News ARM Server CPUs

[Gideon]

With all the upcoming ARM servers (Let's not forget Nuvia, etc) it probably makes sense to have 1 thread from them all, instead of creating new ones for each announcement (if Not, I will rename it).

However:
Anandtech: Marwell Announces 3rd Gen Arm Server Thunder X3: 96 Cores/384 threads
ServeTheHome: Marvell ThunderX3 Arm Server CPU with 768 Threads in 2020

Could be pretty impressive, though 25% Single Threaded performance gain seems a bit meh, compared to X2, which @2.5Ghz was ~50% slower than Xeon @ 3.8Ghz

But Their marketing slides sure show potential:

 

[Doug S]

Maybe, maybe not.
For one thing we have no idea what the transition plan for v9 is.
They COULD have a temporary period during which devices can decode both v8 (hopefully 64-bit only!) and v9 instructions. How feasible this is depends on how ambitious they are in redesigning v9.

Why would they do that, when what they did with ARMv7/ARMv8 was perfect: treat them as separate ISAs, and allow for cores that either run both or run ARMv8 only. Why would they need to define a "transition period", rather than letting the customers decide what's best for them and their market?

Since there's essentially zero commercial (closed source) ARMv8 binaries, an ARM server vendor could quite reasonably go ARMv9 only from day one. Why should they pay the price of a transition period? Why should Android OEMs that serve a market that goes as low as sub $50 phones be forced into a transition before it makes sense for them?

I'm still skeptical there IS an ARMv9, since AFAIK the only evidence was a job posting mentioning that. That might be something they are targeting for 2025, and we'll see several more ARMv8 iterations.

 

[name99]

Why would they do that, when what they did with ARMv7/ARMv8 was perfect: treat them as separate ISAs, and allow for cores that either run both or run ARMv8 only. Why would they need to define a "transition period", rather than letting the customers decide what's best for them and their market?

Since there's essentially zero commercial (closed source) ARMv8 binaries, an ARM server vendor could quite reasonably go ARMv9 only from day one. Why should they pay the price of a transition period? Why should Android OEMs that serve a market that goes as low as sub $50 phones be forced into a transition before it makes sense for them?

I'm still skeptical there IS an ARMv9, since AFAIK the only evidence was a job posting mentioning that. That might be something they are targeting for 2025, and we'll see several more ARMv8 iterations.

You're AGREEING with me!
My point was that
- continuing to advance v8
- does not mean those advances TAKE THE PLACE OF v9
which was the comment to which I was replying.

v8 and v9 could serve different roles and different markets going forward for a few years, even in v9 appears next year.

Andrei has been adamant that v9 is imminent (ie way before 2025) but of course he may have have misinterpreted whatever it was that gave him this confidence.

 

[soresu]

Not got a subscription, so I don't know who it is. My guess would be Marvell?

Yes, the problem is that Neoverse seems to be killing it.

Given this fact just as with Qualcomm and now Samsung on the client side (and nVidia on the Tegra side as of Oren) there is probably a growing thought that custom server core development is simply more trouble/expense than it is worth.

 

[jpiniero]

When it comes to ARM Servers, Amazon is going to be tough to beat. If you can use ARM at this point, you are running Linux, and are most likely okay with the cloud...

 

[name99]

Yes, the problem is that Neoverse seems to be killing it.

Given this fact just as with Qualcomm and now Samsung on the client side (and nVidia on the Tegra side as of Oren) there is probably a growing thought that custom server core development is simply more trouble/expense than it is worth.

Or to put it differently, common sense has finally taken over!
It was always the case that these separate ARM companies (and so many of them) made zero economic sense. ARM in servers ala Neoverse, yes, that makes a huge amount of sense; but not other companies doing the exact same thing, it's just wasted resources.

Amazon is basically a more sensible model (or, alternatively, think Apple A4 and A5) -- ARM does one part of the job with the CPU and bus designs, TSMC/SS does fabbing, AMZ (or Apple back then) puts together the exact package of functionality they need.

