New 2nd Gen AMD EPYC Processors Redefine Performance for Database, Commercial HPC and Hyperconverged Workloads

[AleksandarK]

AMD today announced it is extending the 2nd Gen AMD EPYC processor family with three new processors that combine the balanced and efficient AMD Infinity architecture with higher speed "Zen 2" cores for optimal performance on database, commercial high-performance computing (HPC) and hyperconverged infrastructure workloads.

The three new processors, the AMD EPYC 7F32 (8 cores), EPYC 7F52 (16 cores) and EPYC 7F72 (24 cores), expand 2nd Gen AMD EPYC performance leadership into workloads that can leverage up to 500 MHz of additional base frequency, and large amounts of cache, making AMD EPYC the world's highest per core performance x86 server CPU.



The AMD EPYC 7Fx2 processors provide new performance capabilities for workloads in the heart of the enterprise market including database with up to 17% higher SQL Server performance compared to the competition, hyperconverged infrastructure with up to 47% higher VMmark 3.1 score (using vSAN as the storage tier in a 4-node cluster) compared to the competition for a new world record, and commercial high-performance computing (HPC) with up to 94% higher per core computational fluid dynamics individual application performance compared to the competition.

"AMD EPYC continues to redefine the modern data center, and with the addition of three powerful new processors we are enabling our customers to unlock even better outcomes at the heart of the enterprise market," said Dan McNamara, senior vice president and general manager, server business unit, AMD. "With our trusted partners, together we are pushing the limits of per core performance and value in hyperconverged infrastructure, commercial HPC and relational database workloads."

A Balanced System That's More than Gigahertz
The new 2nd Gen AMD EPYC 7Fx2 processors provide leading per core performance and breakthrough value, while adding the highest per core performance of the EPYC family.

The performance of these new processors comes from a balanced architecture that combines high-performance "Zen 2" cores, innovations in system design like PCIe 4 and DDR4-3200 memory5, and the AMD Infinity architecture, to provide customers with optimum system performance that enables better real world application performance.

Details of the new processors are below.


Ecosystem Growing with AMD EPYC
The ecosystem of OEMs, cloud providers, ISVs and IHVs using 2nd Gen AMD EPYC processors continues to grow, with existing OEMs and new partners adopting the new AMD EPYC 7Fx2 processors.

