Xeon platinum 9282 benchmark: Intel Xeon Platinum 9282 Benchmarks Compared To EPYC 7742 Form Intel

Update to the Intel Xeon Platinum 9282 GROMACS Benchmarks Piece

Intel Server System 9200WK 2U Air Cooled Node CPU Heatsink Cover

I wanted to circle back on this one since apparently a lot of people read the Intel Performance Strategy Team Publishing Intentionally Misleading Benchmarks. As one might imagine, I have been chatting with Intel folks since even before that article came out yesterday. I think we have a fairly good idea of what happened so I wanted to lay this out.

Getting to the Bottom of Intel’s GROMACS Results

Over the last day or so since the piece went live, Intel has re-tested using GROMACS 2019.4. They showed me the results which were largely the same as they presented publicly yesterday.

There were a few things that we addressed in the discussion. The first we are going to use this chart from the discussion yesterday:

Intel Xeon Platinum 9282 GROMACS Config Compared To EPYC 7742 Form Intel

There are two points that needed clarification:

First, the “AVX2 Build” meant that Intel specifically enabled the AVX2 data path and allege it is working properly in their GROMACS 2019. 3 run. One of the big changes in GROMACS 2019.4, was that the tool automatically optimizes for this.
Second, the disclosure #31 threads per core = 1 on AMD, Intel maintains that it was a typo and that the tests were actually done with two threads per core.

During the discussions, I mentioned that this seems to be a process breakdown. I think we agreed that a better proofing process and disclosures need to be part of Intel’s go-forward process. I also offered to have them bounce this type of content off me before they release it to help prevent this type of error to creep into the public domain.

I also confirmed that Intel did the testing using the AMD EPYC 7742 default TDP of 225W, not a 240W cTDP that the chips are capable of. That cTDP has some variability in how much it helps different pieces of silicon, but it can get single-digit performance gains on tasks like this. Frankly, I still think a 240W cTDP is a better proof point if they could not get something like the EPYC 7h22 if one is being completely fair and best optimizing the AMD system for comparison. One could argue that AMD would similarly not extend that courtesy to an Intel part, but this needs to be documented that the feature was available and not being used.

Publishing incorrect information is misleading, but Intel maintains that given its test setup above, it still was directionally correct on its numbers. We do not have a Platinum 9282 system in the lab for our team to run comparisons on, so at some point, we look to Intel for those numbers and it is up to us and our readers to decide whether the optimizations Intel presented are acceptable.

Broader Perspective

I wanted to add a few points to the discussion:

Many of our readers rightly noted that Intel is using its software stack, and it is comparing a 400W TDP CPU versus the default 225W configurable TDP of a competitor. The Platinum 9282 platform is only available from Intel and lacks some substantial features that are also important to the discussion such as PCIe Gen4. At the same time, Intel disclosed what they did, albeit with an error in the initial draft.
While the criticism of the comparison between a mainstream CPU versus a niche part is valid, as it how the test was conducted, it seems like Intel did a better job than their publishing process indicated.
We need to remember this is a marketing exercise. As such, we also need to expect that Intel is going to try presenting its best case.
Intel needs a better process to accurately communicate what they are showing. This is a case where that process broke down. Without a solid process to accurately communicate what is being shown, we end up in situations like these where Intel is trying to show a competitive landscape but end up showing something else. Even if one item was just a typo, which everyone makes (I will be the first to admit I frequently do), it changes what is being presented to the public. A process to maintain accuracy in communication is key to ensuring that content is not misleading and also ensuring reputations are not damaged.
I pointed out that the disclosures are not easy to find given their citations. I was told that the company has a project to improve that.
I actually know the Intel performance teams fairly well, and they are generally nice folks. I know there are some comments out there that are extremely disparaging citing specific individuals that we never mentioned on STH. It is worthwhile to remember that these folks have families and are often not accustomed to the spotlight.

Final Words

The best comparison for the AMD EPYC 7742 to Intel’s lineup is the Platinum 8280. Comparing the EPYC 7742 to the Platinum 9282 is probably not the comparison that is the most useful, however, it is a story that Intel is pushing. Intel is being very aggressive in providing support for the Xeon Platinum 9282 and there are going to be cases where it is able to beat a lower power chip. That is Intel’s story to tell whether many would agree with that comparison or not.

