AMD means business with latest Epyc and Ryzen Pro CPU launches

Bob O'Donnell

Posts: 80   +1
Staff member
The big picture: Chipmaker AMD is on a roll, particularly for consumer-focused products. Not only are its Ryzen 5000 processors now the fastest CPUs you can buy for consumer PCs, its Radeon RX 6000 GPUs are giving market leader Nvidia a run for their money on the performance side. Oh yeah, and let’s not forget that AMD is also powering two of the most highly sought-after tech gadgets currently available (or more likely, still “unavailable” for many eager to purchase them): the Sony PlayStation 5 and Microsoft Xbox Series X. Unbeknownst to many, however, AMD has been building up its business-focused offerings as well.

Rarely has this been more evident than with the back-to-back launches of its latest 3rd-gen Epyc server CPUs and its Ryzen Pro 5000 series mobile processors for enterprise notebooks.

Both the Epyc 7003 family of chips (codenamed “Milan”) and Ryzen Pro 5000 mobile processors (like consumer versions, codenamed “Cezanne”) are based on the company’s Zen 3 core architecture—a 7nm design that continues the impressive performance path that the company began with the original Zen cores that debuted in 2017.

While each of the new chips is optimized for their very different needs and applications, they do share a number of enhancements that are at the heart of Zen 3 cores, including larger, unified caches for faster processing with larger data sets—as many AI-focused workloads on both servers and PCs require.

Cezanne Detailed Die Shot

Zen 3 lets all eight of the available CPUs in each core complex access the full amount of L2 cache (32 MB for Epyc and 16 MB for Ryzen Pro, both of which are twice the size of their predecessors). This is a significant enhancement over Zen 2 designs, which had groups of four cores that were each able to access 16 MB or 8 MB, respectively. This new architecture makes it easier for multi-threaded applications capable of simultaneously using all cores to fully leverage the cache, resulting in significantly improved performance for these types of programs. Again, that’s something that’s very useful in both servers and notebooks, but for entirely different types of applications.

In addition to performance enhancements—including a 19% improvement in the critical Instructions Per Clock (IPC) metric on the Zen 3 cores found in both chips—AMD added several hardware-based security improvements, which are becoming increasingly important for businesses of all sizes. Both the Epyc 7003 and the Ryzen 5000 Pro offer Shadow Stack, for example, which is designed to deter hardware-focused malware attacks, such as Spectre, Meltdown and other memory location-focused threats.

AMD Epyc 7003 Series Processor

For the Epyc 7003, the company also added a suite of security features it calls Infinity Guard that includes support for securing virtual machines and hypervisors, through Secure Encrypted Virtualization-Encrypted State (SEV-ES) and Secure Nested Paging (SEV-SNP). These capabilities are a big reason why Microsoft announced that it would now be offering Confidential Computing-focused VMs in Azure that leverage these third generation Epyc parts.

On the notebook side, the Ryzen Pro 5000 series CPUs add support for Microsoft’s Secured Core PC initiatives as well as FIPS 140-3 level encryption certification. Together, these new security features represent a portion of what the company describes as a multi-layered security model.

For the enterprise data center and cloud computing world, the Epyc 7003 series builds on the impressive performance and TCO (Total Cost of Ownership) metrics that AMD first set with its second generation Epyc line back in 2019. The company is claiming its “Milan” CPUs offer significant performance leadership on key workloads for cloud computing, enterprise data centers, and HPC (high performance computing—I.e., supercomputers).

In fact, the company even cited comparative benchmarks that it says show how a single socket Epyc 7003-based server can outperform many two socket Intel Xeon-powered solutions on certain workloads, with correspondingly lower energy consumption (and a significantly lower purchase price). In another example, AMD cited the ability to run more than double the number of active virtual desktop sessions on an Epyc 7763-based server versus a comparatively configured offering with an Intel Xeon 6258R CPU.

AMD also announced an impressive arrange of new partnerships, including new 3rd generation Epyc-based cloud computing instances from Amazon AWS, Microsoft Azure, Google Cloud, IBM Cloud, Oracle Cloud, Tencent Cloud and Alibaba Cloud, along with new servers from Dell, HPE, Cisco, Lenovo, SuperMicro and many more.

For Ryzen Pro 5000, AMD had big name launch-day partners, including several different notebook designs from HP, Lenovo, and others. The company also cited its rapid increase in both market share and design wins (though, admittedly, from a small base in commercial notebooks). As with the Epyc line, AMD is claiming significant performance wins for the Ryzen Pro 5000 line versus Intel’s latest business notebook offerings, particularly on content creation applications and workloads, such as video encoding and working with Adobe’s Creative Cloud suite of tools. The company was also quick to point out significant improvements in battery life versus previous generation Ryzen mobile designs, as it did with the consumer-focused Ryzen 5000 series that it debuted during CES in January of this year.

