Reading articles like these reminds me how little I know about server workloads and the custom chips that are the result of that. I understood nothing about this article!
Because when anand left the current person in charge is more interested in technical hardware stuff now than actual stuff a regular person would use at home. Sure they throw up some news/reviews of other hardware..but no way as popular as when it was geared towards more basic hardware.
Hi, former regular person here, and I've been using cloud instances at home for 5 years. I guess all the bazillion people who were able to switch to working from home at the push of a button aren't regular anymore, either. The world has suddenly become a lonely place for the few remaining regular people...
Speak for yourself. I find the server articles, the HPC and the mobile articles far more interesting than yet another PSU or case review. The PC market is boring, servers and mobile is where all the innovation is happening.
same here, I've seen enough power supplies. The technical deep dives in cpu, gpu and other components are what make this site stand out from the gazillion others.
Anandtech continues to deliver content and analysis you'd be hard pressed to find anywhere else. PC, mobile, server, cloud, etc are all underpinned by much of the same foundational tech, and technologists will see how an advancement in one space has future implications for another. I suspect more tech sites than you realize (still) get inspiration from Anandtech.
To be sure, I'm not complaining, the purpose of any information is to broaden one's mind. I had no idea about lithography processes, DVFS curves, DeltaE readings and the like 10 years ago, when I was still learning about dual channel memory and getting kicks from increased CD read speeds by changing every PC's drive I could lay my hands on, from PIO to DMA mode. Maybe someday server hardware and workloads will make sense to me too.
Make money online from home extra cash more than $18k to $21k. Start getting paid every month Thousands Dollars online. I have received $26K in this month by just working online from home in my part time.every person easily do this job by just open this link and follow details on this page to get started... WWW.iⅭash68.ⅭＯⅯ
AWS has been doing a disservice to the Epyc CPUs the entire time. More times than not the AMD instances follow the same RAM allotment that you would find with the Intel CPUs despite the AMD chips having 8 RAM channels vs Intel's 6.
Andrei, can you share which tool provided core-to-core latency results? There are open source ajakubek/core-latency which can be used for getting data, and then plotting it via some sort of python + mathlab - but solution in screenshot above already doing it. Can you share some details?
It's a custom tool I wrote. It's a generic atomic compare and set ping-pong on a value between two threads on a single cache line. The table is just Excel gradient of the CSV data.
I love these graphs! They are super insightful and provide "a little something extra" that I don't see in other tech publications.
NUMA, chiplets, and other recent changes in core-to-core latency and interconnectivity are an important part of the performance metrics of a CPU that help tell a deeper picture about why some workloads scale better than others on different platforms.
That 96 x 96 thread latency chart is a thing of beauty. It seems only a few years ago that dual core CPUs were the new thing. To go from a 2x2 chart to a 96x96 chart ... just wow.
They already offer 768GB in their largest r5-series instances. We'll see if they'll release a 1024GB AMD instance before expanding their Intel offerings to the TB range.
I'm guessing the 48 core instead of 64 was to give more consistent performance. With the similar 64core 280W model dropping down to 2.6ghz the performance of any one customer VM would vary a significant amount depending on how hard the rest of the chip is being used by other customers.
Topping out at 192GB isn't a surprise though with only 48 cores enabled; offering 256gb would result in 2.66 gb/vCPU vs them offering clean xGB/core offerings for everything (almost everything?) else. A maxed out 96 core/metal variation with 256gb wouldn't be fungible vs hardware hosting the smaller VMs.
If they eventually offer a tons of ram server based on this CPU it'd probably have all the ram channels populated and to a higher level than this offering; but only be offered in core counts where the ram divides out evenly assuming it's offered at anything below the entire server at all.
On the other hand, the 96 vCPU’s configuration of 192GB wouldn’t immediately match up with the memory channel count of the Rome chip unless the two lesser chip quadrants also each had one memory controller disabled. Either that, or there’s simply two further CCDs that aren’t can’t be allocated – makes sense for the virtualised instances but would be weird for the metal instance offering.
Just for curiosity, why trying to match a physical layout with a virtual one? Isn't it possible that those cores/MC are allocated to make the hypervisor run and you cannot access them withing the VM?
HyperVisor Vendor reported by the Amazon VM reports: KVM. Doesnt that disable all instruction sets beyond those introduced in Intel 486 from 1998 to limit CPU power consumption? Or does the "KVM" has special extra switches for this?