Honestly I'm not optimistic for Nuvia. Having the ambition to be faster than ARM is great, but you need customers who are willing to pay more for that extra speed, you need to manage relationships, you need to be able to provide any extra weirdness that might be desired from NPUs to accelerators to security features. I can't see how Nuvia is big enough to both design a kickass CPU+SoC AND do the sort of handholding and customer management that's required when a Google or a Facebook come by considering buying the chips but asking for one extra thing that will really help us out.

Well, we'll see. I'm sure whatever happens they will land on their feet, acquired by Apple, Google, or nV.

 

[name99]

Or to put it differently, common sense has finally taken over!
It was always the case that these separate ARM companies (and so many of them) made zero economic sense. ARM in servers ala Neoverse, yes, that makes a huge amount of sense; but not other companies doing the exact same thing, it's just wasted resources.

Amazon is basically a more sensible model (or, alternatively, think Apple A4 and A5) -- ARM does one part of the job with the CPU and bus designs, TSMC/SS does fabbing, AMZ (or Apple back then) puts together the exact package of functionality they need.

Honestly I'm not optimistic for Nuvia. Having the ambition to be faster than ARM is great, but you need customers who are willing to pay more for that extra speed, you need to manage relationships, you need to be able to provide any extra weirdness that might be desired from NPUs to accelerators to security features. I can't see how Nuvia is big enough to both design a kickass CPU+SoC AND do the sort of handholding and customer management that's required when a Google or a Facebook come by considering buying the chips but asking for one extra thing that will really help us out.

Well, we'll see. I'm sure whatever happens they will land on their feet, acquired by Apple, Google, or nV.

There's an interesting question here that, as far as I know, remains unanswered: what was the ARM server timeline?
We know ARM has been pushing a particular timeline for its presence in servers since at least 2015. There were earlier timelines that could be mocked, but starting around 2015 they began to be, IMHO, realistic, hoping for a substantial v8 based presence beginning in 2020, which is about what we saw, with Neoverse announced in 2018, and AMZ really taking advantage of this in 2019.

OK, so against that background, consider Centriq. That was announced as an idea in 2014, SoC demoed late 2016, shipped late 2017, then soon cancelled. Cancelling was probably a good idea, given what we know of Neoverse, but the interesting political question is who knew what when...
Was ARM planning Neoverse quietly for years, since 2014 or so?, and just not informing the partners?
Did the partners think (and to be fair, this was not an unreasonable thought at the time, until Apple kicked ARM's butt) that ARM had a history of cores that were great at area+power, not so much at performance, so they had an opening?
Or was ARM willing to leave servers in partner hands until some late point (2016?) by which time they were so disappointed at everything they'd been shown that they concluded "to hell with these people and their companies, they all deserve to fail! If you want something done right do it yourself."

 

[DrMrLordX]

@name99

Honestly I think ARM expected people to do what Amazon did: license the design diretly from ARM (in this case, NeoVerse), maybe tweak it a little, and sell that to the public. I don't think ARM ever intended for players to just drop out of the race.

 

[Thala]

There is an opportunity to develop you own cores but license other IP like the mesh-interconnect used in NeoVerse. You can license lots of the building blocks separately. Thats the direction NUVIA is going.

 

[scannall]

There is an opportunity to develop you own cores but license other IP like the mesh-interconnect used in NeoVerse. You can license lots of the building blocks separately. Thats the direction NUVIA is going.

Just my own opinion/speculation but it looks to me like Nuvia is building IP and making noise for a high dollar buyout. As an actual business model it's pretty weak. Look at how resistant the market is just switching vendors using the same ISA, let alone going to ARM or whatever requiring rebuilds of most of your software and licenses.

 

[NTMBK]

Looks like it was Marvell:

https://twitter.com/x/status/1322253556364140545

 

[DrMrLordX]

So how many ARM server offerings are left outside of Graviton2 (cloud only) and A64FX (extremely niche)? Including ones that are supposed to be coming up real soon now.

