I assume you meant WoA exclusivity + platform instead of ARM exclusivity + platform?
I assume you meant WoA exclusivity + platform instead of ARM exclusivity + platform?The IPC uplifts of Cortex-X5 show ARM is getting more powerful and with ARM's exclusivity ends with Qualcomm, I see almost all CPU vendors are going to join ARM platform......except Intel and that's another point
Its the closer thing they have, they just completely lost the low power market to Intel small cores after Baytrail as they could never make their cores low power enoght.I assumed Mendocino was about low cost, high volume rather than low power.
More about competing with lower end Intel SKUs in the Chromebook market and the like.
To be fair they stopped even trying after Jaguar because of the issues with Bulldozer necessitating that they concentrate on just getting their performance µArch back up to IPC competitive status with Intel.Its the closer thing they have, they just completely lost the low power market to Intel small cores after Baytrail as they could never make their cores low power enoght.
Why don't they just limit the frequency of Zen 4c/5c cores to max 2 GHz? That should keep the cores in the sweet spot of power efficient performance. I bet they could go fanless with that strategy.some variant of zen even more cut down than the C cores
The lower end APUs like Phoenix 2 already seem to be chasing that strategy by also lowering the GPU CU count considerably.Why don't they just limit the frequency of Zen 4c/5c cores to max 2 GHz? That should keep the cores in the sweet spot of power efficient performance. I bet they could go fanless with that strategy.
The problem was more that Intel was dumping Bay Trail on the market with so many discounts that it had negative revenue. Go look up "Intel contra-revenue". Intel was trying desperately to fight the ARM tablet market and was throwing money and manufacturing volumes at the problem, it made it a completely unprofitable market for AMD to target.Its the closer thing they have, they just completely lost the low power market to Intel small cores after Baytrail as they could never make their cores low power enoght.
January 2016+ is when AMD got the Cat cores out of exclusivity with GlobalFoundries.Between that and the failure of TSMC's 20nm node killing Nolan, AMD just gave up on the market and focused on Zen.
Its not just that i had (and i still have somewhere at work) old AMD ITX boards and laptops and their small cores were just bad. Something like the E2-3800 using i think double the power (or more) than a J4105, when intel launched Baytrail they placed socs into tablets with regular battery sizes and aluminium paper sheets as heat spreader in a blink of an eye. It was not just subsiding, they actually had the hardware for it. (i still have my Z3735F tablet somewhere)The problem was more that Intel was dumping Bay Trail on the market with so many discounts that it had negative revenue. Go look up "Intel contra-revenue". Intel was trying desperately to fight the ARM tablet market and was throwing money and manufacturing volumes at the problem, it made it a completely unprofitable market for AMD to target.
Between that and the failure of TSMC's 20nm node killing Nolan, AMD just gave up on the market and focused on Zen.
Coz it would be smaller, simpler and cheaper.If Zen5 is so good, why does AMD need an ARM APU?
This (along with PRC beating its chest over Taiwan) is clearly causing TSMC to become more expansionist - along with the CHIPS act funding subsidizing a new fab in the US.As more companies move to in-house designs, competition for TSMC wafers are expected to increase significantly Yes. Everyone wants to use TSMC to manufacture their in house designs. Many AI accelerator startups are also using TSMC. Competition is higher for TSMC wafers for everyone.
I wrote this in 2021 and updated with commentary in April 2024.
Original post: AMD_Stock/comments/1c14rm7/reviewing_my_amd_vs_competition_predictions_from
Original post (2021) + comments today (Apr 2024 bolded):
ARM Attack on consumer x86 market:
ARM Attack on hyperscalers market:
- x86 market shrank by 10% the moment Apple announced a full transition to ARM
- Even worse, Apple is the forcing function for PC software makers to support ARM, paving the way for Windows ARM Windows ARM is as serious as ever in 2024 with Qualcomm, Mediatek, Samsung releasing better SoCs and Nvidia rumored to join soon
- Microsoft has just announced x86 emulation on their Windows ARM OS It's now core to Windows
- Apple's entry-level M1 is faster than 4900HS by 55% in single-core and 3.7% in multi-core while using 5x less power. You can buy an M1 laptop today while the 4900HS is a highly binned-part.
- Apple's entry-level iGPU blows away the very best AMD iGPU Base M3 GPU still does
- Apple's ARM chip allows the entry-level Macbook Pro to have 20 hours of battery life while staying extremely cool AMD has not achieved the same level of efficiency
- Apple is readying 8,16, 32 core SoCs. For reference, M1 has only 4-high performance cores. "M Extreme" never released
- Apple is poised to significantly increase PC market-share, 100% within 3 years Can't confirm
- In order to compete, PC makers like Dell, Lenovo, and HP will increasingly look to ARM alternatives Absolutely true. See all OEMs that signed up for Qualcomm Snapdragon X Elite platform.
- Microsoft is working on in-house ARM chips for Surface Still rumored but they will use X Elite.
