AWS continues to invest in AMD's Epyc platform

[onetheycallEric]

Why it matters: AMD continues to slowly invade the lucrative server segment where not so long ago they were all but shut out by Intel. Amazon is placing more faith in AMD's Epyc platform by continuing to invest in new AWS instances like the new M5ad and R5ad.

Amazon Web Services announced the availability of more Epyc powered instances. The new M5ad and R5ad instances are variants of the M5 and R5 EC2 (Elastic Compute Cloud), and use custom AMD Epyc 7000-series chips.

According to Amazon, the M5ad instances are geared towards general compute workloads. This could be gaming, web servers, data logging, or media processing. The R5ad instances are aimed at more intense workloads, specifically those that are memory-intensive. Such applications include in-memory analytics, data mining, in-memory data bases, and caching.

Both the M5ad and R5ad instances will run at 2.5 GHz and offer local NVMe SSD-based block level storage with AES 256-bit XTS encryption support. The instances can also be configured with up to 96 vCPUs. Amazon is billing these new instances as an option "for customers who are looking to achieve a 10% cost savings on their Amazon EC2 compute environment." That 10 percent savings is presumably over the nearby competition, the AWS instances powered by Intel's Skylake-based Xeon processors.

The new instances are available in the US East (N. Virginia, Ohio), US West (Oregon), and Asia Pacific (Singapore) AWS Regions.

Permalink to story.

https://www.techspot.com/news/79431-aws-continues-invest-amd-epyc-platform.html

 

[Vulcanproject]

You will be able to have 64 cores on a single socket, or 128 cores on a dual socket board this year with 7nm power efficiency and relatively low purchase cost. Much higher performance per watt than any Xeon that will be available this year and possibly most of next year.

AMD are going to capture great swathes of the server market, that much is certain.

 

[Kashim]

You will be able to have 64 cores on a single socket, or 128 cores on a dual socket board this year with 7nm power efficiency and relatively low purchase cost. Much higher performance per watt than any Xeon that will be available this year and possibly most of next year.

AMD are going to capture great swathes of the server market, that much is certain.

Yes, but we've seen this movie before. AMD will gain a small slice while Intel dust themselves off. Then a couple of years later Intel comes back with a vengeance (performance improvements + price slashing) and pushes AMD back out like they always do. I hope this time it will be different, but no one should expect Intel to take this lying down.

 

[Vulcanproject]

Yes, but we've seen this movie before. AMD will gain a small slice while Intel dust themselves off. Then a couple of years later Intel comes back with a vengeance (performance improvements + price slashing) and pushes AMD back out like they always do. I hope this time it will be different, but no one should expect Intel to take this lying down.

Intel aren't yet behind until AMD launch 7nm CPUs in a couple months, but on that day for the first time in well over a decade they will be. It'll be a major landmark because of that.

With Netburst they went down the wrong fork in the path. To fix it they simply backed out and went down the other they had available with their mobile parts, developed to desktop and server. What helped a lot however is that Intel have always had this position where they had a manufacturing process advantage.

This process advantage was maintained for over 10 years, coupled with an architecture advantage the gap was HUGE. It's now 2019. Their architecture advantage has been reduced to single digit percentages, and the process advantage has completely evaporated.

In fact, AMD will have the better process on 7nm TSMC and almost certainly eliminate the remaining architecture advantage with it. They also have a developed strategy in place, anticipating the chiplet future ahead of Intel. Their plans in this area are clearly more advanced, as these EPYC server parts are showing. They can bolt together faster chips while costing them less. It's a double blow for Intel.

Intel have to fix their manufacturing and catch up on chiplet interconnect design. Both have proven to be difficult multi year projects. They won't be able to charge the prices they used to while maintaining total dominance.

This is good for consumers. It's fantastic. Whether you want a server, workstation, gaming desktop or even a new laptop- later this year is going to probably be the best year for renewing any of those for a long time.

 

[Evernessince]

You will be able to have 64 cores on a single socket, or 128 cores on a dual socket board this year with 7nm power efficiency and relatively low purchase cost. Much higher performance per watt than any Xeon that will be available this year and possibly most of next year.

AMD are going to capture great swathes of the server market, that much is certain.

Yes, but we've seen this movie before. AMD will gain a small slice while Intel dust themselves off. Then a couple of years later Intel comes back with a vengeance (performance improvements + price slashing) and pushes AMD back out like they always do. I hope this time it will be different, but no one should expect Intel to take this lying down.

That's not how it went down the first time. Intel paid companies through an illegal "rebate" program that required vendors to refuse to sell AMD products. They were sued by AMD for this and lost but by then it was too late

Your version of the events are entirely off the mark. And price slashing? Intel most certainly did not win the market by cutting pricing, that much is for certain.

