A peak against the core performance balance

The arrival of AMDs 3rd The generation EPYC processor family, which uses the new Zen 3 core, is highly anticipated. The promise of a new micro-architecture for processors, connectivity updates and new security options while still maintaining platform compatibility is a good measure of an enterprise platform update, but the One True Metric is platform performance. Seeing Zen 3 boosted the ultimate performance leadership per core in the consumer market in November, raised expectations for a similar victim in the enterprise market, and today we see the results.

AMD EPYC 7003: 64 cores from Milan

The key figures that AMD is promoting with the new generation of hardware are an increase in raw production performance of + 19% due to improvements with the new core design. In addition, AMD has new security features, optimizations for different memory configurations and updated performance with the Infinity Fabric and connectivity.


3rd gen EPYC

Anyone looking for the brief descriptions of the new EPYC 7003 series, known under the code name Milan, will be very familiar with the previous generation, but this time AMD is focusing on different design points.

Milan processors offer up to 64 cores and 128 wires using AMD’s latest Zen 3 cores. The processor is designed with eight chiplets of eight core points each, similar to Rome, but this time all eight core points are coupled into the chiplet, allowing for an effective dual L3 case design for a lower overall case delay structure. All processors have 128 lanes of PCIe 4.0, eight memory channels, with most models supporting dual processor connectivity, and new options for channel memory optimization are available. All Milan processors must be compatible with platforms in the Rome series with a firmware update.

AMD EPYC: Generation on Generation
AnandTech EPYC
7001
EPYC
7002
EPYC
7003
Code name Naples Rome Milan
Micro-architecture Zen Zen 2 Zen 3
Core manufacturing 14nm 7 nm 7 nm
Max core / wires 32/64 64/128 64/128
Core complex 4C + 8MB 4C + 16MB 8C + 32MB
Memory support 8 x DDR4-2666 8 x DDR4-3200 8 x DDR4-3200
Memory capacity 2 TB 4 TB 4 TB
PCIe 3.0 x128 4.0 x128 4.0 x128
Safety KMO
SEV
KMO
SEV
KMO
SEV
SNP
Peak Power 180 W 240 W * 280 W
* Rome introduced 280 W for special HPC mid-cycle

One of the highlights here is that the new generation of processors will offer 280 W models to all customers – previous generations only had 240 W models for everyone and then 280 W for specific HPC customers, but this time the customers may have the high performance enables parts with the new core design.

This is illustrated when we make direct comparisons between the best processors:

2P Top of Stack GA Offers
AnandTech EPYC
7001
EPYC
7002
EPYC
7003
Intel
Xeon
Processor 7601 7742 7763 6258R
uArch Zen Zen 2 Zen 3 Cascade
Core 32 64 64 28
TDP 180 W 240 W 280 W 205 W
Basic questions 2200 2250 2450 2700
Turbo frequency 3200 3400 3500 4000
L3 kas 64 MB 256 MB 256 MB 37.5 MB
PCIe 3.0 x128 4.0 x128 4.0 x128 3.0 x48
DDR4 8 x 2666 8 x 3200 8 x 3200 6 x 2933
DRAM Cap 2 TB 4 TB 4 TB 1 TB
Price $ 4200 $ 6950 $ 7890 $ 3950

The new top processor for AMD is the EPYC 7763, a 64-core processor at 280 W TDP that offers a base frequency of 2.45 GHz and a boost frequency of 3.50 GHz. AMD claims that this processor offers + 106% performance in terms of dimensions compared to Intel’s best 2P 28 core processor, the Gold 6258R, and + 17% compared to its previous generation 280 W version the 7H12.

Peak performance versus per core performance

One of AMD’s angles with the new Milan generation will be targeted performance metrics, while the company will not only go for ‘peak numbers’, but also take a broader view of customers who also need high performance per core, especially for software. it is always limited or licensed per-core performance. With that in mind, AMD’s F-Series ‘fast’ processors are now being crystallized.

AMD EPYC 7003 F-Series Processors
Core
Wires
Base
Questions
Turbo
Questions
L3
(MB)
TDP
(W)
Price
F-series
EPYC 75F3 32/64 2950 4000 256
(8 x 32)
280 W $ 4860
EPYC 74F3 24/48 3200 4000 240 W $ 2900
EPYC 73F3 16/32 3500 4000 240 W $ 3521
EPYC 72F3 8/16 3700 4100 180 W $ 2468

These processors have the highest single thread values ​​of anything else in AMD’s offering, along with the full 256 MB L3 case, and in our results we achieve the best scores on a wire basis than anything else we’ve tested for enterprise over x86 . and Arm – more details in the review. The F-series processors will have a slight price above the others.

AMD EPYC: The Tour of Italy

The first generation EPYC was launched in June 2017. At the time, AMD was in fact a phoenix: it had risen from the ashes of its former Opteron business and with a promise to return to high-performance computing with a new processor design philosophy.

At the time, the company’s traditional customer base was initially unconvinced – AMD’s latest foray into the enterprise space with a new generation of paradigm-shifting processor cores, though successful, collapsed because AMD had to stop it from going bankrupt. Opteron customers had no updates in sight at the time, and the willingness to jump on an unfamiliar platform of a company that has stabbed so many in the past was not a positive prospect for many.

