News AMD 4Q25 Earnings

[Thunder 57]

Should make it clear that this is x86 market share. As a whole, x86 is shrinking market share compared to the entire CPU market.

OK. Go post about it in the x86 vs ARM thread. I'm honestly tired of hearing how x86 is dead. It didn't happen in the 90's, it's not happening now.

 

[branch_suggestion]

Should make it clear that this is x86 market share. As a whole, x86 is shrinking market share compared to the entire CPU market.

Every meaningful ARM CPU outside of smartphones is either attached to a hyperscaler, is a surrogate to a GPU that people actually want, or Apple who are not a merchant supplier and will never replace the bulk volume of Windows PCs.
Once Intel has their NVL-C2C CPUs, both DC and client, you can say goodbye to yet another failed Tegra experiment.
Merchant ARM in DC is super dead, nothing is in the same league as Turin let alone Venice.
Windows, well Qualcomm are going through it now and the majestic NV display core has doomed N1X to being too little, too late.

 

[Nothingness]

I always find it amusing when people claim that something will never happen. This reminds me of the late 80s / early 90s when people said x86 would never be in workstations or data center servers. One should never use the never adverb Which doesn't mean I don't agree that it's boring to hear x86 is dying, it isn't, and I'd add alas.

 

[fastandfurious6]

Every meaningful ARM CPU outside of smartphones is either attached to a hyperscaler, is a surrogate to a GPU that people actually want, or Apple who are not a merchant supplier and will never replace the bulk volume of Windows PCs.
Once Intel has their NVL-C2C CPUs, both DC and client, you can say goodbye to yet another failed Tegra experiment.
Merchant ARM in DC is super dead, nothing is in the same league as Turin let alone Venice.
Windows, well Qualcomm are going through it now and the majestic NV display core has doomed N1X to being too little, too late.

About DC: AWS says their ARM Graviton servers are more efficient than x86 and they sell them a lot cheaper than x86. Is that just because Graviton is Amazon's own ARM chip?

Or "merchant ARM in DC is super dead" just means self-hosting companies don't buy ARM servers?

 

[MS_AT]

About DC: AWS says their ARM Graviton servers are more efficient than x86 and they sell them a lot cheaper than x86. Is that just because Graviton is Amazon's own ARM chip?

Or "merchant ARM in DC is super dead" just means self-hosting companies don't buy ARM servers?

Graviton is not merchant ARM. It's in "attached to the hyperscaler category" the original author used

In other words, you cannot buy Graviton to use in your own company. You can only rent it from Amazon. The same story with Google and M$, each one of them has their own ARM cpu.

 

[eek2121]

Graviton is not merchant ARM. It's in "attached to the hyperscaler category" the original author used

In other words, you cannot buy Graviton to use in your own company. You can only rent it from Amazon. The same story with Google and M$, each one of them has their own ARM cpu.

It also doesn’t come close to matching current EPYC chips in terms of performance. It is optimized for cost. Both die area and performance/watt. Amazon does charge less, however they make more money due to higher margins on the compute, so of course they are going to hype it up.

 

[adroc_thurston]

In other words, you cannot buy Graviton to use in your own company.

AWS outposts are theoretically a thing but yeah.
Neoverse is just soviet thirdworldism for hyperscalers.

 

[adroc_thurston]

It also doesn’t come close to matching current EPYC chips in terms of performance. It is optimized for cost. Both die area and performance/watt

Oh no big dawg, GV4 and 5 both are focused on perf.
GV5 in particular is an expensive 192c SoIC shebang with Actual Real LLC too.

 

[Josh128]

$AMD currently down 14%, Micron dipping below $400, looks like the smart money is starting to hedge their AI bets.

 

[adroc_thurston]

$AMD currently down 14%, Micron dipping below $400, looks like the smart money is starting to hedge their AI bets.

Normal market chimpouts. You can ignore them.

 

[jpiniero]

Normal market chimpouts. You can ignore them.

They need to change their name to AID.

 

[adroc_thurston]

They need to change their name to AID.

You would still get a post-ER selloff until they post a blowout DC quarter.

 

[ToTTenTranz]

14% isn't normal...

 

[adroc_thurston]

14% isn't normal...

It is normal.
$MU is down 9% rn and they're gonna poast a year or two of earthshattering quarters.