- Qualcomm is expected to become a serious player in the laptop space Seems more and more like it
ARM Attack on small size cloud companies:
- AWS's Graviton2 workload increased by 10x last year. In just one year, Graviton2 is now 10% of all AWS CPU workloads. Graviton is now the majority of AWS
- Like Apple in consumer market, Amazon is the forcing function for better ARM server software support Almost all popular server apps now have ARM versions
- Amazon will most certainly prioritize in-house designs over x86 chips. In the future, Amazon will likely offer x86 only for legacy software. TBD
- AWS controls 40% of the cloud. If AWS moves most of its future workload to its own CPUs, then AMD can control at best 60% of what Intel used to contol.
- In house ARM chips allow hyperscalers to differentiate, save on cost, and build chips tailored to their unique challenges Yes.
- Microsoft is designing its own ARM cloud chips They're using Ampere but rumors are that they too, is designing custom ARM server chips
- You can expect Google, Baidu, Tencent to follow soon Google just announced Axion server CPUs yesterday. Chinese cloud companies are diversifying to RISC-V
- Hyperscalers don't want to be controlled by a duopoly. If Intel isn't grabbing them by the balls, then it's just AMD. It's the same thing. Hyperscalers want to control their own destiny and ARM allows them to.
GPUs:
- Ampere will allow smaller cloud companies to buy into ARM Oracle cloud is an example that bought a lot of Ampere chips
- Anandtech just said that Ampere's latest 80-core chip is 42% better than Epyc in terms of cost/performance
- Ampere is releasing a 128-core chip in 2021
- Anandtech says Ampere's 2022 N1 Neoverse chip could be 50% faster at minimum
- Very well-funded startup Nuvia is also competing here Nuvia was bought by Qualcomm and switched to laptop first. But Qualcomm has plans to bring Nuvia designs to server later.
Consoles:
- Apple will no longer use AMD GPUs in their computers. Bloomberg reports that Apple is testing 64-core and 128-core GPUs. For reference, M1 has 8 GPU cores. "M Extreme" never came out.
- Nvidia's grip on server GPUs and AI is tight and AMD has a mountain to climb in order to catchup True. Nvidia is now a $2 trillion company.
- With ARM, Nvidia can now offer complete server units from CPU to GPU to interconnects, just like AMD True. Nvidia announced Grace CPU and ditched Epyc and Xeon in their stack.
- Intel is joining the cloud GPU competition, and they're expected to use TSMC to manufacture Xe GPUs. True but Intel is having limited success.
TSMC:
- Mobile gaming is now 3x bigger than consoles
- Consoles are a low-margin business
- It's quite possible that PS5/Xbox Series X are the last consoles ever if AAA gaming moves to the cloud in the next 6-7 years. I'm wrong here but it's not clear that Microsoft will continue to use AMD chips in their next-gen Xbox.
- As more companies move to in-house designs, competition for TSMC wafers are expected to increase significantly Yes. Everyone wants to use TSMC to manufacture their in house designs. Many AI accelerator startups are also using TSMC. Competition is higher for TSMC wafers for everyone.
- Apple is likely to continue to hold a node advantage lead because they have a lot more cash and a lot more volume. Apple's volume will likely increase significantly with Apple Silicon Macs. Ming-Chi Kuo expects Apple to sell 35 million Macs/year by 2023. True. Apple is first to 3nm and likely first to 2nm as well.
- Unlike Intel, AMD does not have a node advantage over Qualcomm, Ampere, Nuvia, Nvidia, Amazon, Microsoft, etc. True.
- Intel is expected to make a decision on whether they will invest in their 7nm process node in Spring of 2021 or use TSMC. If Intel decides to go fabless instead, AMD will lose its node advantage over Intel and wafer prices will increase for everyone. Intel went the IFS + TSMC strategy.
It was a projection made by Ming Chi Ku. I don't have the statistics. Feel free to prove that point wrong.Honestly this reads like an ARM/Apple advert. Apple increase market share by 100% in three years? Yea you can't confirm because the is nonsense. Apple is a walled garden and that tight integretion allows them to extract performance at the cost of freedom of choice.
From what I remember GCN had terrible idle power, something AMD worked hard to improve in RDNA.Something like the E2-3800 using i think double the power (or more) than a J4105
Apple is poised to significantly increase PC market-share, 100% within 3 years Can't confirm
- Microsoft is working on in-house ARM chips for Surface Still rumored but they will use X Elite.
- AWS's Graviton2 workload increased by 10x last year. In just one year, Graviton2 is now 10% of all AWS CPU workloads. Graviton is now the majority of AWS
- In house ARM chips allow hyperscalers to differentiate, save on cost, and build chips tailored to their unique challenges Yes.
- Microsoft is designing its own ARM cloud chips They're using Ampere but rumors are that they too, is designing custom ARM server chips
- You can expect Google, Baidu, Tencent to follow soon Google just announced Axion server CPUs yesterday. Chinese cloud companies are diversifying to RISC-V
ARM Attack on small size cloud companies:
- Ampere will allow smaller cloud companies to buy into ARM Oracle cloud is an example that bought a lot of Ampere chips
- Anandtech just said that Ampere's latest 80-core chip is 42% better than Epyc in terms of cost/performance
- Ampere is releasing a 128-core chip in 2021
- Anandtech says Ampere's 2022 N1 Neoverse chip could be 50% faster at minimum
- Very well-funded startup Nuvia is also competing here Nuvia was bought by Qualcomm and switched to laptop first. But Qualcomm has plans to bring Nuvia designs to server later.