 

[jpiniero]

So how many ARM server offerings are left outside of Graviton2 (cloud only) and A64FX (extremely niche)? Including ones that are supposed to be coming up real soon now.

Makes sense to get out now with nVidia buying ARM.

 

[DrMrLordX]

Makes sense to get out now with nVidia buying ARM.

Well yes, it does. I'm just wondering who is still left? Ampere and . . . ?

 

[NTMBK]

Well yes, it does. I'm just wondering who is still left? Ampere and . . . ?

Nuvia are still in it. And don't be surprised if the other hyperscalers are building their own Graviton-style chip, with ARM IP and a few custom tweaks.

 

[DrMrLordX]

Nuvia are still in it. And don't be surprised if the other hyperscalers are building their own Graviton-style chip, with ARM IP and a few custom tweaks.

Such as Google? They won't market anything like that, though. It'll be in-house only. Just like Amazon.

 

[trexfromouterspace]

There's a very good chance Microsoft is making an in-house platform (presumably ARM-based) for Azure. They've been hiring IC designers like crazy.

 

[LightningZ71]

I wonde if someone will buy the Thunde X3 design?

 

[trexfromouterspace]

I wonde if someone will buy the Thunde X3 design?

I have 2 dollars and 79 cents. Take it or leave it.

 

[DrMrLordX]

I have 2 dollars and 79 cents. Take it or leave it.

I got tree fiddy

 

[Gideon]

Anandtech has an excellent article about Neoverse V1:

This slide definitely shows better than most others that AMD and Intel need a solid long-term plan to counter ARM (It includes Q1 2021 btw).

1. Hyperscalers are > 50% of the entire enterprice market, and they are integrating vertically into ARM surprisingly fast, you can't just ignore it.
2. Apple is gone (for x86) and Qualcomm (and Samsung it seems) are coming to laptops next year.
3. Nvidia is pushing ARM gaming, it's still in babysteps but in ~2023 timeframe they might have some solid offerings around.
4. previous point means that consoles actually have a decent chance of moving to ARM as well.

I'm not sayng all of the above will 100% pan out, but IMO just trying to out-execute competition by improving x86 processors doesn't seem to be enough in the longer term (5 - 10 year timeline).
I wonder what is? Start investing in RISC-V infrastructure looks like an avenue.

 

[ThatBuzzkiller]

Anandtech has an excellent article about Neoverse V1:

This slide definitely shows better than most others that AMD and Intel need a solid long-term plan to counter ARM (It includes Q1 2021 btw).

1. Hyperscalers are > 50% of the entire enterprice market, and they are integrating vertically into ARM surprisingly fast, you can't just ignore it.
2. Apple is gone (for x86) and Qualcomm (and Samsung it seems) are coming to laptops next year.
3. Nvidia is pushing ARM gaming, it's still in babysteps but in ~2023 timeframe they might have some solid offerings around.
4. previous point means that consoles actually have a decent chance of moving to ARM as well.

I'm not sayng all of the above will 100% pan out, but IMO just trying to out-execute competition by improving x86 processors doesn't seem to be enough in the longer term (5 - 10 year timeline).
I wonder what is? Start investing in RISC-V infrastructure looks like an avenue.

2. Apple's efforts don't benefit anyone except for their own platforms
3. AAA games will never catch on ARM because most big studios don't support ARM architectures on their proprietary engines
4. Virtually no chance of this ever happening since at least 2 major console vendors now care about backwards compatibility and their most recent platforms use x86 CPUs too

ARM may have a more accessible ISA license but it doesn't get anymore contributions outside of the licensing fees. There is no path for ARM to universally succeed since there are self-absorbed corporations like Apple who don't care about participating in industry standards with their most recent 5G debacle as an indication and Nvidia does not share their technology if ever. If other ARM vendors did care about ARM architecture in general then we'd have Apple licensing their CPU designs for others and Nvidia licensing their GPUs out to other ARM vendors at reasonable rates ... (anyone can tell that neither Apple or Nvidia has many friends in the industry)

Instead what we have is ARM vendors like Samsung cooperating (?!) with x86 vendors such as AMD and Intel opening up their fab services to other ARM vendors as well which suggest that either x86 vendors are more trustworthy in general or ARM vendors have very bitter relationships towards each other ...