• Dell Technologies will support all three processors across its entire lineup of AMD EPYC based Dell EMC PowerEdge servers, including the R6525 which holds a world record 2P Four-Node Benchmark Result on VMmark 3 with VMware vSAN 1. "These new AMD EPYC 7Fx2 processors enable Dell EMC PowerEdge servers to drive substantial performance benefits for customer business applications like database and hyperconverged infrastructure, where Dell EMC PowerEdge servers hold a world record in benchmark performance. Our customers will truly benefit from these new processors as we continue to grow our AMD EPYC family of PowerEdge platforms." - Rajesh Pohani, vice president, Server Platform Product Management, Dell Technologies
• HPE continues to expand its offerings using 2nd Gen AMD EPYC processors with latest support of HPE SimpliVity, an intelligent hyper-converged infrastructure solution. HPE will also support all three AMD EPYC 7Fx2 processors on the recently announced HPE Apollo 2000 Gen10 Plus system, HPE ProLiant DL385 Gen10 Plus server and HPE ProLiant DX servers. "We are pleased to expand support of the 2nd Gen AMD EPYC processors across our portfolios, which include new additions with the HPE Apollo 2000 Gen10 Plus system, HPE ProLiant DL385 Gen10 Plus server and HPE ProLiant DX servers to meet high-frequency and performance needs for our customers in high-performance computing and database environments." - Peter Ungaro, senior vice president and general manager, HPC and Mission Critical Solutions (MCS), at HPE
• IBM Cloud is the first cloud provider to offer its clients the AMD EPYC 7F72 processors in their bare metal offering, providing access to fast, high core-count dual socket bare metal servers. Additionally, IBM recently announced the availability of its first bare metal server powered by the AMD EPYC 7642 processor. "We are excited to be the first cloud provider to support the new AMD EPYC 7F72 processor. Now, IBM Cloud provides access to another high core-count dual socket bare metal server with high clock speed frequency, giving our clients more optimized platform choices for compute-intense workloads such as analytics, commercial HPC and EDA. We stay committed to enabling flexible and powerful bare metal experiences for clients to enhance performance and throughput." - Satinder Sethi, general manager, IBM Cloud Infrastructure Services
• Lenovo will support the new AMD EPYC 7Fx2 processors on its ThinkSystem SR635 and SR655 platforms. These ThinkSystem platforms are already a great choice for a variety of enterprise workloads including data analytics, software defined storage and infrastructure for remote workers. Lenovo's storage and PCIe capabilities coupled with AMD EPYC core count and I/O density will help provide customers with choice as their business needs evolve. These new higher frequency 2nd Gen AMD EPYC processors, with an increased core clock speed up to 15%, in the single socket ThinkSystem platform, provides customers with greater options for workloads where per core performance is critical. Lenovo's one socket optimized platforms with these new processors allow customers to deploy these platforms where traditionally two socket systems were used, providing power and SW licensing costs savings. "Today's business dynamics are presenting customers with new challenges to improve speed, cost and performance. We feel confident we have the right portfolio to provide our customers with enhanced choice as organizations look to enable remote working capabilities and manage their increased data and storage requirements." - Kamran Amini, vice president and general manager, Server, Storage and Software Defined Infrastructure, Lenovo Data Center Group
• Microsoft recognizes the impact the new AMD EPYC 7Fx2 processors have on providing Microsoft data platform customers the best experience possible, including an up to 17% higher SQL Server2 TPM per core performance. "Microsoft data platform solutions help customers release the potential hidden in data and reveal insights and opportunities to transform a business. A critical part of this process is making sure a database has access to an efficient, powerful and fast processor and that's exactly what the new AMD EPYC 7Fx2 processors provide Microsoft data platform solutions customers." - Jamie Reding, SQL Server program manager, Microsoft
• Nutanix, in conjunction with Hewlett Packard Enterprise, announced that it expects that Nutanix HCI software will be supported on select AMD EPYC based HPE ProLiant servers by May. As well, HPE announced the upcoming availability of AMD EPYC 7Fx2 processors on HPE ProLiant DX servers in Q3. "We are excited to have validated Nutanix's HCI software for 2nd Gen AMD EPYC processor based HPE ProLiant systems. This will bring 2nd Gen AMD EPYC processor support to Nutanix software, giving more flexibility and choice to our customers while unleashing greater workload performance for databases, analytics, VDI and other virtualized business critical applications." - Tarkan Maner, chief commercial officer, Nutanix
• Supermicro is launching the industry's first blade platform built for 2nd Gen AMD EPYC processors with immediate support for the new AMD EPYC 7Fx2 processors combined with integrated 25G Ethernet and optional 100G EDR InfiniBand support with 200G HDR in the near future. In addition, all Supermicro A+ platforms including Ultra, GPU, WIO, Twin and Mainstream systems will support the new AMD EPYC 7Fx2 processors immediately. "Adding the new SuperBlade platform to our extensive portfolio of products supporting the 2nd Gen AMD EPYC processors gives our customers another powerful choice when redefining their modern data center. Leveraging support for the new AMD EPYC 7Fx2 processors, our latest SuperBlade and Supermicro A+ platforms further excel at database, EDA and other data-intensive workloads." - Vik Malyala, senior vice president, Field Application Engineering & Business Development, Supermicro
• VMware, a leading innovator in enterprise software, is pleased to add support for the new 2nd Gen AMD EPYC 7Fx2 processors, enabling customers access to powerful virtualization platforms. "The 2nd Gen AMD EPYC 7Fx2 processors bring new value to VMware customers. They provide a unique balance of strong per core performance coupled with an industry-leading per-processor memory capacity of 4 TB. A key element of VMware vSphere, vSAN, and now VMware Cloud Foundation market success has been our commitment to helping customers quickly adopt the latest hardware innovation." - Richard A. Brunner, chief technology officer, Server Platform Technologies, VMware

The new processors are available now through multiple OEMs and IBM Cloud. You can learn more about where to get systems with the new processors here.