At the end of the day, to increase accuracy, Intel needs to implement a better publishing process to bridge the gap between the work that their benchmarking teams do, and what they show publicly. The company can choose what data to present and how to present it. Ensuring accurate communication of what was, and what was not done in the comparison is important.

From our perspective, I am more than happy to listen to concerns and help find these errors so that accurate information can be presented.

Xeon Platinum 9282 — Technical City

Intel
Xeon Platinum 9282

Buy

Interface

Core clock speed

Max video memory

Memory type

Memory clock speed

Maximum resolution

Summary

Intel started Intel Xeon Platinum 9282 sales 2 April 2019. This is Cascade Lake-AP architecture desktop processor primarily aimed at professional systems. It has 56 cores and 112 threads, and is based on 14 nm manufacturing technology, with a maximum frequency of 3800 MHz and a locked multiplier.

Compatibility-wise, this is
Intel BGA5903
processor with a TDP of 400 Watt. It supports DDR4-2933 memory.

We have no data on Xeon Platinum 9282 benchmark results.

General info

Xeon Platinum 9282 processor market type (desktop or notebook), architecture, sales start time and pricing.

Place in performance rating	not rated
Market segment	Server
Series	Intel Xeon Platinum
Architecture codename	Cascade Lake-AP (2019)
Release date	2 April 2019 (3 years ago)
Current price	$14999

Technical specs

Basic microprocessor parameters such as number of cores, number of threads, base frequency and turbo boost clock, lithography, cache size and multiplier lock state. These parameters can generally indicate CPU performance, but to be more precise you have to review its test results.

Physical cores	56 (Hexapentaconta-Core)
Threads	112
Base clock speed	2.6 GHz	of 4.7 (FX-9590)
Boost clock speed	3.8 GHz	of 5.8 (Core i9-13900K)
Bus support	4 × 8 GT/s
L1 cache	64K (per core)	of 1536 (EPYC Embedded 3401)
L2 cache	1 MB (per core)	of 12 (Core 2 Quad Q9550)
L3 cache	77 MB (shared)	of 32 (Ryzen Threadripper 1998)
Chip lithography	14 nm	of 5 (Apple M1)
Number of transistors	8,000 million	of 57000 (Apple M1 Max)
64 bit support	+
Windows 11 compatibility	+
Unlocked multiplier	—

Compatibility

Information on Xeon Platinum 9282 compatibility with other computer components and devices: motherboard (look for socket type), power supply unit (look for power consumption) etc. Useful when planning a future computer configuration or upgrading an existing one.

Note that power consumption of some processors can well exceed their nominal TDP, even without overclocking. Some can even double their declared thermals given that the motherboard allows to tune the CPU power parameters.

Number of CPUs in a configuration	2	of 8 (Opteron 842)
Socket	Intel BGA5903
Thermal design power (TDP)	400 Watt

Technologies and extensions

Technological capabilities and additional instructions supported by Xeon Platinum 9282. You’ll probably need this information if you require some particular technology.

Instruction set extensions	Intel® AVX-512
AES-NI	+
FMA	+
AVX	+
PowerNow	+
vPro	+
Enhanced SpeedStep (EIST)	+
Speed Shift	+
Turbo Boost Technology	2. 0
Hyper-Threading Technology	+
TSX	+
TSX	+
SIPP	+

Security technologies

Processor technologies aimed at improving security, for example, by protecting against hacks.

Virtualization technologies

Supported virtual machine optimization technologies. Some are specific to Intel only, some to AMD.

AMD-V	+
VT-d	+
VT-x	+

Memory specs

Types, maximum amount and channel number of RAM supported by Xeon Platinum 9282’s memory controller. Depending on the motherboard, higher memory frequency may be supported.

Supported memory types	DDR4-2933	of 5200 (Ryzen 5 7600X)
Maximum memory size	2 TiB	of 786 (Xeon E5-2670 v3)
Max memory channels	12
Maximum memory bandwidth	281. 6 GB/s

Benchmark performance

Single-core and multi-core benchmark results of Xeon Platinum 9282. Overall benchmark performance is measured in points in 0-100 range, higher is better.

We have no data on Xeon Platinum 9282 benchmark results.

Similar processors

Here is our recommendation of several processors that are more or less close in performance to the one reviewed.

Recommended graphics cards

These graphics cards are most commonly used with Xeon Platinum 9282 according to our statistics.

