News [Anandtech] Arm Announces Mobile Armv9 CPU Microarchitectures: Cortex-X2, Cortex-A710 & Cortex-A510

[Gideon]

ARM consumer stack updates:

• Finally a decent "little" core update (A55 -> A510) with 35% performance gain
• Big core is less ambitious (A78 -> A710) with 10% uplift mentioned
• X2 is supposedly 16% faster than X1
• Lots of other changes, Armv9 ISA (with decent vector ops finally), new interconnects, and more L3 cluster designs

Arm Announces Mobile Armv9 CPU Microarchitectures: Cortex-X2, Cortex-A710 & Cortex-A510

www.anandtech.com

 

[insertcarehere]

I hadn't even thought about thread count As I recall, the NeoVerse V1 is a single thread processor. Zen3 and Zen4 are SMT2. Assuming that Genoa is still going to be 96 cores (12 CCDs, 8 core CCX), that 192 threads PER socket, and 384 threads per 2S system. I can't imagine that they will have a problem with system throughput, even if they have to dial back clocks a lot to keep power draw reasonable. This should be an interesting race...

A 2S Genoa platform to match 3 Graviton 3s in power consumption would imply ~150w TDP per 96C Genoa. Is that even possible given the IOD power constraints + CCDs which also need to be designed to be able to hit 5ghz in desktop parts (I. E higher leakage)?

Milan doesn't hit the 150w TDP range until scaled down to 32C part SKUs.

 

[LightningZ71]

Milan, being based on N7 CCDs with a GF 14LP IOD doesn't hit 150W until scaled down to 32C.
How does that change with N5 CCDs and N7/N6 IOD?

 

[DrMrLordX]

The only thing I don't like this announcement is that we won't get board samples sent to reviewers. There will have to be a lot of guesswork about how this product affects Amazon's bottom line. Still a very interesting product development though. Cloud folks be happy yo.

 

[BorisTheBlade82]

Milan, being based on N7 CCDs with a GF 14LP IOD doesn't hit 150W until scaled down to 32C.
How does that change with N5 CCDs and N7/N6 IOD?

IMHO the process of the IOD plays a minor role with regards to power consumption. If the interconnect is still IFoP then the same "thick" wires between dies need to be powered. Only exchanging these wires will net significant gains.

Or in other words: We cannot isolate the consumption of the IOD from the type of interconnect it needs to power, as this is were the power really goes.

 

[eek2121]

In cloud setting, we are getting to the point of "good enough" for ARM. If this thing is 30% faster than Graviton2 at 100W envelope, it won't matter for typical cloud workloads if 225W CPU is say 25% faster.
TCO for Amazon will be highly in favor of Graviton3 and by the time cloud guys will sample Bergamo in 2023, Amazon will be out with Graviton4.

In generic cloud computing, the writing is on the wall for both Intel/AMD.

Ofc, i am dealing with same type of complex requests every day. Running our own servers etc.
But even in this environment we find plenty of uses for "generic" cloud computing - like computing that is perfectly offloadable to cloud.
Even if we were not using it directly, i know 3rd party services we already depend on are using Azure and Amazon stuff. They are probably on x64 for now, but heck, as long as they meet our SLA, they could run on ZX Spectrums, data is data.

So there is an expanding sea of generic computing, that these ARM CPUs excel at. And with Graviton2, 3 things are looking better each day.

And obviuosly even if we look beyond HW - since that Phoronix benchmark software support for ARM64 is moving leaps forward. I think 99% of "server" software like compilers, frameworks,. JDKs are already ARM64 capable and being optimized for ARM as we speak.

A recent example with JDKs - Amazon is maintaining their own openJDK distro

OpenJDK Download - Corretto - AWS

Amazon Corretto is a no-cost, multiplatform, production-ready distribution of the Open Java Development Kit (OpenJDK). Corretto comes with no-cost long-term support. Amazon runs Corretto internally on thousands of production services and Corretto is certified as compatible with the Java SE standard.

aws.amazon.com

That means very important thing - they are testing it on their ARM cloud stuff and hardware and probably do not mind optimizing for their hardware either as they seem to have very capable JDK developers as well! ( for example Announcing preview release for the generational mode to the Shenandoah GC | AWS Developer Tools Blog (amazon.com) )

Mindshare is increasing, i used to say the biggest obsticle is lack of capable workstation for development, i think now M1 derivatives are taking care of it. Need one more vendor with Linux/Windows workstation and we are set.

Man, I have to question whether you folks know the cloud business at all. My company spends a TON of money with AWS. We are on of the top spenders worldwide. We have a direct line of contact to the top folks at AWS. I’ve mentioned this in the past, but we tried ARM for our instances. The cheaper price did not cover the cost of lower performance. Our AWS bills went up significantly. Note that when using thousands of instances, you don’t pick and choose the underlying hardware. You pick the absolute cheapest you can that meets your performance metrics. That typically means perf/$ is king. We do use some Graviton2 instances, but the number is very small (maybe 50 or so out of thousands) and we only use those instances to fill in surplus demand. That number shrinks daily. It was a few hundred earlier this year.

Actually, surprisingly, most of our instances until recently were Intel due to a shortage of AMD powered instances.

 

[JoeRambo]

Man, I have to question whether you folks know the cloud business at all. My company spends a TON of money with AWS. We are on of the top spenders worldwide. We have a direct line of contact to the top folks at AWS. I’ve mentioned this in the past, but we tried ARM for our instances. The cheaper price did not cover the cost of lower performance. Our AWS bills went up significantly. Note that when using thousands of instances, you don’t pick and choose the underlying hardware. You pick the absolute cheapest you can that meets your performance metrics. That typically means perf/$ is king. We do use some Graviton2 instances, but the number is very small (maybe 50 or so out of thousands) and we only use those instances to fill in surplus demand. That number shrinks daily. It was a few hundred earlier this year.

