GPU battle takes toll on NVIDIA

[onethreehill]

NVIDIA announced the GeForce GTX 280, which claimed the overall GPU performance leadership, as well as the GTX 260. Soon after AMD's graphics arm - ATI - announced its Radeon HD 4800 series, which aims to offer superior value for money in the mid-range.

Share prices at the end of last week would seem to imply that markets are keener on AMD's offering than NVIDIA's, with the former's shares rising by 3.2 percent and the latter's falling by an alarming 16 percent.

http://channel.hexus.net/content/item.php?item=13961

[hecktic]

thats not good...

[Macadamia]

It's mostly about them not being able to reach margins as desired with G92.

If they got flak over how the last 3 months (which was pretty much ripoff zone and good for them) I wonder how they're gonna deal with margins on G92b, especially 9800GTX's margins compared to the 4800s.

And of course, GT200. Oh.

AMD's previous stock was just ridiculously bad, so they have a slight improvement that's natural. Given how the US economy is now I wouldn't expect miracles per se on anybody's side, except Dell :p

[bowman]

Their chips cost more to manufacture, they use more RAM chips, they need more traces for their larger memory bus, they need more traces for more RAM, they need better, more expensive coolers, more expensive power circuitry..

And they need to compete with the 4 series. I guess they expected a repeat of R600 so they could rest on their laurels and release another G80 rehash, this time with almost twice the shaders. Guess it didn't cut it.

Let's hope they burn themselves as badly as they did on the FX series, maybe we'll see something new for once (and maybe enough red market share to see some tesselation and DX10.1).

[LuckyNV]

This is not a real problem for any company, its only a problem if its happening 3-4 years on the trot, clearly this has happened to neither ATI or Nvidia.

[loutsos]

More traces for their memory bus? How much is that? 1 cent per wire?

[Macadamia]

bowman, I think the worst decision they made this generation was adding FP64 seperate units instead of more shader power.

They want 1 chip to be used in several cards (Geforce, Tesla, Quadro) but unlike ATI's R6XX the shader units are much less flexible, and the whole architecture is less flexible (512-bit bus = 32 ROPs, no talk and no override, etc)

The fact that you see ATI refining, refining and refining their once also-ran architecture to something that's truly amazing in terms of engineering, is a real testament even when the whole R&D budget went down.

nVidia... I'm sorry I just can't say the same thing. Instead of making 2 chips, one big gamer chip and a mainstream/Tesla GPGPU chip (together), they made 1 instead.

[Dan_c]

That is not good news for the green side!
It's been to quik response to be just that... Someone having problems with there parteners?!

[Helmore]

bowman, I think the worst decision they made this generation was adding FP64 seperate units instead of more shader power.
*snip*
nVidia... I'm sorry I just can't say the same thing. Instead of making 2 chips, one big gamer chip and a mainstream/Tesla GPGPU chip (together), they made 1 instead.

I do not completely agree as they are trying to push CUDA as hard as they can and they want everything that's made for CUDA to also run on any consumer card. I'm not sure how much use we, as 'ordinary' consumers, have at the moment in that FP64 capability, but it could become relevant in the future and there is nothing wrong with some future proofing. Especially as things like GPU accelerated photoshop and video encoding on the GPU start to become mainstream, then I think there can be a use for it. But to be honest, I have no idea about what kind of computations and precision is used to do video encoding for example.

[Shintai]

Shareprice reflects future income. So even if nVidia wont make a huge massive profit but just a massive profit it drops. Its how long ago nVidias stock felt 20-25%? How did it end?

[perkam]

As much as one can feel good about the price drops here and there, and the wishing for more price drops, it always has consequences somewhere, and this time around Nvidia's ROI on 55nm and GT200 will be hit hard due to AMD's offerings.

Perkam

[Macadamia]

I do not completely agree as they are trying to push CUDA as hard as they can and they want everything that's made for CUDA to also run on any consumer card. I'm not sure how much use we, as 'ordinary' consumers, have at the moment in that FP64 capability, but it could become relevant in the future and there is nothing wrong with some future proofing. Especially as things like GPU accelerated photoshop and video encoding on the GPU start to become mainstream, then I think there can be a use for it. But to be honest, I have no idea about what kind of computations and precision is used to do video encoding for example.

Problem is, FP32 is certainly almost good enough for majority of CUDA stuff. Folding uses FP32, video encoding seems to too, and really, the current FP64 speed on the GT200 is abysmal anyway.

They should have seperated the FP64 units for the server/tesla cards so that businesses that really need it will have them, and those cards can sell at high prices so need to worry for margins.

[MomijiTMO]

Shareprice reflects future income.

Well it reflects future dividend yields and therefore future cash flows more so than accounting income.

Tech stocks have always been volatile. They are high risk shares and it is completely normal to have such a variation. This simple drop in price is probably wiping off a hundred million dollars or more out of their equity which reduces the amount of money that could be used to fund future products [R&D].

Hopefully they will bounce back soon.

[bowman]

More traces for their memory bus? How much is that? 1 cent per wire?

Not so much the wire cost but the cost of PCB manufacturing goes up.. Maybe it doesn't matter to you but it matters when you make tens of millions of cards.

Why are you being so cheeky?

[v_rr]

Shareprice reflects future income. So even if nVidia wont make a huge massive profit but just a massive profit it drops. Its how long ago nVidias stock felt 20-25%? How did it end?

Everything as a beginning.
But GTX200 and HD 4800 won´t make difference in Q2. Only on Q3/Q4.

[DeathReborn]

http://channel.hexus.net/content/item.php?item=13961

I expected worse to be honest. The analysts don't seem to like nVidia that much so it will probably drop another 10-15% (if not more) in a few weeks once they get their teeth into them.

[Sparky]

It was bound to happen eventually. Nvidia seemed to be shooting themselves in the foot on more than one occasion recently, and ATI finally has appeared to get its act together, so...

[n-sanity]

16 bloody percent :O !? I'd understand if the CEO left or something... But i guess they sat on their hand during their "freetime" from ati, and are now paying for it. 16% tho... *whistle*.

[Shocker003]

HEHE!!! drop in stocks means they have to drop their new card prices > I want to go with ati this time, if their HD 4870 will support physx in future::

[flippin_waffles]

I wouldn't be surprised if it will be Nvidia scrambling to try and adopt Havok in the not too distant future.

[Helmore]

HEHE!!! drop in stocks means they have to drop their new card prices > I want to go with ati this time, if their HD 4870 will support physx in future::

It has the capability to do so, whether it will ever support it depends on the relation between ATI and NVIDIA and Intel is also to a certain extend involved in what will happen regarding physics acceleration. I'd say there are interesting times ahead of us, I just wish they would quickly settle on a universal standard more akin to what DirectX and OpenGL does for graphics.

[bowman]

The Khronos group has already assembled a work group to determine a standard parallel programming language that every vendor will support. Apple's OpenCL is amongst the proposals so far.. Both Intel, AMD and Nvidia are onboard.

If not DX11, then whatever the Khronos group decides on should be very good for running both Havok and PhysX on any vendor's GPU.

[spicypixel]

Even though its hosted by apple im all for OpenCL myself.

[bowman]

Even though its hosted by apple im all for OpenCL myself.

It's just a proposal at this stage, they might go for something different, maybe even a CUDA, Brook+ or Ct derivative..

Personally I don't care as long as they actually agree on it and each vendor supports it.

[safan80]

AMD is in debt where as Nvidia is not. I see no reason for Nvidia to worry. Those analysts have always been stupid! it took them forever to realize AMD had a good product with the k8.