Well, I disagree on the "cutting the GPU.. entirely" part. Here is why:
If the GPU division is definitely bleeding red ink, then yes, amputate ASAP. But, we don't know. They might actually generate positive cash flow.
Battlemage has actually gotten a good reception and uptake. And, their GPUs are produced entirely fabless. So, in many ways, it should be reasonably straightforward for Intel to figure out if the GPU division is currently (!) a drag or a plus for the bottom line. And then, there's this: You can't be in the Notebook/2-in-1 SoC space without a decent iGPU. Not anymore. Object lesson here was Qualcomm's Snapdragon SoC. One major reason for poor reception in the market was (is) the very underwhelming GPU performance. Which was a letdown, as many of us expected their Adreno prowess from their smartphone SoCs to carry over to their notebook SoC - it didn't.
So, you need a decent iGPU with good software support to succeed in the notebook space. Which is a lot more straightforward if that iGPU is derived from a dGPU based on the same architecture. Xe and Xe2 were and are successful as iGPUs. To keep that going, Intel will need a capable GPU team, even if they decide today to exit the dGPU business entirely.
The need to have good iGPU performance in Mobile SoCs will IMHO be even bigger once the first Mediatek/Nvidia SoCs hit that market. Unlike Qualcomm's Snapdragon, that collaboration's SoC will have a capable GPU, with great firmware and driver support from the start.
One problem I saw while at Intel is lots of legacy. So you have this legacy verification system with legacy tools to check some things, so you need people to keep it alive.
Moving to the new system is a massive effort, and give no immediate benefit - so it is always postponed.
On ANET - I keep hearing the argument that their O/S is top notch and admins love using it hence the preference. Any truth to that? Luckily, I always thought I missed the train and it kept going highner on my face. Seems like they are finally taking the hit.
On another question, you also seem to agree that AI-themed semi value-chain is now taki g a hit for one reason or another. Do you expect some positive news from AI world to re-fire the stocks? If so what'd that catalyst be? If not, should we step aside for a year or so till some cycle normalization? Love to hear your view on this, a bit macro call, I know, but if we can get it roughly right, it would be tremendous.
Well, I disagree on the "cutting the GPU.. entirely" part. Here is why:
If the GPU division is definitely bleeding red ink, then yes, amputate ASAP. But, we don't know. They might actually generate positive cash flow.
Battlemage has actually gotten a good reception and uptake. And, their GPUs are produced entirely fabless. So, in many ways, it should be reasonably straightforward for Intel to figure out if the GPU division is currently (!) a drag or a plus for the bottom line. And then, there's this: You can't be in the Notebook/2-in-1 SoC space without a decent iGPU. Not anymore. Object lesson here was Qualcomm's Snapdragon SoC. One major reason for poor reception in the market was (is) the very underwhelming GPU performance. Which was a letdown, as many of us expected their Adreno prowess from their smartphone SoCs to carry over to their notebook SoC - it didn't.
So, you need a decent iGPU with good software support to succeed in the notebook space. Which is a lot more straightforward if that iGPU is derived from a dGPU based on the same architecture. Xe and Xe2 were and are successful as iGPUs. To keep that going, Intel will need a capable GPU team, even if they decide today to exit the dGPU business entirely.
The need to have good iGPU performance in Mobile SoCs will IMHO be even bigger once the first Mediatek/Nvidia SoCs hit that market. Unlike Qualcomm's Snapdragon, that collaboration's SoC will have a capable GPU, with great firmware and driver support from the start.
Where's the proof PDK is still bad?
One problem I saw while at Intel is lots of legacy. So you have this legacy verification system with legacy tools to check some things, so you need people to keep it alive.
Moving to the new system is a massive effort, and give no immediate benefit - so it is always postponed.
And then there is a newer system, etc.
Removing legacy and improving perf. is not easy.
We already saw from isscc that BSPDN didn’t do squat for SRAM because they had to route all over the place to get to the middle of the SRAM
On ANET - I keep hearing the argument that their O/S is top notch and admins love using it hence the preference. Any truth to that? Luckily, I always thought I missed the train and it kept going highner on my face. Seems like they are finally taking the hit.
On another question, you also seem to agree that AI-themed semi value-chain is now taki g a hit for one reason or another. Do you expect some positive news from AI world to re-fire the stocks? If so what'd that catalyst be? If not, should we step aside for a year or so till some cycle normalization? Love to hear your view on this, a bit macro call, I know, but if we can get it roughly right, it would be tremendous.
You keep saying “PDK is bad” - what do you mean by this?
I’ll probably sell my Arista on this next short term upward swing then move over to pure storage. I see more potential for medium term growth in PSTG.
I’d like to see a post on all your shorts vs the success of them. I’m not disagreeing with anything but you tend to love to short so just curious
“.. hire your own team to make an internal custom AI switching software stack?”
That would need to be some rapid pattern matching 🤔
MSFT reducing CaPex? Wasn't that BS called out?