Both NVIDIA and AMD have opted for the same manufacturer for the chips of their new gaming graphics cards, but although for the first time they will compete head-to-head with the same node, Huang’s are once again one step ahead when it comes to to the acquisition of chip supply. And it is that as we have been able to know NVIDIA has reserved all the production of TSMC in the most advanced node that the company has to mass manufacture, its 4nm.
NVIDIA wants to deal a hard blow to AMD and distance itself from it based on a checkbook and exclusivity. After leaving Samsung and with the consequent problem of pride for Koreans, it is now thrown back into the hands of Taiwan to hit back (at least partially) its most direct rival, since Intel plays in another league.
NVIDIA GH100, the beast will arrive in the most advanced node
Since the 3nm at the moment are a chimera that has no real prospects of reaching the market soon, the 5nm being in the high-performance mass production phase for a few months and having no real alternatives with Samsung, I had little margin NVIDIA to move and be at the forefront.
TSMC’s 4nm is the only real option to distance itself from AMD and as expected the greens have pulled out the checkbook and thereby ensure that any chip that comes out of Taiwan at 4nm is destined to produce the new Hooper GPU.
From here comes the controversy, since there have been rumors about the production ramp of this type of node, while on the other hand and after having already seen the chip itself, it is speculated with the arrival for the third quarter of this year, that is, between July and September.
AD102 at 5nm and NVIDIA GH100 with TSMC at 4nm
TSMC cannot ensure production of its 4nm for NVIDIA and AMD at the volume level that both companies want, therefore, Navi 31 and AD102as well as other inferior chips must be manufactured on the 5nm node.
GH100 is different since the volume is much smaller, many fewer units will be sold proportionally and above all, it will have at most and supposedly two variants that are simply chips that have not passed the internal tests to be GH100 as such in its entirety.
What NVIDIA is looking for here is simple: not to repeat the mistakes of the past with AMD and TSMC, since if we remember in 2020 those of Lisa Su ate their toast by reserving production with the Taiwanese 7nm giant, having to resort to Samsung and its 8nm with a lower monthly volume of chips.
So with the move to TSMC and their 5nm and 4nm for the GH100 it is more than likely that the share of the 80% of the market that NVIDIA has increases if they do a good job and are ahead of AMD, something that is still not clear to us since on paper the data of one and the other has lights and shadows.