Article: Parallelism at HotPar 2010
By: Richard Cownie (tich.delete@this.pobox.com), August 4, 2010 8:23 am
Room: Moderated Discussions
hobold (hobold@vectorizer.org) on 8/4/10 wrote:
---------------------------
>I am not yet ready to count Nvidia out. They can be positively crazy with new and
>innovative designs (GPGPU wouldn't be such a hot topic now if it weren't for them).
>Their lack of integration means that they will be the only game in town when it
>comes to upgrading rather than replacing boxes. Fermi the product may suck, but Fermi the idea is still valid.
>
>I have no idea if a business can be sustained on that, but short term I don't see them die all of a sudden.
Well, I think nVidia has done us a service in pioneering
the GPGPU stuff, if only because it has forced Intel and
AMD to respond by improving their CPU's behavior for
throughput-intensive and parallel-FPU intensive apps.
But being a pioneer doesn't always lead to a sustainable
business. Right now NVidia just announced an
unexpected $150M drop in revenue for the quarter - that's
20% of their business that vanished overnight. And by
mid-2011 we'll have adequate CPU+GPU chips from both
Intel and AMD which will simply obliterate NVidia's
revenue from sub-$100 discrete GPUs. And in the $100-$200
range NVidia still has nothing at all that matches
the Evergreen series on features (DX11, Eyefinity) or
price-performance. A situation that's probably only
going to get worse with AMD's Southern Islands arriving
in Q4 2010 and Q1 2011.
That's a really ugly business prospect over the next
12 months or so.
On the technical side, they've had a couple of years of
unreliable and too-hot chips (leading to the loss of
major OEMs such as Apple). And they don't appear to have
any option in the near future to achieve tighter integration
with a capable scalar cpu, which is what both AMD and Intel
are doing.
And while AMD is doing the integrated CPU+GPU stuff,
that certainly doesn't mean they're going to give up on
the mid/high-end discrete GPUs. On the contrary,
they're making a ton of money on that and will surely
milk it for all it's worth.
Undoubtedly NVidia has some good people, but I just don't
see what rabbit they can pull out of the hat within the
next 2 years. What would make sense would be for Intel
to buy NVidia and take their good GPU technology and
throw away Intel's own dysfunctional GMA. And shrinking
Fermi from TSMC's broken 40nm process onto Intel's
solid 32nm process would probably give you
something pretty good. But that merger would be a big
antitrust issue, which Intel probably wouldn't want to
take on when they just got out of hot water with the FTC.
---------------------------
>I am not yet ready to count Nvidia out. They can be positively crazy with new and
>innovative designs (GPGPU wouldn't be such a hot topic now if it weren't for them).
>Their lack of integration means that they will be the only game in town when it
>comes to upgrading rather than replacing boxes. Fermi the product may suck, but Fermi the idea is still valid.
>
>I have no idea if a business can be sustained on that, but short term I don't see them die all of a sudden.
Well, I think nVidia has done us a service in pioneering
the GPGPU stuff, if only because it has forced Intel and
AMD to respond by improving their CPU's behavior for
throughput-intensive and parallel-FPU intensive apps.
But being a pioneer doesn't always lead to a sustainable
business. Right now NVidia just announced an
unexpected $150M drop in revenue for the quarter - that's
20% of their business that vanished overnight. And by
mid-2011 we'll have adequate CPU+GPU chips from both
Intel and AMD which will simply obliterate NVidia's
revenue from sub-$100 discrete GPUs. And in the $100-$200
range NVidia still has nothing at all that matches
the Evergreen series on features (DX11, Eyefinity) or
price-performance. A situation that's probably only
going to get worse with AMD's Southern Islands arriving
in Q4 2010 and Q1 2011.
That's a really ugly business prospect over the next
12 months or so.
On the technical side, they've had a couple of years of
unreliable and too-hot chips (leading to the loss of
major OEMs such as Apple). And they don't appear to have
any option in the near future to achieve tighter integration
with a capable scalar cpu, which is what both AMD and Intel
are doing.
And while AMD is doing the integrated CPU+GPU stuff,
that certainly doesn't mean they're going to give up on
the mid/high-end discrete GPUs. On the contrary,
they're making a ton of money on that and will surely
milk it for all it's worth.
Undoubtedly NVidia has some good people, but I just don't
see what rabbit they can pull out of the hat within the
next 2 years. What would make sense would be for Intel
to buy NVidia and take their good GPU technology and
throw away Intel's own dysfunctional GMA. And shrinking
Fermi from TSMC's broken 40nm process onto Intel's
solid 32nm process would probably give you
something pretty good. But that merger would be a big
antitrust issue, which Intel probably wouldn't want to
take on when they just got out of hot water with the FTC.