 

[yRaz]

This is good for consumers. It's fantastic. Whether you want a server, workstation, gaming desktop or even a new laptop- later this year is going to probably be the best year for renewing any of those for a long time.

and for those of us not interested in being on the cutting edge of performance, PRICE CUTS!!!!!!! Might try to pick up a few 2700/2700x's after a price drop and maybe change my server over to a cheap threadripper?

 

[akamateau]

Yes, but we've seen this movie before. AMD will gain a small slice while Intel dust themselves off. Then a couple of years later Intel comes back with a vengeance (performance improvements + price slashing) and pushes AMD back out like they always do. I hope this time it will be different, but no one should expect Intel to take this lying down.

Actually you are quite wrong. The last time this happened in HISTORY Intel recovered through ILLEGAL TRADE PRACTICES and was cited and fined by both the FTC and the European Union.

Furthermore AMD never had such a massive advantage not only with core counts, chiplet technology but also Intel bungled it's process node transition. Intel is still stuck on 14nm and even then can not produce enough silicon. AMD will earn market share just because Intel has failed to produce.

AMD 7nm is mature and they are now sampling 3nm for 2021 releases!!

This is a perfect storm that Intel will recover from eventually but not before AMD has grabbed serious market share.

And you still forget the really big point to all of this. AMD is the tail that wags the DOG.

 

[hahahanoobs]

You will be able to have 64 cores on a single socket, or 128 cores on a dual socket board this year with 7nm power efficiency and relatively low purchase cost. Much higher performance per watt than any Xeon that will be available this year and possibly most of next year.

AMD are going to capture great swathes of the server market, that much is certain.

Companies aren't consumers. They don't easily fall for low prices and core counts like your average joe. For instance, are EPYC chips a drop in replacement? How much will it cost to switch over? Do they perform the same or better with their existing software? Is the tech support there? Do EPYC chips match or exceed current speeds and efficiencies? Do companies believe AMD be able to remain competitive 5, 10 or 20 years from now? Etc, etc, etc.

These are all important factors that are almost never mentioned in the comment sections of some tech sties. Your comment is a perfect example.

 

[hahahanoobs]

Unrelated, but I just saw this and my jaw dropped. Intel is not messing around when it comes to graphics and gaming. This is HUGE!

"Intel Snatches Up Tom Peterson (formerly) Of NVIDIA, Will Be Joining As An Intel Engineering Fellow"

 

[Vulcanproject]

Companies aren't consumers. They don't easily fall for low prices and core counts like your average joe. For instance, are EPYC chips a drop in replacement? How much will it cost to switch over? Do they perform the same or better with their existing software? Is the tech support there? Do EPYC chips match or exceed current speeds and efficiencies? Do companies believe AMD be able to remain competitive 5, 10 or 20 years from now? Etc, etc, etc.

These are all important factors that are almost never mentioned in the comment sections of some tech sties. Your comment is a perfect example.

I am well aware of various other major factors that extend well beyond the simple hardware a company will purchase from AMD.

All your points are very valid, but the answer to many of them lies between the lines of this article itself. The world's second largest provider (only just second by revenue) of cloud services is investing heavily into AMD's infrastructure.

That speaks volumes.

 

[hahahanoobs]

I am well aware of various other major factors that extend well beyond the simple hardware a company will purchase from AMD.

All your points are very valid, but the answer to many of them lies between the lines of this article itself. The world's second largest provider (only just second by revenue) of cloud services is investing heavily into AMD's infrastructure.

That speaks volumes.

How long can AMD keep it up is the question. I know I come off as an AMD hater, but they need consistency before I can get excited.

 

[yRaz]

How long can AMD keep it up is the question. I know I come off as an AMD hater, but they need consistency before I can get excited.

keep it up? AMD pulled this off while they were down, imagine what they can do once they're back on their feet!

AMD is down in the single digit percentage points on per-core performance. AMD is DESTROYING Intel on core density. AMD can pack twice as many cores in a 1U server rack as Intel and in multithreaded workloads, that's all that matters. 128 intel cores to 128 AMD cores, Intel performance gains are less than 10% and costs about 4X as much. For half the cost of an Intel setup you can have twice as many AMD cores. Or, simply put(In the layest of layman terms) you can get an 80% performance increase over an intel setup for half the cost in workloads where core count, threads and PCIe bandwidth are the key factors.

 

[Vulcanproject]

How long can AMD keep it up is the question. I know I come off as an AMD hater, but they need consistency before I can get excited.

