Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

The Ampere Altra Review: 2x 80 Cores Arm Server Performance Monster

anandtech.com

Post reply on HN
Print view View original

61–70 of 266 posts

Re: The Ampere Altra Review: 2x 80 Cores Arm Server Performance Monster

#61

Post by tolien »

x86_64Ubuntu wrote: post #49
tlamponi wrote: post #48

So how are the cores interconnected? As ARM server designs love to throw out tens to hundreds of cores which sounds good on paper, but are not really useful to lots of workloads as they have a shit interconnect or lots of NUMA nodes. Interconnect is key, e.g., see the improvements between zen2 and zen3, especially for the `4 An anecdote: I got some hands on a box with a Cavium ThunderX2 96 logical core processor abou…

This is a dumb question. But as you've said, cores per die has a strong influence on how performant a system is and that's because of the processor interconnects. My question is, what are the processors saying to each other?

In the case of Zen/Zen2, some of the dies didn't have direct access to main memory but went through an adjacent die's memory controller and channel. This had a performance and latency effect, and required some extra work on the part of OS schedulers to make sure that tasks weren't being scheduled on related cores.

It's (relatively) easy to have lots of cores on one die, it's not so easy to keep them fed.

Re: The Ampere Altra Review: 2x 80 Cores Arm Server Performance Monster

#62

Post by my123 »

flyinghamster wrote: post #52
simon1573 wrote: post #18

Earlier quoted context omitted.

I'm excited as well. But as an earlier discussion pointed out I'm worried that the transition might impact the openness in PCs, e.g. locked bootloaders etc

Yup. Finding out that Big Sur phones home every time one opens an application is a showstopper for me, never mind Apple's ever-changing ports and its campaign against the right to repair. We've also already seen that Microsoft considers ARM systems to be "special" enough to require locked bootloaders as well. Don't get me started on the mess in the smartphone world. I'd hate to see a world where the only "open" compu…

Windows on arm64 systems aren't locked down.

Meanwhile for arm64 Macs, they aren't either. Yesterday, https://twitter.com/xenokovah/status/1339914714055368704?s=2... was released to run unsigned kernels.

Re: The Ampere Altra Review: 2x 80 Cores Arm Server Performance Monster

#63

Post by dspillett »

tlamponi wrote: post #48

So how are the cores interconnected? As ARM server designs love to throw out tens to hundreds of cores which sounds good on paper, but are not really useful to lots of workloads as they have a shit interconnect or lots of NUMA nodes. Interconnect is key, e.g., see the improvements between zen2 and zen3, especially for the `4 An anecdote: I got some hands on a box with a Cavium ThunderX2 96 logical core processor abou…

> slower than on my...

May have consumed far less power doing the job though. For server farms performance-per-watt can be far more important than absolute speed per unit, especially for massively parallel tasks.

Re: The Ampere Altra Review: 2x 80 Cores Arm Server Performance Monster

#64

Post by walki »

panpanna wrote: post #27

Has anyone tried machine learning on this or the Graviton2? I understand TPUs and GPUs easily beat these guys but it would still be interesting to see what raw cpu power can achieve in 2020.

> Has anyone tried machine learning on this or the Graviton2?

I have not done any machine learning on AWS Graviton2 CPUs but I ran many other CPU benchmarks on Graviton2 CPUs and overall I have been disappointed by their performance. They are still much slower than current x64 CPUs (x64 CPUs are up to 2x faster in single thread mode).

According to the benchmarks from Anandtech the Ampere Altra should have much better performance than Graviton2 CPUs as its performance is neck to neck with the fastest x64 CPUs.

Re: The Ampere Altra Review: 2x 80 Cores Arm Server Performance Monster

#65

Post by Sebb767 »

SoapSeller wrote: post #29
Sebb767 wrote: post #2

These things look awesome, but are pretty expensive, still. I'd love to put an ARM server in my homelab, but the price range between 100$ (RPi & Co) and 10000$ (enterprise server) seems pretty empty right now :/

You can get SolidRun HoneyComb[0] for ~$1000, it has 16 cores with supposedly reasonable performance. [0] https://www.solid-run.com/arm-servers-networking-platforms/h...

This looks what I was searching for. Thank you!

Re: The Ampere Altra Review: 2x 80 Cores Arm Server Performance Monster

#66

Post by flyinghamster »

my123 wrote: post #62
flyinghamster wrote: post #52

Earlier quoted context omitted.

Yup. Finding out that Big Sur phones home every time one opens an application is a showstopper for me, never mind Apple's ever-changing ports and its campaign against the right to repair. We've also already seen that Microsoft considers ARM systems to be "special" enough to require locked bootloaders as well. Don't get me started on the mess in the smartphone world. I'd hate to see a world where the only "open" compu…

Windows on arm64 systems aren't locked down. Meanwhile for arm64 Macs, they aren't either. Yesterday, https://twitter.com/xenokovah/status/1339914714055368704?s=2... was released to run unsigned kernels.