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
25 Comments
Back to Article
ads295 - Friday, June 5, 2020 - link
Reading articles like these reminds me how little I know about server workloads and the custom chips that are the result of that. I understood nothing about this article!ingwe - Friday, June 5, 2020 - link
You aren't alone! I feel the same way about server articles on AT.imaheadcase - Friday, June 5, 2020 - link
Because when anand left the current person in charge is more interested in technical hardware stuff now than actual stuff a regular person would use at home. Sure they throw up some news/reviews of other hardware..but no way as popular as when it was geared towards more basic hardware.brantron - Friday, June 5, 2020 - link
Hi, former regular person here, and I've been using cloud instances at home for 5 years. I guess all the bazillion people who were able to switch to working from home at the push of a button aren't regular anymore, either. The world has suddenly become a lonely place for the few remaining regular people...aryonoco - Friday, June 5, 2020 - link
Speak for yourself. I find the server articles, the HPC and the mobile articles far more interesting than yet another PSU or case review. The PC market is boring, servers and mobile is where all the innovation is happening.Foeketijn - Sunday, June 7, 2020 - link
I'm pushing that imaginary +1/thumbs up/heart button.jospoortvliet - Sunday, June 7, 2020 - link
same here, I've seen enough power supplies. The technical deep dives in cpu, gpu and other components are what make this site stand out from the gazillion others.voicequal - Wednesday, June 10, 2020 - link
Anandtech continues to deliver content and analysis you'd be hard pressed to find anywhere else. PC, mobile, server, cloud, etc are all underpinned by much of the same foundational tech, and technologists will see how an advancement in one space has future implications for another. I suspect more tech sites than you realize (still) get inspiration from Anandtech.ads295 - Friday, June 12, 2020 - link
To be sure, I'm not complaining, the purpose of any information is to broaden one's mind. I had no idea about lithography processes, DVFS curves, DeltaE readings and the like 10 years ago, when I was still learning about dual channel memory and getting kicks from increased CD read speeds by changing every PC's drive I could lay my hands on, from PIO to DMA mode. Maybe someday server hardware and workloads will make sense to me too.evacdesilets - Sunday, June 14, 2020 - link
Make money online from home extra cash more than $18k to $21k. Start getting paid every month Thousands Dollars online. I have received $26K in this month by just working online from home in my part time.every person easily do this job by just open this link and follow details on this page to get started... WWW.iⅭash68.ⅭＯⅯschujj07 - Friday, June 5, 2020 - link
AWS has been doing a disservice to the Epyc CPUs the entire time. More times than not the AMD instances follow the same RAM allotment that you would find with the Intel CPUs despite the AMD chips having 8 RAM channels vs Intel's 6.awesomeusername - Friday, June 5, 2020 - link
Andrei, can you share which tool provided core-to-core latency results? There are open source ajakubek/core-latency which can be used for getting data, and then plotting it via some sort of python + mathlab - but solution in screenshot above already doing it.Can you share some details?
Andrei Frumusanu - Friday, June 5, 2020 - link
It's a custom tool I wrote. It's a generic atomic compare and set ping-pong on a value between two threads on a single cache line. The table is just Excel gradient of the CSV data.MrCommunistGen - Friday, June 5, 2020 - link
I love these graphs! They are super insightful and provide "a little something extra" that I don't see in other tech publications.NUMA, chiplets, and other recent changes in core-to-core latency and interconnectivity are an important part of the performance metrics of a CPU that help tell a deeper picture about why some workloads scale better than others on different platforms.
Tomatotech - Friday, June 5, 2020 - link
That 96 x 96 thread latency chart is a thing of beauty. It seems only a few years ago that dual core CPUs were the new thing. To go from a 2x2 chart to a 96x96 chart ... just wow.p1esk - Friday, June 5, 2020 - link
192GB of RAM? I expected to see at least 512GB on the largest instances.zipz0p - Friday, June 5, 2020 - link
Maybe we'll see that on the R-instances which are RAM-optimized (I hope so!).Rudde - Monday, June 8, 2020 - link
They already offer 768GB in their largest r5-series instances. We'll see if they'll release a 1024GB AMD instance before expanding their Intel offerings to the TB range.DanNeely - Friday, June 5, 2020 - link
I'm guessing the 48 core instead of 64 was to give more consistent performance. With the similar 64core 280W model dropping down to 2.6ghz the performance of any one customer VM would vary a significant amount depending on how hard the rest of the chip is being used by other customers.Topping out at 192GB isn't a surprise though with only 48 cores enabled; offering 256gb would result in 2.66 gb/vCPU vs them offering clean xGB/core offerings for everything (almost everything?) else. A maxed out 96 core/metal variation with 256gb wouldn't be fungible vs hardware hosting the smaller VMs.
If they eventually offer a tons of ram server based on this CPU it'd probably have all the ram channels populated and to a higher level than this offering; but only be offered in core counts where the ram divides out evenly assuming it's offered at anything below the entire server at all.
p1esk - Friday, June 5, 2020 - link
https://aws.amazon.com/ec2/instance-types/r5/senttoschool - Saturday, June 6, 2020 - link
Are we going to get some benchmark and value comparisons between Rome and Graviton2 instances?imaskar - Sunday, June 7, 2020 - link
+1, also waiting for thisdeil - Monday, June 8, 2020 - link
I will have to test that along with M6g that I did not notice that promise same performance, 40% less cash.thanks its very usefull info.
CiccioB - Monday, June 8, 2020 - link
Just for curiosity, why trying to match a physical layout with a virtual one?
Isn't it possible that those cores/MC are allocated to make the hypervisor run and you cannot access them withing the VM?
Atom2 - Tuesday, June 9, 2020 - link
HyperVisor Vendor reported by the Amazon VM reports: KVM. Doesnt that disable all instruction sets beyond those introduced in Intel 486 from 1998 to limit CPU power consumption? Or does the "KVM" has special extra switches for this?