Mass Confusion about NVIDIA's G70
And Other Parts
by Josh Walrath
Last week some of the first good looking information on the G70 from
NVIDIA was leaked. Now, this info pointed towards the G70 being a 110 nm
part clocked at 430 MHz and it featured 24 pixel pipelines (six quads), and
gave some other pertinent information. The materials leaked with the specs
also made it look like it was legitimate. Now, I am just not so sure.
At the J.P. Morgan technology conference, NVIDIA gave a 15 minute
presentation with a short Q&A. Marv Burkett, CFO of NVIDIA, gave the
presentation. Most of the presentation talked about NVIDIA's current
financial position, how their products are positioned in the market, and how
well certain aspects of the business are growing (GPU's and MCP's being the
main growth areas). He also went on to state that while the Consumer
Electroncis group (those in charge of products like the X-Box) will have
very flat growth until around Q3, when they will start receiving income from
the RSX (PS3 graphics unit). Their WMP (Wireless Media Products) division
had a big customer last year, but that has since dried up. However, they
are expecting two major customers to come on board next quarter, so that
area should be shored up.
In his talk he covered quite a few topics, and some of the bigger ones
were that of the RSX and 90 nm products. Currently the RSX is still in
development, and no actual silicon is available as of yet. Looking at
Sony's timeline, I would expect the RSX to be taped out by the end of this
Summer, and that first silicon will be available in late Fall. Once all the
little problems are fixed and the design is working as it should, Sony will
take over production and pay NVIDIA a royalty for the use of their
technology. While overall revenue from this deal will be lower than the
X-Box, NVIDIA will not have to worry about things such as production
schedules, poor yields, and the other pitfalls of handling the production
portion of a GPU. This will of course have a positive effect on net profits
though, since this will essentially be "free money" from work previously
done. Sony has laid out a good chunk of change for the current design work,
and I would imagine that delivery of first silicon will be faster than I am
quoting because Sony owns and runs the Fab that the silicon will be produced
on (without having NVIDIA pay out the waazoo for an accelerated first run,
you can expect Sony to give that product top priority in its Fab).
The demos that were running at E3 were apparently mainly running on SLI
machines, as well as G70 parts. Marv talked about how these demos were run
on an upcoming product with many similar capabilities as the RSX chip. So,
while the RSX will have more features that are aimed at the PS3, we can
expect this next generation of cards to nearly match the overall performance
and feature-set of the RSX.
Now for the confusion. Earlier this year at a conference call with
Jen-Hsun and the gang, it was stated that the first 90 nm parts were going
to be introduced this Fall. Now we are hearing something different. At the
J.P. Morgan conference, Marv Burkett clearly stated that the first 90 nm
part will be introduced this quarter (which definitely cannot be
characterized as "Fall"), and that all "large" products will be 90 nm from
here on out. This suggests, in very strong language, that the G70 will be
90 nm (as it has not been released as of yet, and it is a large part). So,
was the leak last week legitimate? If Marv really meant what he said, then
no, the G70 will not be a 110 nm part.
The amount of confusion that NVIDIA has spread about their products in
the past two years in terms of leaks has been pretty astonishing. Nobody
has a handle on what is going to be introduced, and while the big picture is
fairly well known, the details are not. We all know that the next gen of
products will have a faster clockspeed, and that they will feature at least
24 pixel pipelines. Other than that, it is a lot of guesswork. Now, one
noted hoax that NVIDIA perpetrated was that of hinting the NV40 was a 8x2
architecture. Apparently NVIDIA delivered "special" cards to some
developers that showed up as 8x2, and of course this information was leaked
to the internet community, and ATI was able to see what was going on. At
this point ATI thought they were sitting pretty with their X800 Pro and X800
XT PE. A 12 pixel pipeline card running at 475 MHz should just destroy a
8x2 architected 350 MHz part. Of course the XT PE would wipe the floor with
the competition. Then April rolled around last year and we saw that the
NV40 was a 16 pipeline design, the 6800 GT was significantly faster than the
X800 Pro, and the 6800 Ultra matched the X800 XT PE. As we saw, ATI had to
introduce the X800 XT near the end of Summer of last year to be able to
offer a part more competitive with the NVIDIA range of cards (and gave users
something between the middling performance of the X800 Pro and the
outstanding performance of the X800 XT PE). Unfortunately for ATI, they had
some serious supply issues, and their XT and XT-PE parts were very hard to
find.
