Sinatech: ATI/AMD Aquisition Agreement Reached: ATI facing a big shake-up

If the merge does go through, it will NOT kill Nvidia. Nvidia will flurish so long as Nvidia keeps producing kick-ass chipsets and video cards for both platforms.

 

I don't think ondie GPU can compete with high-end (or even main-stream) GPU in terms of performance. How can you integrate a GPU with half billion transistors into CPU when CPU itself has fewer than 300 millions transitions? Ondie GPU, if there is, can not even meet vista premium requirement given the die area that can be allocated to GPU IMHO, if the semiconductor technology keep advancing as predicted in ITRS road map.

 

I actually don't think that was 3dfx's problem, and ATI used to have no 3rd party graphics makers.
3dfx problem were no OEM prescence, which was a rapidly growing market. As far as the retail sector go, they still sold well over 50% of the add-in cards during the voodoo3 days, and dropped to slightly below half (I believe 30% to 40%) during voodoo 5 (though they still remained the single largest supplier of cards to the retail market). Of course, they only kept up such high levels of sales due overly aggressive pricing, and their margins went to practically nothing from the voodoo 2 days, despite market share staying the same or possibly even going up.

Is that an x86 cpu? Actually, I guess it doesn't matter for ondie integrated graphics. Still, they could do VIA and focus on low end systems that way.

 

Pardon? What about PackardBell V3-3k TV-out, MB MSI-6168 (V3-2k onboard) used for PackardBell Club 800P PC series, Compaq V3 - 1000 16MB (special design for Compaq), Compaq V3-3500 (cheap design for Compaq, w/o TV tuner, for Presario 5000 series), Dell V3-3kD (cheaper design fo Dell, w/o TV-out, optional for Dell Dimension XPS T Series), Gateway Bonesteel V3-1000G (Velocity 100 with different BIOS), Gateway V3-3kG (optional TV-out), Velocity 100 and many many other cards designed specially for OEMs ...?

 

It was reported on that Sina website one month ago that Intel had already planned the strategy maybe 1 to 2 months ago if the deal does go through. The negotiations between AMD and ATI has been going on for months. Many people think Intel purposedly leaves the highest end chipset market with no successor to 975 so as to leave the market to Nvidia.

This is the future roadmap and you will see the extreme high end is left untackled by Intel.
http://images.dailytech.com/nimage/2100_large_intel_ww28_apac_motherboards.png

It is actually a very brilliant strategy since every major customer would want to have a legitimate second source of supply. Nvidia will be given the highest end chipset segment for their leadership in SLI. Nvidia meanwhile, should have plan to become second chipset supplier in AMD world too. AMD clearly will be reluctant to focus resources on the highest end chipset since they simply could not play all-out in every front. Essentially, Nvidia will then be de-factor highest end chipset player with relatively low volume in both AMD and Intel world. GPU side business should thrive in short and mid term since restructuring will be a very big hit on ATI and their GPU strategy simply will be second thought in realizing the synergy of platform bundling.

All in all, this isn't really news for all the major players. Everyone has planned strategy and how it plays out would be crucial to their future outlook.

 

So? Just because there was models carrying them does not mean they were selling well. Its well known that the OEM market was one area where 3dfx had great trouble with.

 

NV loves high margins. If ATI is redirected toward integrated chipsets, I see no reason why NV couldn't flourish in the top 1/2 of the market.

I just would hate to see it, personally. I want robust competition at the bleeding edge performance front.

I think the effects of this would take 3-5 years to really shake themselves out. R600 and probably R700 are pretty much already in the engineering bank, so to speak. It's R800+ that I'm losing sleep over.

 

Ok, they had a minimal marketshare in the OEM market, even if they had various products designed for it. Compared to their dominating prescence in the retail market, they had next to nothing of the OEM market-share.
I recall most PCs of the era shipping with ATI or nvidia cards, I had never seen a PC advertised with a voodoo card from a major OEM. For a market-leader, 3dfx really failed to break into the market.

 

I still say I will believe it when it's announced, not before.

To speculate though, AMD has made comments about targetting the mainstream, it's about the "platform", not the niche high end. They moved into the server space successfully, not dominant, but they are a player. That was doing things different than what they had done before, this may be more along the same lines, thinking outside of their traditional role, innovating. They are a business first, not a sporting team trying to win a trophy. If they could supply the world with a complete solution, fast enough cpu, fast enough graphics, say $500 cheaper than what would exist separately, sold by Dell etc, and gained pedestrian but large volumes of the consumer pc market, it would be a success for them.

You can have infinitely fast cpus, have all the geeks in the world impressed, but if the bottleneck is really the video card for your target market, what value does that have, what future is in that? I know there are good reasons for faster cpus, obviously, servers, scientific applications, etc, but the majority of the consumer market doesn't need it for the applications they run, the most demanding applications are games and that is gfx card territory for the most part. Whether a dvd rip takes 40 mins or 30 mins isn't significant enough to upgrade the entire computer. The traditional role for AMD would be just to keep competing with Intel at a disadvantage for king of the hill in a race few care much about. They still have to compete obviously, and for marketing being on top is great, but they have to do something to shake things up as well. When a fast cpu like a conroe is $200 at the low end and a good video card is $400+, you can see where the money in the PC is to be made, especially if you get a large contract with somebody like Dell.