 

[jpiniero]

You would still get a post-ER selloff until they post a blowout DC quarter.

Isn't that part of the problem? AMD's never going to get as much of the AI market as Wall Street hopes. Figure most will just be the Cloud Guys rolling their own.

 

[adroc_thurston]

Isn't that part of the problem?

no.

AMD's never going to get as much of the AI market as Wall Street hopes.

yes they will.
It's gonna be selloffs every Q until AMD posts a blowout DC quarter driven by rack-level sales and then everyone on the Street will orgasm.

Figure most will just be the Cloud Guys rolling their own.

Only GOOG has good enough hw/sw teams to do that.

 

[Jan Olšan]

Every meaningful ARM CPU outside of smartphones is either attached to a hyperscaler, is a surrogate to a GPU that people actually want, or Apple who are not a merchant supplier and will never replace the bulk volume of Windows PCs.
Once Intel has their NVL-C2C CPUs, both DC and client, you can say goodbye to yet another failed Tegra experiment.
Merchant ARM in DC is super dead, nothing is in the same league as Turin let alone Venice.
Windows, well Qualcomm are going through it now and the majestic NV display core has doomed N1X to being too little, too late.

Nah, Nvidia has ways to push their stuff into the market. So their Grace/etc processors probably manage to sell even if they aren't that good, iirc.

Back when gaming GPUs were they main driver, I think they had a scheme where they basically extorted the card vendors and OEMs like this: if you wanted their highest performance GPUs (during the periods of high demand), you had to order bundles of relatively trash products (that wouldn't sell so well otherwise) together with them, like the MX chips, or old obsolete lowend chips that kept shipping for years and years.
When that worked, it killed two birds with one stone, earning extra revenue but also marketshare - since the Nvidia clients now held bags of those trash chips, it was now their task to sell them, so they might have cut orders of competing GPUs and go out of their way to put trash lowend dedicated Nvidia GPUs into laptops. It might have helped Nvidia get the nigh monopoly marketshare they have.

Now they can use similar tactics in servers, since the big corps have high demand for the datacenter GPUs so they may be similarly open to extortion.

 

[adroc_thurston]

So their Grace/etc processors probably manage to sell even if they aren't that good, iirc.

Nope.
Grace-grace was peddled literally everywhere and to everyone. 0 customers.

 

[Tangopiper]

Nope.
Grace-grace was peddled literally everywhere and to everyone. 0 customers.

It would take an actually competent Arm implementation to gain any customers. And NVIDIA doesn’t have a good track record with custom Arm cores lol

 

[adroc_thurston]

It would take an actually competent Arm implementation to gain any customers. And NVIDIA doesn’t have a good track record with custom Arm cores lol

Oh that mesh of theirs is a crime against humanity.

 

[fastandfurious6]

How it's possible Nintendo chose Tegra with 8nm tech from 2021 for their brand new Switch 2 while the rest of the world is putting a 4nm literal desktop flagship CPU + high-mid range GPU in a 8 inch handheld is beyond me

 

[poke01]

How it's possible Nintendo chose Tegra with 8nm tech from 2021 for their brand new Switch 2 while the rest of the world is putting a 4nm literal desktop flagship CPU + high-mid range GPU in a 8 inch handheld is beyond me

Handhelds don’t need more than 8nm for the mass market. Also Nintendo has ip

 

[jpiniero]

Handhelds don’t need more than 8nm for the mass market. Also Nintendo has ip

Just the opposite. They really cut down the clock speed in portable mode. Would have been nice with some Samsung node later than 8 nm.

How it's possible Nintendo chose Tegra with 8nm tech from 2021 for their brand new Switch 2 while the rest of the world is putting a 4nm literal desktop flagship CPU + high-mid range GPU in a 8 inch handheld is beyond me

$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$$

 

[jpiniero]

Speaking of consoles, I am impressed that they are still getting ~800M from Gaming. XBox Series sales are kinda dead with the price hikes, and it's only a matter of time before Sony hikes some more too.

 

[adroc_thurston]

Speaking of consoles, I am impressed that they are still getting ~800M from Gaming. XBox Series sales are kinda dead with the price hikes, and it's only a matter of time before Sony hikes some more too.

Radeon is doing well nuff.