GeForce RTX
3090

15.4%

GeForce RTX
3090 Ti

6.5%

GeForce RTX
3080 Ti

5. 7%

GeForce GTX
1050 Ti

5.5%

TITAN
RTX

5.3%

GeForce GTX
TITAN X

3.9%

GeForce RTX
3090 SUPER

2.4%

GeForce GTX
750 Ti

UHD
Graphics 630

Radeon RX
6900 XT

1.8%

User rating

Here is the rating given to the reviewed processor by our users. Let others know your opinion by rating it yourself.

Questions and comments

Here you can ask a question about Xeon Platinum 9282, agree or disagree with our judgements, or report an error or mismatch.

Please enable JavaScript to view the comments powered by Disqus.

AMD Epyc 7742 vs Intel Xeon Platinum 9282: What is the difference?

92points

AMD Epyc 7742

79points

Intel Xeon Platinum 9282

61 facts in comparison

AMD Epyc 7742

Intel Xeon Platinum 9282

Why is AMD Epyc 7742 better than Intel Xeon Platinum 9282?

16 more CPU threads?
128vs112
175W lower TDP?
225Wvs400W
1 newer version of PCI Express (PCIe)?
4vs3
179MB bigger L3 cache?
256MBvs77MB
Supports ECC memory?
512KB bigger L1 cache?
4096KBvs3584KB
2000GB larger maximum memory amount?
4000GBvs2000GB
267MHz higher ram speed?
3200MHzvs2933MHz

Why is Intel Xeon Platinum 9282 better than AMD Epyc 7742?

71. 56GB/s more memory bandwidth?
262.26GB/svs190.7GB/s
4 more memory channels?
12vs8
0.4GHz higher turbo clock speed?
3.8GHzvs3.4GHz
24MB bigger L2 cache?
56MBvs32MB
0.5MB/core more L2 cache per core?
1MB/corevs0.5MB/core

Which are the most popular comparisons?

AMD Epyc 7742

AMD Epyc 7h22

Intel Xeon Platinum 9282

AMD Epyc 7763

AMD Epyc 7742

AMD Epyc 7702

Intel Xeon Platinum 9282

AMD Epyc 7h22

AMD Epyc 7742

AMD Epyc 7763

Intel Xeon Platinum 9282

AMD Epyc 7773X

AMD Epyc 7742

AMD Epyc 7713

Intel Xeon Platinum 9282

Intel Xeon Platinum 8168

AMD Epyc 7742

AMD Epyc 7282

Intel Xeon Platinum 9282

AMD Epyc 7702

AMD Epyc 7742

AMD Epyc 7542

Intel Xeon Platinum 9282

Intel Xeon Platinum 9242

AMD Epyc 7742

AMD Epyc 7642

Intel Xeon Platinum 9282

AMD Epyc 7513

AMD Epyc 7742

AMD Epyc 7773X

Intel Xeon Platinum 9282

Intel Xeon Phi 7230

AMD Epyc 7742

AMD Epyc 7302

Intel Xeon Platinum 9282

Intel Xeon Gold 6230R

Price comparison

Cheap alternatives

User reviews

Performance

1. CPU speed

64 x 2.25GHz

56 x 2.6GHz

The CPU speed indicates how many processing cycles per second can be executed by a CPU, considering all of its cores (processing units). It is calculated by adding the clock rates of each core or, in the case of multi-core processors employing different microarchitectures, of each group of cores.

2.CPU threads

More threads result in faster performance and better multitasking.

3.turbo clock speed

3.4GHz

3.8GHz

When the CPU is running below its limitations, it can boost to a higher clock speed in order to give increased performance.

4.L3 cache

A larger L3 cache results in faster CPU and system-wide performance.

5.L1 cache

4096KB

3584KB

A larger L1 cache results in faster CPU and system-wide performance.

6.L2 cache

A larger L2 cache results in faster CPU and system-wide performance.

7.L2 core

0.5MB/core

1MB/core

More data can be stored in the L2 cache for access by each core of the CPU.

8.clock multiplier

Unknown. Help us by suggesting a value. (AMD Epyc 7742)

Unknown. Help us by suggesting a value. (Intel Xeon Platinum 9282)

The clock multiplier controls the speed of the CPU.

9.Has an unlocked multiplier

✖AMD Epyc 7742

✖Intel Xeon Platinum 9282

Some processors come with an unlocked multiplier which makes them easy to overclock, allowing you to gain increased performance in games and other apps.

