When do you think AMD will start manufacturing its GPUs itself?

Techno+

Regular
hello,

When do u think AMD will dump TSMC and manufacture GPUs itself? In doing this i see the folloing advantages:

1) AMD no longer has to pay TSMC, resulting in lower manufacturing costs and more competetive pricing

2) GPUs gain access to advanced manufacturing techninques more quickly.

What other advantages\disadvantages can u think of?
 
hello,

When do u think AMD will dump TSMC and manufacture GPUs itself? In doing this i see the folloing advantages:

1) AMD no longer has to pay TSMC, resulting in lower manufacturing costs and more competetive pricing

2) GPUs gain access to advanced manufacturing techninques more quickly.

What other advantages\disadvantages can u think of?

The only GPU I think AMD is likely to manufacture themselves anytime soon is one integrated into the CPU. The main reason for this is that from everything I've heard CPUs make substantially more profit than GPUs. So unless they can satisfy the demand for CPUs they effectively don't have any capacity to spare for GPUs.
 
AMD doesn't have capacity to produce GPU's. When they build and equip FAB38, they will start producing FUSION hybrids (CPU/GPU), and maybe some volumes of GPU's just to test, and adopt APM to the GPU manufacturing. Only with New York FAB online, AMD will have enough resources on its disposal to start GPU manufacturing in full on it's own. But that will not happen in the next 5 years!
 
  • Like
Reactions: Geo
65nm gpus on AMD's process means they have to be converted to SOI, which means delay. But if they are going to integrate they will have to try this out sometime, and if I were then I would be doing testing on it now. They are limited capacity wise, so it would have to be profitable, as in low end, small, low power gpus for laptops, or perhaps even the xbox 360. Key word being profitable.
 
I think never. Not really viable, TSMC and co. doing it is cheaper in the long run.

Why?

The cost model that is used when producing any type of silicon at TSMC/UMC/any other external fab is very predictable and not very flexible in terms of pricing: high cost initially, steep steady drops, petering out in the end when the next process becomes mainstream. Client don't have much choice in deciding how to the silicon is accounted for.

It'd be a mistake to assume that they use the same forumula with their own fabs : it's entirely acceptable to write off major parts of the fabs on CPU's first and only add only a small remaining fraction to the cost of the GPU die. In that case, cost of the GPU dies would be vastly lower compared to the ones at TSMC.

I'm not saying that's necessarily what they are going to do (what do I really know about accounting?). Just saying that they have a large flexibility in the way costs will be assigned and that the cost will almost always be much lower that the TSMC route.
 
I guess it comes down to economics. If AMD makes a greater markup on CPUs than their GPUs and has CPU demand that exceeds their fab capacity, then maybe not for a very long time. Who knows how those things will shift.

But there could also be a strategic change: AMD could decide that they can increase the margins on their GPUs by doing them internally, as well leverage their process knowhow and leadtime to gain significant marketshare wins in the GPU market which could lead to large longterm windfalls.
 
If AMD is producing GPUs, it will rather use their cutting edge process for CPUs, and the previous for GPUs / Chipsets. That way they can get more life out of each process-generation.

But NEVER is most probable, unless Intel develops and produces cutting edge GPUs.
 
I guess it comes down to economics. If AMD makes a greater markup on CPUs than their GPUs and has CPU demand that exceeds their fab capacity, then maybe not for a very long time. Who knows how those things will shift.


Very true, this is one of the financial concerns of the CPU market, the synergy of the market or the antogism of Intel and AMD, depending on each course they take the outlook of either company is on shaky or solid ground.
 
Ok did a little digging for numbers.

TSMC said:
Average gross profit margin of 43.6%
Average operating profit margin of 35.2%

The TSMC figures are from '05 but I'd expect those percentages to remain about the same. I'd imagine those numbers translate onto GPUs at about the same rates. Not positive on these figures, they seem kind of low, but from AMDs earnings statements margins on "Computation Products" looks to be ~15%. It's possible with them price warring that number is close to correct I suppose or they could have been stockpiling inventory but that's based on a 9 month timespan.

Until TSMC catches up to AMD's process technology Nvidia could effectively be left in the dust. Similar to how AMD currently has a hard time competing with the 65nm Intel chips. The sales from those GPUs could really drive sales of CPUs as well. Especially if OEMs start prefering to keep AMD based systems tied to AMD/ATI chipsets/GPUs. Also if OEMs start moving PCs towards being media centers a higher powered graphics chips are almost a must to accelerate 1080I/P.

Not to mention margins on those stream processors they're aiming to sell must be insane. Seriously, $2600 for a R580 that says stream processor on the box? Not to mention with the market they're targeting you don't exactly buy just one of the things. I wouldn't be surprised to see rackmount setups loading up on SPs to make mini supercomputers.
 
Last edited by a moderator:
Just saying that they have a large flexibility in the way costs will be assigned and that the cost will almost always be much lower that the TSMC route.

That's where I disagree. First, they need all the resources they can get for the CPU/chipset business. Setting up high-end production lines for GPU's is very expensive. And correcting failures, respins, whatnot eats up lots of resources. The workforce is more costly in AMD fabs then at TSMC too.

TSMC is not only cheaper, but also much easier in every regard.
 
Back
Top