Unlike those CPUs, however, the Ryzen Pro line also has a strong focus on manageability features, for IT departments, as well as image stability, to ensure that companies can have a stable, consistent software image and set of drivers on systems that pass their validation tests.

All told, the two days of business-focused CPU debuts offered a good opportunity to shine the spotlight on increasingly important, albeit lesser-known, parts of AMD’s business. Despite its strength in the consumer market, the company still has a long way to go to win over business computing buyers of all types. As these new Zen 3-based Ryzen and Epyc offerings highlight, however, the benefits of strong competition have never been clearer.

Bob O’Donnell is the founder and chief analyst of TECHnalysis Research, LLC a technology consulting firm that provides strategic consulting and market research services to the technology industry and professional financial community. You can follow him on Twitter .

Permalink to story.

 
Hopefully AMD will continue to gain strength, so it can finally also slay Nvidia.

Yes, I know, it will take time and effort and most important, undo the fantasy aura created by Nvidia marketing dept.

Intel and Nvidia need to be cut down a peg or two, thanks to their anticonsumer bs pushed down on us.

And if down the line AMD becomes like them, well, another slayer should arrive.
 
I think this chip really shows how badly AMD needs to move the IO chip to 7nm.

Otherwise another homerun by AMD.
 
Why do you say that? At most they might save a few watts from it, but there shouldn't be performance improvements.
idle power usage. Also at load it does use a fair amount of power.

It just makes their chips not as power efficient as it could be.

The Anandtech report does a pretty good job going over the new chip.

Zen 4 will have a new IO die on a smaller process.
 
I'm wondering how long before AMD starts slacking since there's no competition on Intels side. Boiling lake looks atrocious. Whatever you might save on these CPUs you'll pay through your electricity bill.
 
idle power usage. Also at load it does use a fair amount of power.

It just makes their chips not as power efficient as it could be.

The Anandtech report does a pretty good job going over the new chip.

Zen 4 will have a new IO die on a smaller process.

They already said the IO die doesn't scale as well with lower node plus AMD already struggles with demand, keeping the IO on 12nm is better for now until they figure out the best way to move it to 7nm and when the core dies are moved to 5nm
 
Last edited:
AMD will be the majority share in the data centers within the next 5 years unless intel can pull a miracle of innovation.
 
Assuming they can supply their customers with CPUs of course. This is one market where they need to deliver.
 
idle power usage. Also at load it does use a fair amount of power.

It just makes their chips not as power efficient as it could be.

The Anandtech report does a pretty good job going over the new chip.

Zen 4 will have a new IO die on a smaller process.


Part of the reason they did the Zen 2 I/O die from GF was to avoid any additional load on an already-expensive/overtaxed fab line (back in 2019). The reason they stayed with that is because, they have eveen more oad on that same supply line.

If your I/O die size would be even larger than your chip, the fab impact would be 66% less Zen 3 core shipping.
 
idle power usage. Also at load it does use a fair amount of power.

It just makes their chips not as power efficient as it could be.

The Anandtech report does a pretty good job going over the new chip.

Zen 4 will have a new IO die on a smaller process.
I doubt it's that good and worth the investment. AMD chips are already efficient. Saving 2-5W under load and maybe 1-2 in idle is not that amazing and even the size won't scale that well.

Zen 4 should have a smaller I/O because the core production will move to 5nm and it will free up 7nm.
 
I think this chip really shows how badly AMD needs to move the IO chip to 7nm.

Otherwise another homerun by AMD.

lol, it's a obvious strategy in where the more mature proces ensures more I/O dies out of it.

This chip is cheaper for AMD to produce then intel could ever do at this point.
 
I doubt it's that good and worth the investment. AMD chips are already efficient. Saving 2-5W under load and maybe 1-2 in idle is not that amazing and even the size won't scale that well.

Zen 4 should have a smaller I/O because the core production will move to 5nm and it will free up 7nm.

Zen4 will use 12/14nm I/O die. Not any uncertainty about that one.

Why? Because AMD is still under WSA until 2024. That means AMD either must use something GF has to offer (12nm/14nm are best ones) or pay penalty. Since I/O die contents also do not scale well, it makes no sense for AMD to use 7nm or anything better than 12/14nm for Zen4 I/O die.

2024 things may be different.

Part of the reason they did the Zen 2 I/O die from GF was to avoid any additional load on an already-expensive/overtaxed fab line (back in 2019). The reason they stayed with that is because, they have eveen more oad on that same supply line.
Only reason was, again, WSA. AMD had permission to use TSMC without penalties because GlobalFoundries strictly said they will not have enough 7nm capacity. AMD maximized TSMC 7nm output with chiplet+I/O die combo. AMD was still supposed to get majority of 7nm output from GF but as we know that just didn't work.
 