They now rely on TSMC, and TSMC look to have the process lead and are going to reach yet another smaller node before anyone else, including Intel. So for the foreseeable future AMD are looking very good on the server side.

As for things like future platform support most customers have seen how keen AMD are making their platform forwards compatible for as long as possible, unlike Intel. If you built an Epyc 'Naples' setup 18 months ago you will be able to drop in Epyc Zen 2 'Rome' parts and effectively double your core count per socket this year while dramatically increasing power efficiency. You will also be able to use them in Epyc Zen 3 'Milan' in another 18 months!

It's another big bonus to companies looking at the costs of these platforms weighing them against Xeons. That must be extremely attractive to cloud based services who are expanding as fast as possible it seems.

 

[hahahanoobs]

They now rely on TSMC, and TSMC look to have the process lead and are going to reach yet another smaller node before anyone else, including Intel. So for the foreseeable future AMD are looking very good on the server side.

Yields are better at TSMC.

As for things like future platform support most customers have seen how keen AMD are making their platform forwards compatible for as long as possible, unlike Intel.

Source?

If you built an Epyc 'Naples' setup 18 months ago you will be able to drop in Epyc Zen 2 'Rome' parts and effectively double your core count per socket this year while dramatically increasing power efficiency. You will also be able to use them in Epyc Zen 3 'Milan' in another 18 months!

That's a big IF! You'd have to already be using AMD to make this an advantage, and current market share doesn't show companies moving to AMD at any time in droves. And who are these customers? I'd like to see the list.

It's another big bonus to companies looking at the costs of these platforms weighing them against Xeons. That must be extremely attractive to cloud based services who are expanding as fast as possible it seems.

There you go again with costs. Companies looking for high(est) performance have no problems paying a premium. This is a fact easily found on the internet. Also, you have yet to mention two very important factors such as power efficiency and clock speeds.

Please stop focusing on cost and core counts.

 

[yRaz]

Please stop focusing on cost and core counts.

what you don't seem to understand is that enterprise workloads are highly multi threaded meaning that core count has more to do with performance than clock speed. With AMD we are getting close to negligible differences in per-core performance, but we are getting ABSURDLY more cores.

For consumer performance, fewer, faster cores are the way to go, The opposite is true in the enterprise space. Go to newegg and compare any consumer CPU to it's server analog. Consumer CPU's have few cores and higher clocks, server CPU's have more cores and low clocks. These are what people are buying, go look. Stop arguing on the internet and go do some CPU shopping as your own personal research. Amazon is putting their money where their mouth is. I don't understand what you are trying to argue at this point.

 

[Evernessince]

How long can AMD keep it up is the question. I know I come off as an AMD hater, but they need consistency before I can get excited.

Lisa Sue stated that AMD's goal was to deliver consistent product gains each new generation. This is why you want an engineer at the helm, they are product focused. Not like the bean counter Intel just appointed.

 

[redgarl]

Unrelated, but I just saw this and my jaw dropped. Intel is not messing around when it comes to graphics and gaming. This is HUGE!

"Intel Snatches Up Tom Peterson (formerly) Of NVIDIA, Will Be Joining As An Intel Engineering Fellow"

Yay.... one guy army... /sarcasm

One guy is not going to make a difference. By they way... graphic gaming...! Are you nutz? They are working on everything EXCEPT gaming GPU. They have a datacenter business and this is where their GPU division is going, nowhere else.

 

[redgarl]

I am well aware of various other major factors that extend well beyond the simple hardware a company will purchase from AMD.

All your points are very valid, but the answer to many of them lies between the lines of this article itself. The world's second largest provider (only just second by revenue) of cloud services is investing heavily into AMD's infrastructure.

That speaks volumes.

How long can AMD keep it up is the question. I know I come off as an AMD hater, but they need consistency before I can get excited.

People are not seeing it, but Google Stadia, MS Xbox servers and PS5, are a statement that AMD is gaming... and that AMD means business for the next decade. If I was Nvidia and Intel, I would thread carefully. A mistake like 10nm and RTX can pass when AMD is norming, but not performing.

 

[redgarl]

They now rely on TSMC, and TSMC look to have the process lead and are going to reach yet another smaller node before anyone else, including Intel. So for the foreseeable future AMD are looking very good on the server side.

Yields are better at TSMC.

As for things like future platform support most customers have seen how keen AMD are making their platform forwards compatible for as long as possible, unlike Intel.

Source?

If you built an Epyc 'Naples' setup 18 months ago you will be able to drop in Epyc Zen 2 'Rome' parts and effectively double your core count per socket this year while dramatically increasing power efficiency. You will also be able to use them in Epyc Zen 3 'Milan' in another 18 months!