Actually, surprisingly, most of our instances until recently were Intel due to a shortage of AMD powered instances.

Not sure how any of this contradicts any of what i said. We are minor user of cloud ourselves, but we use 3rd parties that are obviously using Azure and AWS. We have SLA with those 3rd parties, and as long as they meet it today and meet it tomorrow with potentially Graviton3 we are perfectly fine. Everyone can do tests and price/perf calculations for themselves.
ARM "on cloud" is fast evolving situation, in the past software support was lacking, nowadays it will run just as well as X64 instances.

 

[moinmoin]

Cloud folks be happy yo.

AWS customers be happy yo. That's quite a restricted audience.

 

[DrMrLordX]

AWS customers be happy yo. That's quite a restricted audience.

They still have a lot of customers though. If Graviton3 sets the bar for future ARM server offerings (such as those from Ampere), then overall the picture for ARM cloud server should continue to improve.

Would be great to see Graviton3 vs Altra MAX benchmarks.

 

[moinmoin]

If Graviton3 sets the bar for future ARM server offerings (such as those from Ampere), then overall the picture for ARM cloud server should continue to improve.

If Graviton3 sets any significant bar, it sets the bar for all server offerings, not only ARM ones. That's the issue Arm faces with closed market offerings like Amazon's Graviton and Apple's M1, they don't really push ARM cores forward. Instead they show everybody what could be, but in the open market the competition remains tight.

If Ampere manages to catch up to Amazon and offer an equivalent product on the open market in the near future, that will be a big deal indeed which could improve the overall the picture for ARM servers.

 

[DrMrLordX]

If Graviton3 sets any significant bar, it sets the bar for all server offerings, not only ARM ones.

See post by @eek2121 above. Some orgs aren't getting good use out of Graviton2 now (even versus Intel AWS instances, many of which are still Cascade Lake). Graviton3 isn't likely to change that picture for parties that have workloads that aren't cost-effective on ARM AWS instances. What will happen is that outliers like Ampere will be competing for a niche within the greater enterprise cloud market with Graviton3 indirectly - of course you can't buy Graviton3 directly, but you can outsource datacentre/cloud functions to AWS instances. In order for Graviton3 and Altra to better compete for a larger share of the enterprise cloud market in the future, they'd best address those workloads that are (for whatever reason) less-performant on ARM hardware.

If Ampere manages to catch up to Amazon and offer an equivalent product on the open market in the near future, that will be a big deal indeed which could improve the overall the picture for ARM servers.

If you look at the specs for Graviton2 and Altra/Altra MAX, it should be clear that Ampere has the better product overall, at least in terms of hardware capabilities. TCO is muddied a bit due to the way Amazon has their hardware locked up in AWS. Regardless, Graviton3 may swing the pendulum in the other direction.

 

[moinmoin]

See post by @eek2121 above. Some orgs aren't getting good use out of Graviton2 now (even versus Intel AWS instances, many of which are still Cascade Lake).

That's within AWS' pricing scheme. We found AWS overall pricey within the cloud competition, and more dedicated (but still virtualized) systems easily have better price performance ratios than these cloud ecosystem offerings (where a big selling point but also danger of getting locked in is the increased proprietary software abstraction).

 

[DrMrLordX]

That's within AWS' pricing scheme. We found AWS overall pricey within the cloud competition, and more dedicated (but still virtualized) systems easily have better price performance ratios than these cloud ecosystem offerings (where a big selling point but also danger of getting locked in is the increased proprietary software abstraction).

Well the point was, at least in the above post in the given workload, older Cascade Lake instances were more cost-effective for the workload than Graviton2. So yes AWS may be pricey compared to dedicated systems, but within the confines of Amazon's own pricing scheme, they hadn't priced Graviton2 competitively enough for its performance in that workload to be worth the money. It's not likely that Amazon is subsidizing Intel instances to make Graviton2 less-competitive in comparison. The cost differential is based on TCO of the underlying hardware factored against performance. Graviton3 probably won't improve the situation much since it will be competing against Sapphire Rapids, Milan, and Genoa. And maybe Milan-X and Bergamo.

For workloads that are more-amenable to NeoVerse implementations, I expect the opposite. Those already investing in the ARM server ecosystem are going to have to take a serious look at what AWS has to offer in the future.

 

[Nothingness]

Man, I have to question whether you folks know the cloud business at all. My company spends a TON of money with AWS. We are on of the top spenders worldwide. We have a direct line of contact to the top folks at AWS. I’ve mentioned this in the past, but we tried ARM for our instances. The cheaper price did not cover the cost of lower performance. Our AWS bills went up significantly. Note that when using thousands of instances, you don’t pick and choose the underlying hardware. You pick the absolute cheapest you can that meets your performance metrics. That typically means perf/$ is king. We do use some Graviton2 instances, but the number is very small (maybe 50 or so out of thousands) and we only use those instances to fill in surplus demand. That number shrinks daily. It was a few hundred earlier this year.

It's not because your company is a big user that it proves anything. My company has thousands of instances and most of them are Graviton. But I'm not silly enough to claim that this means it applies to all uses.

 

[moinmoin]

Well the point was, at least in the above post in the given workload, older Cascade Lake instances were more cost-effective for the workload than Graviton2.

And my point was that that only matters to the AWS customers either way. Possibly to general cloud customers but there AWS as a whole isn't particularly cost-effective to begin with. In any case it doesn't matter for the general market as Graviton is proprietary to Amazon.