Zen4 will use 12/14nm I/O die. Not any uncertainty about that one.

Why? Because AMD is still under WSA until 2024. That means AMD either must use something GF has to offer (12nm/14nm are best ones) or pay penalty. Since I/O die contents also do not scale well, it makes no sense for AMD to use 7nm or anything better than 12/14nm for Zen4 I/O die.

2024 things may be different.


Only reason was, again, WSA. AMD had permission to use TSMC without penalties because GlobalFoundries strictly said they will not have enough 7nm capacity. AMD maximized TSMC 7nm output with chiplet+I/O die combo. AMD was still supposed to get majority of 7nm output from GF but as we know that just didn't work.
the 2019 revised WSA doesn't limit AMD's ability to use 7nm for its I/O dies in Zen4. They can still make 12nm dies for the 5000 series which will be used by many older AMD CPU owners for upgrades (moving from 2000/3000 series). (and AMD will definitely not keep it on 12nm till 2024)
 
the 2019 revised WSA doesn't limit AMD's ability to use 7nm for its I/O dies in Zen4. They can still make 12nm dies for the 5000 series which will be used by many older AMD CPU owners for upgrades (moving from 2000/3000 series). (and AMD will definitely not keep it on 12nm till 2024)
WSA allows AMD to use any manufacturing tech from anywhere that is under 12nm. However, AMD must still meet purchase targets for 12nm and above from GF. AMD must either meet purchase targets or pay for nothing.

Those targets are negotiable but then again, GF didn't took care of AMD's manufacturing for free and WSA ends no sooner than 2024. Targets from 2019 WSA are set to end on this year. Now, AMD has quite strong negotiation position since GF don't have any modern process (compared to TSMC, Intel, Samsung) available. That way AMD could get Very good deal from GF. But, AMD also does not have many good products for 12/14nm. Those are same old ones: chipsets and I/O dies.

For servers, Zen3 Epyc's AMD just released are more than good enough against Intel's 2024 server parts. AMD just don't need anything better until 2024 (they will release new Epyc's for sure but even current Zen3 ones are more than good enough and so are future Epyc's with 12nm/14nm IO dies).

For manufacturing capacity, TSMC 7nm is more than good enough for other than top end products. That way AMD could get much more market share. Using 7nm capacity for I/O dies, does not help there.

All in all, AMD has no need or reasons to make IO dies anything "better" than 12/14nm. Perhaps they will change it before 2024, since old products will sell for some time and AMD will 2023 products on 2024 too.
 
WSA allows AMD to use any manufacturing tech from anywhere that is under 12nm. However, AMD must still meet purchase targets for 12nm and above from GF. AMD must either meet purchase targets or pay for nothing.

Those targets are negotiable but then again, GF didn't took care of AMD's manufacturing for free and WSA ends no sooner than 2024. Targets from 2019 WSA are set to end on this year. Now, AMD has quite strong negotiation position since GF don't have any modern process (compared to TSMC, Intel, Samsung) available. That way AMD could get Very good deal from GF. But, AMD also does not have many good products for 12/14nm. Those are same old ones: chipsets and I/O dies.

For servers, Zen3 Epyc's AMD just released are more than good enough against Intel's 2024 server parts. AMD just don't need anything better until 2024 (they will release new Epyc's for sure but even current Zen3 ones are more than good enough and so are future Epyc's with 12nm/14nm IO dies).

For manufacturing capacity, TSMC 7nm is more than good enough for other than top end products. That way AMD could get much more market share. Using 7nm capacity for I/O dies, does not help there.

All in all, AMD has no need or reasons to make IO dies anything "better" than 12/14nm. Perhaps they will change it before 2024, since old products will sell for some time and AMD will 2023 products on 2024 too.
Not only that, 12/14nm GF is a lot cheaper per wafer than TSMC 7nm. If AMD shifts to 7nm for the IO, that extra cost would be passed onto consumers. Zen 3 has already upped it's price, although I assume Lisa Su wants to pay off debt so funds can be spread better within the company to make better products.
 
Not only that, 12/14nm GF is a lot cheaper per wafer than TSMC 7nm. If AMD shifts to 7nm for the IO, that extra cost would be passed onto consumers. Zen 3 has already upped it's price, although I assume Lisa Su wants to pay off debt so funds can be spread better within the company to make better products.

This. Zen3 pricing is justified because it's much better than anything Intel has.

Also AMD still don't have "7nm RX580" (mid level GPU on 7nm). Meaning AMD has still many 7nm products to be released.
 
Back