You don't need the same pool of talent to design a SoC as to design a CPU.As I said in another discussion in this thread, that is an irrelevant debate. If you boil down to "it's not a question of ISA", then the only question becomes in house talent. AMD has a ton there, so they still get my vote.
I have heard that for 20 years when Arm started raising the performance of its cores. "Arm in the server? LOL never. Apple switching to Arm? LOL not gonna happen." Yes it took time. Does it mean they will rule out x86? No, but they are surely displacing/complementing x86.And up to now, "cheaper for them" has simply not been anything but x86 yet. And I find no reason to think that the ARM world, who comes from smaller, lesser chips, is going to bridge the gap and leap above the x86 vendors.
Well, I say vendors, I mean AMD. Intel is dying due to its own problems.
And you're gonna stand to hear it for years yet. It's probably just as annoying to you as it is annoying to us to hear the "ARM will overtake x86 soon" for the past 10 years.I have heard that for 20 years when Arm started raising the performance of its cores. "Arm in the server? LOL never. Apple switching to Arm? LOL not gonna happen." Yes it took time. Does it mean they will rule out x86? No, but they are surely displacing/complementing x86.
You think Intel/AMD sells at cost? Obviously doing in house is cheaper if the R&D isn't sinking you. It's fine for all the hyperscalers.Baffled by claims that ARM isn't cheaper for hyperscalers... and if hyperscalers build their own ARM SoC's because it's cheaper, that doesn't count for some reason. Ooooookay.
You bear your username well, that's like waiting for the return of the Overmind.Whatever. I will continue to think the industry is fundamentally silly until they bring me my Itanium roadmap back. Still waiting on Kittson...
You think Intel/AMD sells at cost? Obviously doing in house is cheaper if the R&D isn't sinking you. It's fine for all the hyperscalers.
You bear your username well, that's like waiting for the return of the Overmind.
And your argument in favour of ARM would hold water if it was for "workloads, generic".It doesn't matter if AMD or Intel sell at cost. What matters in terms of gaining market share is what they're selling for externally. Clearly, judging by the three largest cloud providers opting to roll their own for specialised workloads that they fully own, that's cheaper.
So ARM is a "better deal" , that's what I'm saying. I'm not even sure anyone I've argued with is even clear on what we're arguing about.Being able to spend a couple of tens of millions of USD on licensing and taping out a SoC with IP that already exists, especially now that you can get it as a complete subsystem (Neoverse CSS), allows you to get incremental CPUs for the price of manufacturing wafers. That's innately difficult economics to compete with.
Burrowed Baneling philosophy.I'm a patient woman.
And your argument in favour of ARM would hold water if it was for "workloads, generic".
So ARM is a "better deal" , that's what I'm saying. I'm not even sure anyone arguing against me has had a point today.
"ARM works for its niches, nothing indicates that it'll do better or even equal x86 at generic CPU workloads" is all I'm saying.
Burrowed Baneling philosophy.
Appreciate the effort, but without a price tag attached to Graviton manufacturing costs, what the equivalent to Graviton would've cost if it were a Xeon/EPYC of same price, etc, it's not a super good metric.I feel obligated to point out that I've benchmarked three gens of Graviton on real applications (EDA synth flows) and found it compelling. Single-thread perf (mainly talking about 3/3E here) is great. Perf across whole chips (bare metal instances) is less great but very good.
They're selling VMs on Graviton in solid numbers. Customers are finding they do just fine at generic workloads.
It's a compliment. I've always admired the patience of GM players having spots with burrowed banes and somehow not missing their bombs in a game with hundreds of APM and schizoidly checking another corner of the map every .2 seconds...I've been called worse.
Appreciate the effort, but without a price tag attached to Graviton, what the equivalent to Graviton would've cost if it were a Xeon/EPYC of same price, etc, it's not a super good metric.
The day I see Graviton being picked OVER x86 not for internal AWS stuff but for your generic web service because it's flatly better than just using what works, then ARM will start having real wins in my book.
It's a compliment. I've always admired the patience of GM players having spots with burrowed banes and somehow not missing their bombs in a game with hundreds of APM and schizoidly checking another corner of the map every .2 seconds...