- It's quite possible that PS5/Xbox Series X are the last consoles ever if AAA gaming moves to the cloud in the next 6-7 years. I'm wrong here but it's not clear that Microsoft will continue to use AMD chips in their next-gen Xbox.
I never said X Elite is inhouse. But their next Surface lineup is prioritizing Elite X before Ryzen.X Elite is not in-house. Neither is the Nvidia ARM chip rumoured to come, neither again is the AMD ARM SoC. So... if they're doing this, it's going to be 2027+.
It was already 49% of new EC2 instances by 2020. Graviton4 is out.Can't remember the numbers off the top of my head, but I'm very certain this is untrue from both a new instances and overall instances perspective. Pretty sure when it comes to new instances Graviton still sits at significantly below the 50% mark at AWS, and then obviously from the overall perspective that number is much lower still.
I don't understand this point.Which is why the only other company with in-house ARM chips - Oracle - is doing so well, right?
Oh wait, Ampere is freaking out because AmpereOne is such a disaster, and Oracle is now holding onto that bag of chips wondering what the hell to do with it. Seriously though Google is the only other one with their own solution (for now). So it's very early days to treat this like an industry-wide fact. The bigger problem with these initiatives isn't making something custom in the first place, it's maintaining and developing newer solutions in the future that's the issue.
Why not? Ampere is used by Microsoft, Oracle, and smaller clouds like hetzner.*chuckling* yeah the whole Ampere thing is going great for them
They also make ARM server chips.The Chinese companies have a very different incentive behind diversifying - the issue isn't Intel/AMD, it's the threat of sanctions preventing them from getting non-homegrown parts. That's also why they went towards RISC-V instead of ARM.
No they don't. They invested in Ampere.Oracle own Ampere.
Waiting for your analysis.That's compared to Rome, and also isn't even a proper TCO comparison. Try and compare the TCO for a Milan or Genoa part. Comparing perf/$ with MSRPs isn't even remotely accurate for what any "small cloud company" will be paying for it
You don't think QC is going to make a server chip with their Oryon cores? wow.That's what QC has claimed, but we don't have an actual timeframe for it. Not even in rumour-land. So really it's a bit of a nothing statement for the time being.
Amazon and Google are doing their own in-house Arm SoC for their hyperscaler needs.Which is why the only other company with in-house ARM chips - Oracle - is doing so well, right?
You cite Graviton and then forget it's done in-house the same way Axion is done in-house.Seriously though Google is the only other one with their own solution (for now). So it's very early days to treat this like an industry-wide fact. The bigger problem with these initiatives isn't making something custom in the first place, it's maintaining and developing newer solutions in the future that's the issue.
No, they don't.Oracle own Ampere.
I never said X Elite is inhouse. But their next Surface lineup is prioritizing Elite X before Ryzen.
It was already 49% of new EC2 instances by 2020. Graviton4 is about to come out.
View attachment 96833
I don't understand this point.
Why not? Ampere is used by Microsoft, Oracle, and smaller clouds like hetzner.
You don't think QC is going to make a server chip with their Oryon cores? wow.
Evidently you need to learn what the word other means.Amazon and Google are doing their own in-house Arm SoC for their hyperscaler needs.
Unless if you are talking about the Arm core itself in which case Oracle isn't doing it in-house since Ampere doesn't belong to Oracle contrary to your claim.
You cite Graviton and then forget it's done in-house the same way Axion is done in-house.
No, they don't.
You really need to refresh your knowledge about Arm ecosystem before making claims.
Indeed. Nuvia had a server Arm Soc ready and QC decided not to use it and ask Nuvia team to take their core and plug'em into a consumer SOC.Nuvia was created to create server products, where are they now?
Will QC eventually create server chips? Of course I can't say they never will. What I can say though, is there's currently no indication that they're coming soon.
Qualcomm plans to put Oryon cores everywhere. They're just starting with laptops first because they see a clear opportunity where AMD/Intel have failed to compete with Apple Silicon.Indeed. Nuvia had a server Arm Soc ready and QC decided not to use it and ask Nuvia team to take their core and plug'em into a consumer SOC.
If they really wanted a server Arm Soc, they'd have released it immediately.
Again, that's an article from 2 years ago. At that point, it was already long since rumoured that 8g4 would use an iteration of the core we'll see in X Elite.Qualcomm plans to put Oryon cores everywhere. They're just starting with laptops first because they see a clear opportunity where AMD/Intel have failed to compete with Apple Silicon.
Next, they'll put Oryon on phones. Then servers.
Qualcomm's Oryon cores will come to the Snapdragon 8 Gen 4
Qualcomm has announced that its computing Oryon cores will be coming to mobile in 2024.www.xda-developers.com