Of course, I don't know anything about anything, just having fun with the idea..

 

I believe this will eventually be beneficial to the industry, and to the customers. My magic-ball says we will have gigantic trio-poly leading CPU/GPU market in 10 years - Intel, AMD, NVIDIA. It all makes sense now as I previously mentioned: Intel dropping CrossFire support w/ Conroe launch, Intel dropping ATI chipsets from their lineup, Intel's brand new IGP launch, AMD's awful quietness as of late, and NV CEO's indirect warning and their much less aggressive marketing on NForce 500 series for AMD.

In the short term, the biggest fire would be on NV's. Not only ATI will all of a sudden have competent chipsets for AMD, but R600 might as well be customized for 65nm process and I expect nothing less than 1GHz clock speed.

When do you think NV will plan for their fabs? (I think alot of people underestimate NVIDIA's ambition. Their CEOs won't be happy until their company climb up to the position where Intel sits right now.)

 

Oh and another question: Will D. Bauman be able to keep his position?

 

Well, Jen-Hsun Huang could always build a taller building than theirs and look at Paul Otellini from above (NV's headquarters is, literally, almost across the street from Intel's at Santa Clara, CA)

 

How likely is AMD to divert fabspace away from CPU-products (~150mm^2 die = $100-900 income), to fab GPU-products of inferior margin (150mm^2 die = $30-100 income)?!? Well, I guess a way to phrase this question: if AMD were in a position where it could fab 1 more CPU or 1 more GPU (but not both): under what circumstances would the GPU yield (heh, pun intended) more revenue?

I can't think of a case where a recent fabless company transitioned to an IDM (foundry + design-house + IC-supplier.) Conversely, There are plenty of examples of the opposite: companies outsourcing wafer-production from internal-fabs to contract mfgs (TI, LSI, Motorola, Western Digital, Lucent, more than I can think of), and outright ditching their fabs altogether.

 

Now would be the worse time for Nvidia to build their own fab.

If ATi starts using AMD fabs (which I doubt will happen anytime soon) then all of the fabs will look right to Nvidia and give them nice deals to stick with them.

 

Remember, ATI doesn't have to make all their GPUs from AMD's fabs. Like I previously mentioned, high-end GPU demand is not anywhere near that of CPUs. ATI can make R600 (the top-of-the-line SKU and its derivatives) from AMD's fabs and the rest of their X2xxx series can still be outsourced. (Is there any reason for TSMC to turn a customer down?)

How about AMD? Think about their recent price cuts. Starting tomorrow, the official price of A64 3500+ is $89. Depending on Intel's ability to meet the Core 2 Duo demand, the prices (especially for single-cores) will go down even further. Compared to that, high-end GPUs will get them alot higher margins per-die, considering their superiority over the rest of fab houses. (sans Intel, of course) Once Fab 36 runs in full force, and the 3rd fab coming into the picture, it won't be a problem for AMD to share a tiny bit of fab space for ATI's high-end GPUs.

More than anything, AMD finally gets their own platforms. Considering that everybody is desparately get their platforms out, this could be very beneficial to AMD in the long run. (Hell, NV even sell their name on memory sticks.)

See above. But I agree that it might not be the best for NV to build their fabs. They might buy one, though. On another note, I'd be willing to bet that AMD made a suggestion to NV first and NV surely refused it down-right.

Edit: I just remembered Torrenza (which basically connects everything via HyperTransport), and a rumour that AMD will integrate PCI-E on-die sometime in the future (just like they integrated memory controllers on-die). This merger makes even more sense.

 

I dont think AMD played a game of going big. I think they wanted the biggest cheap guy they could get.... enter ATi. Nvidia would be far to expensive for AMD right now, hell I think ATi is far to expensive for AMD right now.

I think AMD as a whole would be much better off purchasing a smaller company out there then buying something as large as ATi.

 

High-end GPUs are HUGE, G70 and R580 are around 350 mm². (G70 to compare them on equal process size, but a G71 at ~200 mm² is no slouch either)

Think about yields, and think about the ~2.7 dual-core Athlon 64s you sacrifice for every shot at one of these 350mm² monsters. Realistically speaking, you'd probably give up 4+ 65nm X2s for every working 350 mm² GPU (defect yields, logic vs. cache, and even edge losses)

4 X2s get you at least $400, does it still seem like a brilliant idea to process even a small number of wafers in your expensive new FAB with high-end GPUs on them?

Torrenza's set out to solve a problem (CPU <-> Co-processor latency) GPUs don't have.

 

huge sizes partly because they use "standart" cells, no?
And why would they put GPU in co-processor slot? Amybe they can create custom co-processor...

 

True profit margins are much higher on the CPU end per chip. Also there is less overhead for other parts (ram, circuit board, etc, etc), the CPU market is much more profitable and lucrative, it is highly unlikely AMD will focus on GPU's in thier own fabs. Possible make chipsets though. Depending on sucess of future ATi GPU's vs. nV sales wise will be the deciding factor on the amount of focus AMD will give their GPU segment

 

If they went to a custom process, I don't believe they could keep up the development pace that's required in the GPU field.