On the contrary, I don't think you're a liar. I just don't hear enough of these stories to really notice a rise. 99% of what I hear about ARM is "works for workloads A, B, C, D, E, rest is all x86" and tons of internal service instances.I know of someone - an ex, actually - running a couple dozen instances on Graviton for exactly that (PostgreSQL and a Rails app.) But that's just anecdote and you have no particular reason to believe it.
Now now, don't judge Bezos for those fake compute allocations to make you pay for an unused AWS account, or for these surprising "errors" when you want to cancel a Prime subscription after clicking "cancel" 3 times on 3 different pages with 3 increasingly better hidden "yes I want to cancel" buttons.All I can say is that I'm seeing real Graviton use in my industry, and I sure as hell don't work for Amazon. (In general, I avoid them like the plague - likewise Microsoft, Google, and cloud stuff in general. I am old-fashioned.)
Before that point the thread and most posts including mine were squarely about client PC in the form of Windows PCs.Beyond laptops, hyperscalers use of Arm is another thing to consider. Desktop is another discussion (I wouldn't consider anything but AMD here), but all other markets are seeing a push for Arm. Except for Apple, Arm laptops might prove a complete failure, but I doubt we'll see hyperscalers drop Arm in the medium term.
Arm provides the possibility for hyperscalers to design their own SoC using a good enough core from Arm. That's a definitive advantage over AMD and Intel both from the flexibility point of view and from the cost point of view (there likely is a TCO advantage but given that no official figure for hyperscaler Arm SoC power consumption is known, it's hard to say). And if anyone thinks the hyperscalers are buying the "RISC hype", they have no clue what they are talking about.
So sorry for Arm bears and x86 apologists, I doubt Arm will decline soon. And that applies to x86 too, it's alas here to stay for way too long.
I plead guilty. I was basing my answer on the thread title, but indeed re-reading the first post, the focus was on client PC. My apologies.I thought this thread was about the client PC market and ARM's chances to take that over? Seems it's about hyperscalers now.
I mostly agree, but I won't digress anymoreRegarding servers, especially hyperscalers, the environment has been a completely different one anyway. While Intel originally had a similar dominance, Microsoft didn't and doesn't so open source is way more widespread. So proprietary software was and is mainly limited to Intel-specific optimizations. Opening that up to AMD ironically thus also makes porting to other ISAs easier.
The first ARM based servers were all developed in reaction to Intel's monopoly at the time, but development on ARM server slowed down a lot since AMD started competing again. What survived from that time are mainly the in-house solutions of cloud elephants like AWS which can afford the ongoing R&D involved. Open market ARM solutions for servers are still in a dire state in comparison.
I know they are really strong overall, but they have no need to invest in compilers and the like for things outside of its market domain.Apple's cores have been well-characterized across a wide range of workloads. They are serious cores, with superb general-purpose performance. This isn't a great take IMO.
I am not sure what you're trying to say. Apple compiler is based on LLVM (they basically funded its development). Do you think LLVM only targets "Apple market domains" (whatever that means)?I know they are really strong overall, but they have no need to invest in compilers and the like for things outside of its market domain.
No, Apple does a damn good job to not have weak exceptions to the rule.I am not sure what you're trying to say. Apple compiler is based on LLVM (they basically funded its development). Do you think LLVM only targets "Apple market domains" (whatever that means)?
Baffled by claims that ARM isn't cheaper for hyperscalers... and if hyperscalers build their own ARM SoC's because it's cheaper, that doesn't count for some reason. Ooooookay.
Whatever. I will continue to think the industry is fundamentally silly until they bring me my Itanium roadmap back. Still waiting on Kittson...
This kind of “they’re only good for a few tasks” remind me of when the M1 came out and you saw takes like “the P cores are like ASICS for macOS”.Apple's cores have been well-characterized across a wide range of workloads. They are serious cores, with superb general-purpose performance. This isn't a great take IMO.
The thing people miss about Arm is this:You don't need the same pool of talent to design a SoC as to design a CPU.
I have heard that for 20 years when Arm started raising the performance of its cores. "Arm in the server? LOL never. Apple switching to Arm? LOL not gonna happen." Yes it took time. Does it mean they will rule out x86? No, but they are surely displacing/complementing x86.