View at TechPowerUp Main Site

 

[notb]

Finally!
I honestly don't know why these CPUs weren't launched in 2019 with everything else. Anyone knows?

Looking at earlier discussions, some people on this forum may have a hard time accepting an 8-core server CPU in 2020. Let alone one for $2100.

 

[deu]

Finally!
I honestly don't know why these CPUs weren't launched in 2019 with everything else. Anyone knows?

Looking at earlier discussions, some people on this forum may have a hard time accepting an 8-core server CPU in 2020. Let alone one for $2100.

My guess would be: 8 chiplets + binning combined with added complexity compared to mainstream cpus (3000-series)

 

[Chrispy_]

Finally!
I honestly don't know why these CPUs weren't launched in 2019 with everything else. Anyone knows?

Looking at earlier discussions, some people on this forum may have a hard time accepting an 8-core server CPU in 2020. Let alone one for $2100.

These are the results of selective binning - presumably these are chiplets that meet the efficiency/clock requirements for use in an Epyc CPU and have some kind of core defects but still functioning cache.

My server estate is mostly old Xeons but of the two Epyc servers, it's all about cache sizes and PCIe bandwidth for minimising storage bottlenecks (remote GPU crunching nodes for massive datasets on pointcloud site surveys and environmental analysis)

Not every server needs a bazillion cores.

 

[xkm1948]

Would they put in AVX512 for Zen3 based EPYC??

Finally!
I honestly don't know why these CPUs weren't launched in 2019 with everything else. Anyone knows?

Looking at earlier discussions, some people on this forum may have a hard time accepting an 8-core server CPU in 2020. Let alone one for $2100.

Like everyone else said, they probably need to collect enough top of the tier silicon to offer lots of EPYC at once. After all datacenters purchase in bulk. You can declare something is available for sell while only have a handful of them in stock,

Just my $0.02

 

[TheLostSwede]

AMD’s New EPYC 7F52 Reviewed: The F is for ᴴᴵᴳᴴ Frequency

www.anandtech.com

 

[notb]

My guess would be: 8 chiplets + binning combined with added complexity compared to mainstream cpus (3000-series)

Well, 7F32 uses 4 chiplets, 7F52 uses 8. Both need just 2 cores/chiplet, at frequencies that all consumer SoCs can achieve.
So the way I see it, most chiplets that ended up in a Ryzen 5 3600 could have been used for any of these new server chips. You agree?

3600 launched at $200.
7F32 launches at $2100.

So I imagine 7F32 is more profitable way of using chiplets.
And it addresses a really large niche in on-premise servers, so why not?

 

[hat]

Finally!
I honestly don't know why these CPUs weren't launched in 2019 with everything else. Anyone knows?

Looking at earlier discussions, some people on this forum may have a hard time accepting an 8-core server CPU in 2020. Let alone one for $2100.

You're right this time, that $2100 price tag for an 8 core chip seems just a tad high... and then the 16 core chip has more cache and is more expensive than the 24 core chip. What?

 

[notb]

Like everyone else said, they probably need to collect enough top of the tier silicon to offer lots of EPYC at once. After all datacenters purchase in bulk. You can declare something is available for sell while only have a handful of them in stock,

And as I said: I really don't understand what differs these chiplets from what AMD has been putting in all those Ryzen SoCs for almost a year.
Wasn't this the whole point of the chiplet idea?

And this is not really a "datacenter chip" for cloud or server providers (albeit could be used like that as well). The high-core models are more cost-effective for providers.
It's a typical workhorse for all those 8-16 core racks that run databases or ERP instances in pretty much every large company.
The direct competition would be high-frequency Intel Xeon Gold (like 6244).

Still, obviously, they have to make tens of thousands of these to attract large OEMs like Dell. But they should definitely be capable of that *if* this can be made up of the same chiplets as consumer stuff.