It's encouraging to see an effort like this, but unless it can become a first-class citizen on M1 hardware, it will be at best like trying to keep an iDevice jailbroken, or having a custom Android ROM that lacks important basic functionality like VoLTE.

One can hope, I guess. If I could be sure I could run Linux without it being hobbled, and that Apple wouldn't pull the rug out from under me, I could actually see myself adding a Mac Mini to the stable.

Edit:

> Windows on arm64 systems aren't locked down.

Has this changed? I recall that on ARM, Microsoft requires that UEFI Secure Boot be enabled, cannot be disabled, and cannot load custom keys. So, you could boot Linux as long as it's been "blessed" by Microsoft, assuming they don't pull the rug out, much like the leaving Secure Boot enabled on an x86 PC (except on a PC you can usually load your own keys).

https://arstechnica.com/information-technology/2012/01/windo...

Re: The Ampere Altra Review: 2x 80 Cores Arm Server Performance Monster

#67

Post by spamizbad »

dspillett wrote: post #63
tlamponi wrote: post #48

So how are the cores interconnected? As ARM server designs love to throw out tens to hundreds of cores which sounds good on paper, but are not really useful to lots of workloads as they have a shit interconnect or lots of NUMA nodes. Interconnect is key, e.g., see the improvements between zen2 and zen3, especially for the `4 An anecdote: I got some hands on a box with a Cavium ThunderX2 96 logical core processor abou…

> slower than on my... May have consumed far less power doing the job though. For server farms performance-per-watt can be far more important than absolute speed per unit, especially for massively parallel tasks.

Cavium ThunderX2 is a 205 TDP chip vs a circa 2017 i7 (~105W) so probably not.

Re: The Ampere Altra Review: 2x 80 Cores Arm Server Performance Monster

#68

Post by nine_k »

cameron_b wrote: post #42

This line is the kicker for me: "Intel’s current Xeon offering simply isn’t competitive in any way or form at this moment in time." The rest of the paragraph twists the knife more, and declares there to be an open market on server hardware for AMD and Ampere to fight it out. With application stacks being what they are and optimizations for Xeon so deeply integrated, it's unlikely that folks running private datacenter…

I can imagine that every cloud hosting provider expands their fleet of machines, and also retires and replaces some of it.

If the expansion and replacement will mostly involve AMD and ARM-based solutions, I won't be surprised.

Of course Xeons are going to stay for a relatively long time, because some customers are deeply invested in them, e.g. relying on AVX512 heavily, etc.

I also don't think that Intel is going to go under. AMD was in the position of a second-rate player for decades, and had expensive failures, too (remember Bulldozer?). So I expect them to regroup and show us something amazing in, say, 5 years. But one thing they are.going to lose: their sweet big margins.

Re: The Ampere Altra Review: 2x 80 Cores Arm Server Performance Monster

#69

Post by mhh__ »

cameron_b wrote: post #42

This line is the kicker for me: "Intel’s current Xeon offering simply isn’t competitive in any way or form at this moment in time." The rest of the paragraph twists the knife more, and declares there to be an open market on server hardware for AMD and Ampere to fight it out. With application stacks being what they are and optimizations for Xeon so deeply integrated, it's unlikely that folks running private datacenter…

Intel are also miles ahead when it comes to software and documentation.

Intel's optimization guide alone is about 1000 pages long, AMD's isn't as good and they've been at it for decades now.

If you have the money Intel aren't as competitive any more, but for the middle ground where you don't have the budget to bring up your own software it's not the same calculus.

Re: The Ampere Altra Review: 2x 80 Cores Arm Server Performance Monster

#70

Post by MarioMan »

x86_64Ubuntu wrote: post #49
tlamponi wrote: post #48

So how are the cores interconnected? As ARM server designs love to throw out tens to hundreds of cores which sounds good on paper, but are not really useful to lots of workloads as they have a shit interconnect or lots of NUMA nodes. Interconnect is key, e.g., see the improvements between zen2 and zen3, especially for the `4 An anecdote: I got some hands on a box with a Cavium ThunderX2 96 logical core processor abou…

This is a dumb question. But as you've said, cores per die has a strong influence on how performant a system is and that's because of the processor interconnects. My question is, what are the processors saying to each other?

Modern CPUs like these are shared memory multiprocessors. Programs on these CPUs are often designed to run on multiple cores at the same time to improve performance, necessitating communication between cores, to coordinate and share work between each other. Since each core typically has a private cache, most communication in a system like this will typically involve cache coherence protocols to ensure cores have a consistent view of memory contents. Adding more cores increases the overall complexity of the system. The latency of these protocols grows with more cores as interconnections become more "distant", across processor interconnects. Minimizing latency and effectively scaling to more cores is a difficult problem to solve while staying within silicon and thermal budgets.
Post reply on HN