Irrelevant, how many times am I gonna read that today?The fastest ST today and most efficient is on Arm.
Just as irrelevant, Intel is failing as a company. All their products are duds or barely make the cut, they had to boost power draw beyond sane limits, they keep having increasing trouble to compete.We will soon have two core lineups competing with Intel/AMD on Windows - Qualcomm’s and Arm’s Cortex — via QC, MediaTek and Nvidia.
By contrast I remember in 2013 with phones and tablets people kept saying “If anything Arm has to worry, Intel is gonna take low power serious!1!1!”
The ARM hype is not any more justified now that it was then. Nobody's saying ARM "can't catch up", we're saying there is no sane reason to believe that it will overtake x86, on the contrary the odds are much stronger that they'll fall a bit short due to x86 just having that much more experience in really heavy stuff.Shows how that landed.
The Arm hype was largely justified, it was just too soon in almost every case, be it servers, Apple, WoA, Cortex matching Intel IPC (surpassed them now fwiw), etc. These things are tough. But not insurmountable.
I thought this thread was about the client PC market and ARM's chances to take that over? Seems it's about hyperscalers now.
Before that point the thread and most posts including mine were squarely about client PC in the form of Windows PCs.
Regarding servers, especially hyperscalers, the environment has been a completely different one anyway. While Intel originally had a similar dominance, Microsoft didn't and doesn't so open source is way more widespread. So proprietary software was and is mainly limited to Intel-specific optimizations. Opening that up to AMD ironically thus also makes porting to other ISAs easier.
The first ARM based servers were all developed in reaction to Intel's monopoly at the time, but development on ARM server slowed down a lot since AMD started competing again.
Agreed on this part.What survived from that time are mainly the in-house solutions of cloud elephants like AWS which can afford the ongoing R&D involved. Open market ARM solutions for servers are still in a dire state in comparison.
A consumer doesn’t care why. They just care what the final product looks like. Qualcomm is showing they can beat AMD sell to some merchant providers. They even got the XPS lineup, which AMD never could do. I doubt Zen 5 will get them in the XPS.Irrelevant, how many times am I gonna read that today?
AMD doing a tiny cheap core with Z2/Z4 isn't meant to compete with the fat cores Apple put out. If AMD put a huge core out vs Apple's huge core, and they both sold at Apple prices with Apple production costs, then this comparison would make some sense.
Intel is a clusterf&@& yes, though AMD probably won’t have a LNL competitorAnd this comparison is possible in about 2 weeks so please just stop with the danged Zen 4 comparisons.
Just as irrelevant, Intel is failing as a company. All their products are duds or barely make the cut, they had to boost power draw beyond sane limits, they keep having increasing trouble to compete.
Actually, that used to be exactly what people said.The ARM hype is not any more justified now that it was then. Nobody's saying ARM "can't catch up", we're saying there is no sane reason to believe that it will overtake x86, on the contrary the odds are much stronger that they'll fall a bit short due to x86 just having that much more experience in really heavy stuff.
First part, yes, latter part, no, not because of X86, but because of AMD’s priorities and competencies.Put down a fat ARM core vs a fat x86 core, see that neither of them will outclass the other in terms of perf or efficiency,
Zen 5 will be exciting.and this entire hype will have proven to be hot air.
And we're getting that within 2 weeks.
Conclusion: Implementation Matters, not ISA
I’m excited to see competition from ARM. The high end CPU space needs more players, but ARM players aren’t getting a leg up over Intel and AMD because of instruction set differences. To win, ARM manufacturers will have to rely on the skill of their design teams. Or, they could outmaneuver Intel and AMD by optimizing for specific power and performance targets. AMD is especially vulnerable here, as they use a single core design to cover everything from laptops and desktops to servers and supercomputers.
That’s where we want to see the conversation go. Hopefully, the info presented here will avoid stuck-in-the-past debates about instruction sets, so we can move on to more interesting topics.
By this do mean the size of a core? ie it’s areaPut down a fat ARM core vs a fat x86 core
Have you considered the possibilty that what you are hyping up could also be hot air?Put down a fat ARM core vs a fat x86 core, see that neither of them will outclass the other in terms of perf or efficiency, and this entire hype will have proven to be hot air.
And we're getting that within 2 weeks