Throughout the past 5 months we have been hearing many conflicting reports
about what the G70 will be. If Burkett is giving us a true glimpse (which I
think he is), then we can speculate on what we can expect to see. First off
the G70 will be 90 nm (and not the 110 nm that we were all expecting), and
it will probably be clocked significantly higher than the 430 MHz that the
leaked presentation documented. We can also expect a part that is around
300 million transistors. Depending on how NVIDIA has allocated those
transistors, I think we will see a minimum of 24 pixel pipelines. There has
been a lot of talk about the possibility of 32 pixel pipelines, but I just
don't know if that will happen. My conservative nature says no, but it is a
distinct possibility that there could be essentially 32 pixel pipelines. I
think we will also see a new multi-sampling unit that will be able to handle
HDR content (unlike the current unit). Other things such as PureVideo will
of course be included, and we will probably see a couple of new wrinkles.
The "GT" version of this part could be clocked around 450 MHz, while the
"Ultra" edition of this part will probably be 500 MHz+. Power consumption
will still be around 6800 Ultra levels.
With that out of the way, we can move onto the fun stuff! Now, this is
all speculation as essentially NOTHING of the other G7x products has been
leaked. I have a feeling that with the overall success of TSMC's 90 nm
process (which is apparently very, very healthy) we can expect to see NVIDIA
phasing out its NV40/41/45/48 chips. These are very large at 130 nm, and
are not as cost effective as they once were. I feel that there is going to
be a large turnover in the $250 to $400 range with a new set of products.
The NV43/44/44a will continue to address the low end to the $200 market, but
the large 130 nm NV4x parts will soon be replaced by smaller, more cost
effective 90 nm parts. I think we will see some true competition to ATI's
110 nm X800 series (the X800 and X800 XL). The new series of 90 nm products
will feature the same pixel pipeline design of the G70, and all of the
optimizations that entails. If my speculation is correct then the low end
90 nm part will be a 12 pixel pipeline product running between 450 MHz to
500 MHz. This will compete with the X800, and from past indications on per
clock performance of the NV4x architecture, this should be faster than the
X800, yet still be priced around the $249 level. The next step up will be a
full 16 pixel pipeline design running around 500 MHz. This will compete
with the X800 XL in price, but will of course be faster. If this product
does in fact exist, and is sold around the $299 mark, then it could
seriously be the best bang for the buck that we have seen since the X800 XL.
This leaves room for one more product.
A G7x part with 16 pixel pipelines and running at 600 MHz would exist
at the $350 to $400 price range. This part would of course spank all of the
current high end cards (6800 Ultra, X800/X850 XT PE), yet be offered at a
lower price point. While this card would be very fast, it will still not be
able to compete with the high end G70 parts priced at $450 and above. A
massive move to 90 nm would give NVIDIA a pretty solid segmentation of
products, and allow them to stop their 130 nm production of large parts.
The only real question here is what will happen to the 110 nm NV42? Would
NVIDIA be better off keeping that part and moving it down to the $200 price
point and keep the 6200 and 6600 parts at sub-$175? Or will the 6600 GT
still be the best part at just under $200 and phase out the NV42? My gut
feeling is that NVIDIA will stop production on the NV42, as it honestly
gives about the same overall performance as the cheaper 6600 GT. So, by the
end of this summer, NVIDIA will only be producing 110 nm NV43/44/44a and the
90 nm G7x parts.
Again, much of this is speculation based on comments by Marv Burkett,
as well as some other small leaks and info that is floating around. When
ATI released their R300 in the form of the 9700 Pro, and NVIDIA was left
sitting with the aging GeForce 4 Ti series to compete with this product and
the NV30 had not seen the light of day, Jen-Hsun challenged his people to
match ATI, and he essentially said, "This is war!" If NVIDIA is continuing
with that philosophy, then we can expect to see a lot more disinformation on
coming products, and the smoke will get amazingly thick. The only thing we
shouldn't do is underestimate NVIDIA. It is a very aggressive company, and
their engineering talent is seriously second to none. Hopefully ATI will
have taken this challenge seriously, and we can expect to see some
impressive parts from them as well. The R520 does not look to be a slouch,
but hopefully ATI has not been lulled into complacency with the rumors that
the G70 is a lower clocked 110 nm part.
.
Seriously though, I may have a liking for what Sony are doing technically with PS3, but that doesn't mean I don't like X360. I've had this avater since pretty much the first X360 pics hit the web, it may be time for a change soon.