Memory

1.Supports ECC memory

✔AMD Epyc 7742

✖Intel Xeon Platinum 9282

Error-correcting code memory can detect and correct data corruption. It is used when is it essential to avoid corruption, such as scientific computing or when running a server.

2. maximum memory bandwidth

190.7GB/s

262.26GB/s

This is the maximum rate that data can be read from or stored into memory.

3.DDR memory version

DDR (Double Data Rate) memory is the most common type of RAM. Newer versions of DDR memory support higher maximum speeds and are more energy-efficient.

4.maximum memory amount

4000GB

2000GB

The maximum amount of memory (RAM) supported.

5.RAM speed

3200MHz

2933MHz

It can support faster memory, which will give quicker system performance.

6.memory channels

More memory channels increases the speed of data transfer between the memory and the CPU.

7.bus speed

Unknown. Help us by suggesting a value. (AMD Epyc 7742)

Unknown. Help us by suggesting a value. (Intel Xeon Platinum 9282)

The bus is responsible for transferring data between different components of a computer or device.

8.bus transfer rate

Unknown. Help us by suggesting a value. (AMD Epyc 7742)

Unknown. Help us by suggesting a value. (Intel Xeon Platinum 9282)

The bus is responsible for transferring data between different components of a computer or device.

9.eMMC version

Unknown. Help us by suggesting a value. (AMD Epyc 7742)

Unknown. Help us by suggesting a value. (Intel Xeon Platinum 9282)

A higher version of eMMC allows faster memory interfaces, having a positive effect on the performance of a device. For example, when transferring files from your computer to the internal storage over USB.

Features

1.uses multithreading

✔AMD Epyc 7742

✔Intel Xeon Platinum 9282

Multithreading technology (such as Intel’s Hyperthreading or AMD’s Simultaneous Multithreading) provides increased performance by splitting each of the processor’s physical cores into virtual cores, also known as threads. This way, each core can run two instruction streams at once.

2.SSE version

SSE is used to speed up multimedia tasks such as editing an image or adjusting audio volume. Each new version contains new instructions and improvements.

3.Has AES

✔AMD Epyc 7742

✖Intel Xeon Platinum 9282

AES is used to speed up encryption and decryption.

4.Has AVX

✔AMD Epyc 7742

✔Intel Xeon Platinum 9282

AVX is used to help speed up calculations in multimedia, scientific and financial apps, as well as improving Linux RAID software performance.

5.bits executed at a time

Unknown. Help us by suggesting a value. (AMD Epyc 7742)

Unknown. Help us by suggesting a value. (Intel Xeon Platinum 9282)

NEON provides acceleration for media processing, such as listening to MP3s.

6.front-end width

Unknown. Help us by suggesting a value. (AMD Epyc 7742)

Unknown. Help us by suggesting a value. (Intel Xeon Platinum 9282)

The CPU can decode more instructions per clock (IPC), meaning that the CPU performs better

7.Has F16C

✔AMD Epyc 7742

✔Intel Xeon Platinum 9282

F16C is used to speed up tasks such as adjusting the contrast of an image or adjusting volume.

8.Has FMA3

✔AMD Epyc 7742

✔Intel Xeon Platinum 9282

FMA3 is used to speed up tasks such as adjusting the contrast of an image or adjusting volume.

9.Has FMA4

✖AMD Epyc 7742

✖Intel Xeon Platinum 9282

FMA4 is used to speed up tasks such as adjusting the contrast of an image or adjusting volume.

Benchmarks

1.Cinebench R20 (multi) result

Unknown. Help us by suggesting a value. (Intel Xeon Platinum 9282)

Cinebench R20 is a benchmark tool that measures a CPU’s multi-core performance by rendering a 3D scene.

2.Cinebench R20 (single) result

Unknown. Help us by suggesting a value. (Intel Xeon Platinum 9282)

Cinebench R20 is a benchmark tool that measures a CPU’s single-core performance by rendering a 3D scene.

3.Geekbench 5 result (multi)

Unknown. Help us by suggesting a value. (Intel Xeon Platinum 9282)

Geekbench 5 is a cross-platform benchmark that measures a processor’s multi-core performance. (Source: Primate Labs, 2022)

4.Geekbench 5 result (single)

Unknown. Help us by suggesting a value. (Intel Xeon Platinum 9282)

Geekbench 5 is a cross-platform benchmark that measures a processor’s single-core performance. (Source: Primate Labs, 2022)

5.PassMark result

Unknown. Help us by suggesting a value. (Intel Xeon Platinum 9282)

This benchmark measures the performance of the CPU using multiple threads.

