IBM as the secret customer for Intel's 18 A node makes a lot of sense. In addition to your point about IBM getting sick of Samsung's EUV nodes not delivering the promised performance, there is also this: Intel got a chunk of money from the Feds specifically to serve as a "Secure Enclave". From what I understand, it's a further step from what used to be called "Trusted Foundry". IBM's Telums for its mainframes and Power CPUs for servers and workstations are heavily used by various governmental agencies and departments, not least the DoD. And even though South Korea is a close ally, it's still a lot easier to have a close eye on labs and fabs that are located stateside, never mind developing sensitive tech for the military.
Legacy fab can buy equipment that the advanced fab can not? I don't think you need to be an expert on loopholes to see that this was going to be exploited one way or another.
I am very confused regarding the future of x86 in laptops after the release of Qualcomm Oryon V2 chips and alliance of Nvidia and MediaTek to make an ARM chip for Windows, may be Nvidia is preparing for a future where ARM dominates. Plus, the entire AI and HPC compute is shifting purely from CPUs to custom accelerators. Do you think x86 is on the path of obsolesce? If this is true, AMD's future doesn't look bright, but Intel can hedge the decline of x86 to Intel Foundry plus Intel also has Mobileye but whether it will be successful or not that's a different thing.
AMD has Xilinx FPGA and DC GPU. Intel Foundry is like H2 2026 and 2027 real revenue. Yes, I do think x86 is in a position of weakness. The advisory committee exists out of fear.
My take is different. Yes, x86 is the weakest it has been maybe in 30-40 years,
but for the kind of extreme out of order CPU cores that are in use these days, microarchitecture matters more than ISA. Yes, a clean ISA is nice but it's fine x86 manages with its warts. In fact, it's great that there is finally choice and competition for PC users. Intel didn't field competitive designs till AMD caught up and now ARM is forcing both of them to become better at perf/power but x86 is not going to disappear or die. On a similar like process mode there is no reason why AMD or Intel should be significantly behind. BTW when AMD could not manage more than about 20% client CPU share with better designs, it is quite unlikely that ARM will take a majority share with a different ISA. There isn't even a pure perf competitive ARM server yet. AMD won't make as much money with x86, but they have a huge market waiting in DC GPUs.
This alliance between Team Red and Team Blue might also help to finally trim some of the truly dead weight from the x86 ISA. Both companies have individually talked about doing so off and on, but none of them wants to be holding the colostomy bag when the invariable pushback comes because someone's DOS Software isn't running so well anymore.
In that regard, I see Microsoft's move to Windows on ARM also as Redmond trying to shake off some of their own legacy burdens. Apple has been able to do just that several times (68xxx to PowerPC to x86/64 to Apple's own ARM designs) , because their customers are truly a captive audience, and because every forced ISA transition came with significant performance and performance/W improvements.
Agreed. Things like the x86S mode are great to remove all the legacy crap from the ISA. Anybody who needs to run old real mode or 16 bit stuff can easily run it emulated much faster than the hardware of the time. The x86 advisory committee is a good sign that AMD will adopt it along with APX. Although APX will take a while before it becomes widely used as it's not going to be widely targeted by software till most folks have the hardware.
Interestingly enough APX shows how little major ISA changes gets you. If load stores improve by 10-20%, then code that is not purely memory access dominated will likely improve no more than 5-10%. Again, shows that microarchitecture, process nodes and physical memory speeds dominate.
It’s because they are at 33-37% Q2 they were taking underutilization charges. They didn’t have to take them in q3 because the wrote down the equipment and “poth” they have no more underutilization charges. If they didn’t do the 3bn write down the margins would be were you expected. Pat is determined. Also I think he probably kitchen sinked the guide in q2 expecting to completely beat it but then only just
Onto feels like it is losing out vs competitors, given all other inspection and metrology companies have reported quite good numbers. Panel level packaging is still far out. Why are you still confident in the story?
Also, you might find this article interesting (link below) on Samsung's apparent plans to shut down (mothball) several of their EUV fabs all the way down to their 4 nm node due to very low demand. In light of this, AMD must have really gotten an incredible deal on fabbing some of their chips on Samsung's 4 nm node. The Capex on these fabs is enormous, and mothballing one is really something nobody wants to do, so AMD likely low-balled the heck out of them. Also means that Samsung is unlikely to be buying any more standard N.A. EUV equipment from ASML anytime soon.
IBM as the secret customer for Intel's 18 A node makes a lot of sense. In addition to your point about IBM getting sick of Samsung's EUV nodes not delivering the promised performance, there is also this: Intel got a chunk of money from the Feds specifically to serve as a "Secure Enclave". From what I understand, it's a further step from what used to be called "Trusted Foundry". IBM's Telums for its mainframes and Power CPUs for servers and workstations are heavily used by various governmental agencies and departments, not least the DoD. And even though South Korea is a close ally, it's still a lot easier to have a close eye on labs and fabs that are located stateside, never mind developing sensitive tech for the military.