That's a big IF! You'd have to already be using AMD to make this an advantage, and current market share doesn't show companies moving to AMD at any time in droves. And who are these customers? I'd like to see the list.

It's another big bonus to companies looking at the costs of these platforms weighing them against Xeons. That must be extremely attractive to cloud based services who are expanding as fast as possible it seems.

There you go again with costs. Companies looking for high(est) performance have no problems paying a premium. This is a fact easily found on the internet. Also, you have yet to mention two very important factors such as power efficiency and clock speeds.

Please stop focusing on cost and core counts.

Hmmmm... you are funny...

Cost do matters especially when you are google... and that you have 10 000 to 100 000 datacenters.

Same for AWS, cost matters and it is one of the three main focus of any project, cost, schedule and scope.

https://image.slidesharecdn.com/fin.../95/amd-next-horizon-64-638.jpg?cb=1541695599

Also, if AWS or Google wants to use AMD IF scalability feature, they need PCIe 4.0 and probably an EPYC platform. As you see, there IS some ground for companies to invest in EPYC.

https://www.techpowerup.com/img/9FEhHYVOQeNFApsr.jpg

 

[hahahanoobs]

Hmmmm... you are funny...

Cost do matters especially when you are google... and that you have 10 000 to 100 000 datacenters.

Same for AWS, cost matters and it is one of the three main focus of any project, cost, schedule and scope.

https://image.slidesharecdn.com/fin.../95/amd-next-horizon-64-638.jpg?cb=1541695599

Also, if AWS or Google wants to use AMD IF scalability feature, they need PCIe 4.0 and probably an EPYC platform. As you see, there IS some ground for companies to invest in EPYC.

https://www.techpowerup.com/img/9FEhHYVOQeNFApsr.jpg

Never said their wasn't ground. I'm just reminding you EPYC right now still isn't the overall winner against Xeons, no matter what Intel charges for them.

I'm no expert, but I don't see RTG being successful in the long run given its history and current marketshare, and AMD should look to sell and focus on CPU's and maybe consoles if Navi isn't a home run.

 

[wiyosaya]

That Amazon is investing in AMD is a good thing.

As I have said before, AMD was wise to go after the server market with Zen - even though gamers were not happy with the result. Zen was ahead in multi-threaded benchmarks from its release - and that is what counts in the server market. By far, the server market is where the money is.

 

[wiyosaya]

Never said their wasn't ground. I'm just reminding you EPYC right now still isn't the overall winner against Xeons, no matter what Intel charges for them.

You might want to have a look at this - https://www.phoronix.com/scan.php?page=article&item=power9-epyc-xeon&num=1

 

[Puiu]

That Amazon is investing in AMD is a good thing.

As I have said before, AMD was wise to go after the server market with Zen - even though gamers were not happy with the result. Zen was ahead in multi-threaded benchmarks from its release - and that is what counts in the server market. By far, the server market is where the money is.

It is simple. AMD is not targeting server replacements of modern xeon servers. they are targeting the new server market. as long as OEMs offer adequate support, which they generally do, you'll see more and more buy AMD based servers. they are rapidly approaching the 15% mark.
thing aren't as complicated as you think.

 

[Dimitrios]

INTEL's time has past even with all the cheating and dirty tricks this is what happens when they have to play fair. Good to see AMD doing well, karma which I don't believe in really well this is a perfect example of Karma. I hope INTEL hurts, been watching both sides since I was a kid around the early 1990's and building systems by both teams. Deep down I bleed GREEN both for AMD and Kawasaki ( street bike rider).

 

[msroadkill612]

You will be able to have 64 cores on a single socket, or 128 cores on a dual socket board this year with 7nm power efficiency and relatively low purchase cost. Much higher performance per watt than any Xeon that will be available this year and possibly most of next year.

AMD are going to capture great swathes of the server market, that much is certain.

Yes, but we've seen this movie before. AMD will gain a small slice while Intel dust themselves off. Then a couple of years later Intel comes back with a vengeance (performance improvements + price slashing) and pushes AMD back out like they always do. I hope this time it will be different, but no one should expect Intel to take this lying down.

I am reminded of a Stalin quote when I hear your tired script. When told the pope wouldnt approve of some action of his... "How many divisions does the Pope have?"

Apart from your mantra - with what is Intel to respond? They have nothing, not even credible promises.

AMD only have to cross some Ts and dot some I's on ~demonstrated, real, existing tech to fulfil their astonishing roadmap.

Dont regurgitate that "trust me" mantra. Give us credible arguments and we will listen.