You're right this time, that $2100 price tag for an 8 core chip seems just a tad high...

It absolutely isn't. Xeon 6244 (8C/16T) is $3000.

A lot of software is licensed based on core count - databases being the prime example.
These low-core, high-frequency server chips really matter. Seriously. That's the type of server chip AMD should have started with...

and then the 16 core chip has more cache and is more expensive than the 24 core chip. What?

7F52 needs 8 chiplets. 7F72 only needs 6, so it's cheaper to make.

 

[Nater]

You're right this time, that $2100 price tag for an 8 core chip seems just a tad high... and then the 16 core chip has more cache and is more expensive than the 24 core chip. What?

And like notb said, you'll end up money ahead by not having to pay for per core software licensing.

 

[AnarchoPrimitiv]

You're right this time, that $2100 price tag for an 8 core chip seems just a tad high... and then the 16 core chip has more cache and is more expensive than the 24 core chip. What?

A Tad high compared to what? It's still several hundred dollars cheaper than it's most direct Xeon Gold competitor, so I'm really interested as to what you're basing the "tad high" analysis on?

 

[hat]

It just seems a lot for an 8 core chip in 2020... that's all.

 

[Indurain]

EPYC is all about low power, memory channels and PCIE lanes
7f52 = 8 memory channels and 128 PCIE Gen 4 lanes, for example, an a heck of a lot less power hungry than Threadripper with only 1/2 the memory channels and PCIE lanes.

 

[bug]

It absolutely isn't. Xeon 6244 (8C/16T) is $3000.

The Xeon is also a 150W part and may make that money back in electrical bills. I guess the real differentiator here is that huge cache.

 

[dicktracy]

And for a low price of $2100... Of course, no community outrage because it's AMD.

 

[Chrispy_]

A lot of software is licensed based on core count - databases being the prime example.
These low-core, high-frequency server chips really matter. Seriously. That's the type of server chip AMD should have started with...

QFT. When the annual license fee is a 4-digit figure per core, then the cost of the processor is irrelevant as long as the per-core performance is best in class.

 

[remixedcat]

Damn still too high for mah stim check. Looks nice tho

 

[notb]

The Xeon is also a 150W part and may make that money back in electrical bills. I guess the real differentiator here is that huge cache.

I wouldn't overestimate the importance of electricity bills with these CPUs, but there's certainly a TDP gap.

Frankly, Intel really doesn't have a direct competitor. The fastest 16-core Xeon is the 6246R. It's 3.4/4.1GHz for 205W.
205W is the max TDP for current Xeons with less than 28 cores. Only a handful of Xeon Platinum go above.
AMD went for 240W and OEMs accepted it, so we're likely to see new models in Intel's lineup very soon.

As for performance: Anandtech compared it to the 6226R (150W).
EPYC is faster, but it obviously has way more power to work with. 6246R would close the gap considerably.

AMD’s New EPYC 7F52 Reviewed: The F is for ᴴᴵᴳᴴ Frequency

www.anandtech.com

EPYC is all about low power, memory channels and PCIE lanes
7f52 = 8 memory channels and 128 PCIE Gen 4 lanes, for example, an a heck of a lot less power hungry than Threadripper with only 1/2 the memory channels and PCIE lanes.

Funny that now, since AMD launched a high-TDP EPYC, you suddenly compare it to a HEDT chip. What's the point?
You shouldn't stop there. Why not compare it to 9900K?

And for a low price of $2100... Of course, no community outrage because it's AMD.

Absolutely. AMD started with high core-count, low power and relatively cheap models. We had to endure all the moaning about Xeons pulling 250W (or 400W).
These new EPYCs are obviously more power hungry and more expensive per core than what AMD launched earlier.
A 400W, 32-core EPYC is probably around the corner (if the socket can handle it).

I'm looking forward to comments from AMD fans, although I'm slightly worried we'll see more comparisons to products from different segments - like the one above. Well... at least he didn't compare to a microwave...

 

[R0H1T]

These new EPYCs are obviously more power hungry and more expensive per core than what AMD launched earlier.