6.PassMark result (single)

Unknown. Help us by suggesting a value. (Intel Xeon Platinum 9282)

This benchmark measures the performance of the CPU using a single thread.

7.performance per watt

Unknown. Help us by suggesting a value. (Intel Xeon Platinum 9282)

This means the CPU is more efficient, giving a greater amount of performance for each watt of power used.

8.Blender (bmw27) result

33.9seconds

Unknown. Help us by suggesting a value. (Intel Xeon Platinum 9282)

The Blender (bmw27) benchmark measures the performance of a processor by rendering a 3D scene. More powerful processors can render the scene in less time.

9.Blender (classroom) result

95.8seconds

Unknown. Help us by suggesting a value. (Intel Xeon Platinum 9282)

The Blender (classroom) benchmark measures the performance of a processor by rendering a 3D scene. More powerful processors can render the scene in less time.

Price comparison

Cancel

Which are the best Enterprise CPUs?

Intel Xeon Platinum 9282 | 61 factors

79points

Intel Xeon Platinum 9282

Why is Intel Xeon Platinum 9282 better than others?

Processor thread?
112 vs 24.35
CPU speed?
56 x 2.6GHz vs 31.14GHz
Maximum memory bandwidth?
262.26GB/s vs 86.5GB/s
Size of semiconductors?
14nm vs 15.82nm
L3 cache?
77MB vs 25.7MB
Memory channels?
12 vs 4.5
Turbo clock speed?
3.8GHz vs 3.43GHz
L1 cache?
3584KB vs 805.41KB

Which comparisons are the most popular?

Intel Xeon Platinum 9282

AMD Epyc 7742

Intel Xeon Platinum 9282

AMD Epyc 7763

Intel Xeon Platinum 9282

AMD Epyc 7h22

Intel Xeon Platinum 9282

AMD Epyc 7773X

Intel Xeon Platinum 9282

Intel Xeon Platinum 8168

Intel Xeon Platinum 9282

0004 Intel Xeon Platinum 9242

Intel Xeon Platinum 9282

AMD Epyc 7513

Intel Xeon Platinum 9282

Intel Xeon Phi 7230

Intel Xeon Platinum 9282

Intel Xeon Gold 6230R

Price Match

User Reviews

Performance

1. CPU Speed

56 x 2.6GHz

CPU Speed indicates how many processing cycles per second a processor can perform, given all its cores (processors). It is calculated by adding the clock speeds of each core or, in the case of multi-core processors, each group of cores.

2nd processor thread

More threads result in better performance and better multitasking.

3.speed turbo clock

3.8GHz

When the processor is running below its limits, it can jump to a higher clock speed to increase performance.

4.L3 cache

More L3 scratchpad memory results in faster results in CPU and system performance tuning.

5.L1 cache

3584KB

More L1 cache results in faster results in CPU and system performance tuning.

6.L2 Cache

More L2 scratchpad memory results in faster results in CPU and system performance tuning.

7.core L2

1MB/core

More data can be stored in L2 scratchpad for access by each processor core.

8.hour multiplier

Unknown. Help us offer a price.

The hour multiplier controls the speed of the processor.

9.Unlocked multiplier

✖Intel Xeon Platinum 9282

Some processors come with an unlocked multiplier and are easier to overclock, allowing for better performance in games and other applications.

Memory

1. Supports memory recovery code

✖Intel Xeon Platinum 9282

Memory troubleshooting code can detect and repair data corruption. It is used when necessary to avoid distortion, such as in scientific computing or when starting a server.

2.max memory bandwidth

262.26GB/s

This is the maximum rate at which data can be read from or stored in memory.

3. DDR version

DDR (Dynamic Dynamic Random Access Memory Double Data Rate) is the most common type of RAM. New versions of DDR memory support higher maximum speeds and are more energy efficient.

4.Max memory

2000GB

Maximum memory (RAM).

5.RAM speed

2933MHz

Can support faster memory which speeds up system performance.

6.Memory channels

More memory channels increase the speed of data transfer between memory and processor.

7.bus frequency

Unknown. Help us offer a price.