Legacy fab can buy equipment that the advanced fab can not? I don't think you need to be an expert on loopholes to see that this was going to be exploited one way or another.
awesome piece as always. curious though, why do you think Apple internal 5G = more RFFE content for AVGO?
Qualcomm bundles RFFE with their modem. I see the RFFE content QCOM is about to lose (in addition to modem/baseband) as AVGO gain mostly.
Right, QCOM losing RFFE content makes complete sense. But that doesn't really impact QRVO right?
Honestly not quite sure what happened to QRVO now that more news is out. They clearly lost share to someone. Not sure who.
i love the way you explain why Intel advanced packaging is unpopular. There is a reason why its called Industry Standard.
I am very confused regarding the future of x86 in laptops after the release of Qualcomm Oryon V2 chips and alliance of Nvidia and MediaTek to make an ARM chip for Windows, may be Nvidia is preparing for a future where ARM dominates. Plus, the entire AI and HPC compute is shifting purely from CPUs to custom accelerators. Do you think x86 is on the path of obsolesce? If this is true, AMD's future doesn't look bright, but Intel can hedge the decline of x86 to Intel Foundry plus Intel also has Mobileye but whether it will be successful or not that's a different thing.
AMD has Xilinx FPGA and DC GPU. Intel Foundry is like H2 2026 and 2027 real revenue. Yes, I do think x86 is in a position of weakness. The advisory committee exists out of fear.
My take is different. Yes, x86 is the weakest it has been maybe in 30-40 years,
but for the kind of extreme out of order CPU cores that are in use these days, microarchitecture matters more than ISA. Yes, a clean ISA is nice but it's fine x86 manages with its warts. In fact, it's great that there is finally choice and competition for PC users. Intel didn't field competitive designs till AMD caught up and now ARM is forcing both of them to become better at perf/power but x86 is not going to disappear or die. On a similar like process mode there is no reason why AMD or Intel should be significantly behind. BTW when AMD could not manage more than about 20% client CPU share with better designs, it is quite unlikely that ARM will take a majority share with a different ISA. There isn't even a pure perf competitive ARM server yet. AMD won't make as much money with x86, but they have a huge market waiting in DC GPUs.
This alliance between Team Red and Team Blue might also help to finally trim some of the truly dead weight from the x86 ISA. Both companies have individually talked about doing so off and on, but none of them wants to be holding the colostomy bag when the invariable pushback comes because someone's DOS Software isn't running so well anymore.
In that regard, I see Microsoft's move to Windows on ARM also as Redmond trying to shake off some of their own legacy burdens. Apple has been able to do just that several times (68xxx to PowerPC to x86/64 to Apple's own ARM designs) , because their customers are truly a captive audience, and because every forced ISA transition came with significant performance and performance/W improvements.
Agreed. Things like the x86S mode are great to remove all the legacy crap from the ISA. Anybody who needs to run old real mode or 16 bit stuff can easily run it emulated much faster than the hardware of the time. The x86 advisory committee is a good sign that AMD will adopt it along with APX. Although APX will take a while before it becomes widely used as it's not going to be widely targeted by software till most folks have the hardware.
Interestingly enough APX shows how little major ISA changes gets you. If load stores improve by 10-20%, then code that is not purely memory access dominated will likely improve no more than 5-10%. Again, shows that microarchitecture, process nodes and physical memory speeds dominate.
It’s because they are at 33-37% Q2 they were taking underutilization charges. They didn’t have to take them in q3 because the wrote down the equipment and “poth” they have no more underutilization charges. If they didn’t do the 3bn write down the margins would be were you expected. Pat is determined. Also I think he probably kitchen sinked the guide in q2 expecting to completely beat it but then only just
*it’s because they were at 33-37% gm in q3, in q2 they were taking underutilization charges….
Onto feels like it is losing out vs competitors, given all other inspection and metrology companies have reported quite good numbers. Panel level packaging is still far out. Why are you still confident in the story?
Also, you might find this article interesting (link below) on Samsung's apparent plans to shut down (mothball) several of their EUV fabs all the way down to their 4 nm node due to very low demand. In light of this, AMD must have really gotten an incredible deal on fabbing some of their chips on Samsung's 4 nm node. The Capex on these fabs is enormous, and mothballing one is really something nobody wants to do, so AMD likely low-balled the heck out of them. Also means that Samsung is unlikely to be buying any more standard N.A. EUV equipment from ASML anytime soon.
Here the link:
https://biz.chosun.com/it-science/ict/2024/11/01/L2OXTRVJTVGM5JCTSOZMZ22DNA/