And you obviously know why, right & yet keep coming back to the same inane arguments as ever?

A 400W, 32-core EPYC is probably around the corner (if the socket can handle it).

Okay so if customers want it then AMD will make it, this product with full cache wasn't borne our of some nerd or enthusiast's wet dream ~ it obviously had takers & backers!

I'm looking forward to comments from AMD fans, although I'm slightly worried we'll see more comparisons to products from different segments - like the one above. Well... at least he didn't compare to a microwave...

Great, looking forward to more but but but ~ from you

Spoiler

 

[notb]

And you obviously know why, right & yet keep coming back to the same inane arguments as ever?
Okay so if customers want it then AMD will make it, this product with full cache wasn't borne our of some nerd or enthusiast's wet dream ~ it obviously had takers & backers!

But I precisely said I like these CPUs and it's the type of product AMD was missing.

Do you still read what you're replying to? Or have you already labeled everyone as pro-AMD (=> give a like) and anti-AMD (=> bash with a generic comment)?

 

[R0H1T]

We had to endure all the moaning about Xeons pulling 250W (or 400W).

You mean like these? You do know which CPU's from Intel consume 400W, actually more at full load, don't you?
Wonder what do you want me to infer from this

 

[notb]

You mean like these? You do know which CPU's from Intel consume 400W, actually more at full load, don't you?

Every CPU can pull more than it's TDP (AMD's as well). And every CPU can be set to comply with TDP, which actually comes as standard in server chips.
If you have difficulties tracking this, find a review that tests chips running at TDP / PL1, like this from Anandtech:

AMD’s New EPYC 7F52 Reviewed: The F is for ᴴᴵᴳᴴ Frequency

www.anandtech.com

You can see server chips are just under TDP. This means that performance results this review shows can be analyzed in relation to TDP (unlike in some reviews where CPUs are allowed to boost or are limited - depending on what platform allows).

It's not that hard to notice you're totally uninterested in server products and you only come to the server-related topics to attack someone - usually using some consumer-product observations.
Still, I have to quote one part of the earlier comment:

Okay so if customers want it then AMD will make it, this product with full cache wasn't borne our of some nerd or enthusiast's wet dream ~ it obviously had takers & backers!

I absolutely agree with this.
But why do you criticize Intel chips all the time? Doesn't this apply as well?

 

[R0H1T]

Every CPU can pull more than it's TDP (AMD's as well). And every CPU can be set to comply with TDP, which actually comes as standard in server chips.

Server chips generally have very good "representative" TDP, no need to remind me of PL1 or PL2 unless we're also going to microanalyse the use case? I have more issues with Intel's desktop or mobile lineup's "TDP" though! The issue, my issue, is where did this 250-400W TDP come into play?
Everyone (informed) knows that with higher clocks & extra cache you have to raise the TDP, the "400W" parts were more stunt than anything else!
AT sums up the general feeling really well ~

The chips you're seeing from AMD, although not as high volume as regular EPYC, are definitely not for show!

I absolutely agree with this.
But why do you criticize Intel chips all the time? Doesn't this apply as well?

Well I like to criticize Intel, never been a fan of "Big Blue" except during a brief period around Conroe, my reasons may sound selfish (or too idealistic) so I won't go into a deep dive over them. I do however still use Intel, more than AMD, though that's largely due to the exorbitant AMD prices over here & AMD's lack of class leading mobile chips, at least till now.

 

[deu]

Well, 7F32 uses 4 chiplets, 7F52 uses 8. Both need just 2 cores/chiplet, at frequencies that all consumer SoCs can achieve.
So the way I see it, most chiplets that ended up in a Ryzen 5 3600 could have been used for any of these new server chips. You agree?

3600 launched at $200.
7F32 launches at $2100.

So I imagine 7F32 is more profitable way of using chiplets.
And it addresses a really large niche in on-premise servers, so why not?

We agree, but I honestly think AMD planned it to go down this way when it comes to marketshare in consumermarket first, and then next gen in server. The consumer market NEEEDED to be won with 3000 series. The market have been dry for a long time due to demand.