At the time, AMD posted a three-year chart outlining the next generations and the path the company would take to overcome the 99% market share in performance and offerings. These are considered high goals, and many have reluctantly agreed to watch others take the risk.


1st generation EPYC launch

Introducing the first generation Naples, it offered some impressive performance figures. It did not quite compete in all areas, and as with any new platform, there were a few issues that started. AMD held the initial cycle with some of its key OEM partners before slowly expanding the ecosystem. Naples was the first platform to offer extensive PCIe 3.0 and lots of memory support, and the platform was initially aimed at storage or PCIe-heavy implementations.


2nd generation EPYC launch

The second generation Rome, launched in August 2019 (+26 months), has created much more fanfare. AMD’s latest Zen 2 core was competitive in the consumer space, and there were a number of major design changes in the SoC layout (such as moving to a NUMA flat design) that encouraged a number of skeptics to launch the platform begin to evaluate. It was the interest that AMD even told us that they had to be selective with which OEM platforms they were going to help with before the official launch. Rome’s performance was good and he achieved some high-profile supercomputer victories, but more importantly, perhaps it showed that AMD was able to execute the roadmap in June 2017.

That flat SoC architecture, along with the updated Zen 2 processor core (which actually borrowed elements from Zen 3) and PCIe 4.0, enabled AMD to compete on performance as well as simple IO, and AMD’s OEM partners advertised Rome processors throughout. as computer platforms, which often replace two Intel 28-core processors for one AMD 64-core processor that also has higher memory support and more PCIe offerings. It also allows for computer density, and AMD was at a point where it could also help to achieve software optimizations for its platform, and extract performance, but also to move to the edge cases for which its competitors were highly optimized. All of the major hyperscales also rated and deployed AMD-based offerings for their customers as well as internally. AMD’s sticker of approval was there.


3rd generation EPYC CPU

And today AMD continues the tour through Italy with a trip to Milan, about 19 months to Rome. The underlying SoC layout is the same as Rome, but we have better performance on the table, with added security and more configuration options. The hyperscales have been getting the final hardware for their deployment for six months now, and AMD is now able to enable more OEM platforms during launch. Milan is compatible with Rome, which certainly helps, but with Milan covering more optimization points, AMD believes it’s in a better position to target more of the market with high-performance processors and high-performance processors than ever before. previously.

AMD sees the launch of Milan as the third step in the roadmap shown in June 2017, and the affirmation of the ability to perform reliably for its customers, but also offers better performance improvements than the industry for its customers.

The next stop on the tour through Italy is Genoa, which will use AMD’s upcoming Zen 4 microarchitecture. AMD also said that Zen 5 is in the offing.

Competition

AMD launches this new generation of Milan processors, about 19 months after the launch of Rome. At that time, we saw the launch of Amazon Graviton2 and Ampere Altra, built on Arm’s Neoverse N1 core family.

Milan Top-of-Stack Competition
AnandTech EPYC
7003
Amazon
Graviton2
Ampere
Other
Intel
Xeon
Platform Milan Graviton2 QuickSilver Cascade
Processor 7763 Graviton2 V80-33 6258R
uArch Zen 3 N1 N1 Cascade
Core 64 64 80 28
TDP 280 W ? 250 W 205 W
Basic questions 2450 2500 3300 2700
Turbo frequency 3500 2500 3300 4000
L3 kas 256 MB 32 MB 32 MB 37.5 MB
PCIe 4.0 x128 ? 4.0 x128 3.0 x48
DDR4 8 x 3200 8 x 3200 8 x 3200 6 x 2933
DRAM Cap 4 TB ? 4 TB 1 TB
Price $ 7890 Nvt $ 4050 $ 3950

From Intel, the company has split its efforts between large sockets and small socket configurations. For large pedestals (4+) there is Cooper Lake, a Skylake-derived tool for exceptional customers. For smaller socket configurations (1-2), Intel will launch its 10 nm Ice Cream portfolio at some point this year, but it still remains silent on exact dates. Therefore, we only need to compare Milan with Intel’s Cascade Lake Xeon Scalable platform, which was the same platform with which we compared Rome.

Interesting times for sure.

This review

For this review, AMD gave us remote access to several identical servers with different processor configurations. We focus on the top-of-the-line EPYC 7763, a 280 W 64-core processor, the EPYC 7713, a 225 W 64-core processor, and the EPYC 7F53, a 280 W 32-core processor that designed as the Halo Milan processor for core performance.

On the next page, we go through AMD’s processor stack of Milan, and its comparison with Rome, as well as the comparison with current Intel offerings. We then go through our testing systems, discussions about our SoC structure tests (cache, core-to-core, bandwidth), processor power and thereafter to our full benchmarks.

  1. This page, the overview
  2. Deals for processors in Milan
  3. Test bed setups, compilation options
  4. Topology, memory subsystem and latency
  5. Processing power: Core vs IO
  6. SPES: Multi-thread performance
  7. SPECIES: Single wire performance
  8. SPEC: Per core performance wins for 75F3
  9. SPECjbb MultiJVM: Java Performance
  10. Criteria for composition and calculation
  11. Conclusions and concluding remarks

Access these pages is by clicking on the links or by using the menu below.

Source