl. And removing CUDA support won't reduce the die size. CUDA is a natural consequence of GPU becoming more and more general purpose. The only difference is Jen "arrogance" Hsun decided to go with CUDA instead of a something vendor-agnostic. If you want to use the full capacity of your GPU for rendering pictures, you can just turn off PhysX in the game.
I may be buying an ATI 5-series card soon. I don't really care for nvidia cards with 2.7 wood screws.
First, Nvidia renames their 8-series to 9-series then GTX2xx series, which are signs of trouble. Then they lose the chipset license and now can't make chipsets for Intel platforms. And now, this.
Jensen must be up to his ears in cow shit right now.
I'm still not sold on the idea of physics being dealt with by a 700-900Mhz GPU on a 100Mhz PCIe bus. Not for all of us consumers anyway. Might help an E5xxx/X2 chipped system but how about a hyper-threaded quad core? I'd think it could chug the code. Just a big "what if" since we can't compare GPU vs. CPU physics with Physx code being the defacto.
But couldn't the nVidia cards be much better if that computing space on the die went back to drawing pictures? Sure would be smaller and less of a power hog I'm pretty sure. Surely this has to do with all the delays as well.
I just think they'd be better off leaving Cuda to a specialty market outside of gaming. They could then pump non-complicated chips out and make those that absolutely need Physx wait for it. Get it right in $1,000+ workstation cards then move it over to consumer products when they got it worked out. Just Opinion...
Very good idea.. err Idea. However, like Intel, I see pride and stubbornness winning over corporate common sense. We'll see.
I hope they do. We can see what lack of good competition does to prices and innovation with Intel/AMD. I'd hate to see the same in the GPU market.
I hope they have an ample supply of their cans of whoop-ass to survive it.
From what I've seen of DX11 it appears to do some of the work with regards to flags and wind effects and maybe others.
DX12+ will make it redundant in time I reckon.
So Nvidia could unlock physixs on their cards to work with ATI GPU... I would be one of many who would buy a Nvidia GPU to use it with the ATI GPU for physixs... this could help to sell something more