 

[Thala]

Instead what we have is ARM vendors like Samsung cooperating (?!) with x86 vendors such as AMD and Intel opening up their fab services to other ARM vendors as well which suggest that either x86 vendors are more trustworthy in general or ARM vendors have very bitter relationships towards each other ...

Either that or perhaps its just that Intel happens to have fabs and AMD happens to have a good GPU architecture - both of which many others simply could not offer.

 

[Thala]

Anandtech has an excellent article about Neoverse V1:

This slide definitely shows better than most others that AMD and Intel need a solid long-term plan to counter ARM (It includes Q1 2021 btw).

They do not need to counter ARM - they can offer their own ARM designs to the market as an option.

 

[ThatBuzzkiller]

Either that or perhaps its just that Intel happens to have fabs and AMD happens to have a good GPU architecture - both of which many others simply could not offer.

Intel isn't the only IDM offering fab capacity and AMD isn't the only one to have good GPU architectures either. Many ARM vendors simply choose to not share their technology like Apple or Nvidia and instead act like petty children if other corporations are involved. ARM vendors don't understand the value of interoperability and try to find some way to actively sabotage it. At least x86 vendors know how to share responsibility with their partners whether friend or foe like any other adults ... (this is reflected in their age since AMD and Intel are older than most ARM vendors and had their share of dark ages too)

Apple wants people to target their walled gardens at everyone else's expense ...
If you want to use CUDA then this doesn't help anyone but Nvidia ...
Qualcomm has been accused of bundling their baseband technology with their SoCs to limit the amount of competition in specific countries ...

The only thing they have in common are their toxic relationships with their partners so how is ARM supposed to threaten traditional x86 markets if they have very little respect among others in the industry ?

 

[Gideon]

4. Virtually no chance of this ever happening since at least 2 major console vendors now care about backwards compatibility and their most recent platforms use x86 CPUs too

Yes not right now, but during the next cycle in ~5 years they very well might.

IMHO this is a weak argument because:

1. The console vendors have always broken backwards compatiblity when it offered decent performance gains.
2. By next gen they should be easily able to run current gen games under emulation (especially if it's an AMD's ARM CPU with a tightly integrated FPGA on board for exotic instructions).

I mean, just take a look at this video.
This is Metro Exodus (2019):

• Running x86 code under rosetta 2 emulation
• On a fanless Arm M1 laptop with an integrated GPU
• at 1080p with Medium settings at around 30FPS (the entire video with both indoor and outdoor scenes)

Show me one x86 laptop that can do the same with integrated graphics, especially in a fanless chassis?

96EU Intel Iris Xe in the latest tiger lake can only get 30FPS at 720p with Low settings (see this video) and it will run at ~15-20FPS on the same settings on 28W laptops, confirmed by notebookcheck's results:

It only manages to do 15FPS (on an asus Zenbook 14 ) and 20 FPS (on an Acer Swift 5) both running with a 28W TDP (that shortly draws up to 58W with 100% CPU and GPU load) . A far cry from a fanless.

Therefore a decent ARM CPU 3-5 years from now shoud run PS5 x86 games under emulation. Yes It requires hardware investment (like Apple has memory synchronized memory instructions pretty much only for x86 emulation). And It might be only the most popular games requiring patches, but it can.

If other ARM vendors did care about ARM architecture in general then we'd have Apple licensing their CPU designs for others

ARM custom core licences don't allow licencing their cores onwards. Only designing cores for their own products.