The bus is responsible for transferring data between different components of a computer or device

8. bus baud rate

Unknown. Help us offer a price.

The bus is responsible for transferring data between various components of a computer or device.

9.eMMC version

Unknown. Help us offer a price.

A newer version of eMMC — built-in flash memory card — speeds up the memory interface, has a positive effect on device performance, for example, when transferring files from a computer to internal memory via USB.

Features

1.uses multithreading

✔Intel Xeon Platinum 9282

like streams. Thus, each core can run two instruction streams at the same time.

SSE version 2.

SSE is used to speed up multimedia tasks such as editing images or adjusting audio volume. Each new version contains new instructions and improvements.

3. Has AES

✖Intel Xeon Platinum 9282

AES is used to speed up encryption and decryption.

4. Has AVX

✔Intel Xeon Platinum 9282

AVX is used to help speed up calculations in multimedia, scientific and financial applications, and to improve the performance of the Linux RAID program.

5 bits transmitted at the same time

Unknown. Help us offer a price.

NEON provides faster media processing such as MP3 listening.

6.interface width

Unknown. Help us offer a price.

The processor can decode more instructions per clock (IPC), which means that the processor performs better

7. Has F16C

✔Intel Xeon Platinum 9282

F16C is used to speed up tasks such as adjusting image contrast or adjusting volume .

8. Has FMA3

✔Intel Xeon Platinum 9282

FMA3 is used to speed up tasks such as image contrast adjustment or volume control.

9. Has FMA4

✖Intel Xeon Platinum 9282

FMA4 is used to speed up tasks such as image contrast adjustment or volume control.

Geotagging

1. Cinebench R20 result (multi-core)

Unknown. Help us offer a price.

Cinebench R20 is a test that measures the performance of a multi-core processor by rendering a 3D scene.

2.Cinebench R20 result (single core)

Unknown. Help us offer a price.

Cinebench R20 is a test to evaluate the performance of a single core processor when rendering a 3D scene.

3.Geekbench 5 result (multi-core)

Unknown. Help us offer a price.

Geekbench 5 is a cross-platform benchmark that measures the performance of a multi-core processor. (Source: Primate Labs,2022)

4.Geekbench 5 result (single core)

Unknown. Help us offer a price.

Geekbench 5 is a cross-platform benchmark that measures the single-core performance of a processor. (Source: Primate Labs, 2022)

5. PassMark result

Unknown. Help us offer a price.

This test measures processor performance using multithreading.

6. PassMark result (single)

Unknown. Help us offer a price.

This test measures processor performance using a thread of execution.

7. performance per watt

Unknown. Help us offer a price.

This means that the processor is more efficient, giving more performance per watt of power used.

8. Blender test result (bmw27)

Unknown. Help us offer a price.

Blender benchmark (bmw27) measures CPU performance by rendering a 3D scene. More powerful processors can render a scene in a shorter time.

9. Blender result (classroom)

Unknown. Help us offer a price.

The Blender (classroom) benchmark measures CPU performance by rendering a 3D scene. More powerful processors can render a scene in a shorter time.

Price Match

Cancel

Which Enterprise Processors are Best?

Intel’s New HPC Benchmark Results Are Wrong (Update)

Intel’s marketing department compared the performance of its own high-end Xeon processors to AMD’s. The reason for this comparison was the question of the need for more cores. Modern AMD EPYC processors provide up to 64 cores, which in a two-socket system gives 128 cores and 256 threads. Intel doubted that such a large number of cores always gives advantages, reinforcing the thesis with its own tests.

Intel believes there are other factors to consider here. Including the performance of individual cores, the ability to optimize software for hardware resources, memory bandwidth and the ability to scale such systems in a cluster (more than two sockets).

In particular, two Xeon Platinum 9282 and two EPYC 7742 processors were compared. Two Xeon processors belong to the 9200 line, they have an MCM design, that is, two Xeon crystals are located in one package. As a result, for each Xeon CPU we get 2 x 28 = 56 cores, 12-channel memory interface running at 407 GB/s bandwidth.

2nd generation Xeon Scalable processor comparison
Model	Cores/Threads	Base frequency / Turbo	L3 cache	TDP	Memory Bandwidth	Price
Xeon Platinum 9282	56 / 112	2.6 / 3.8 GHz	77MB	400 W	407 GB/s	—
EPYC 7742	64 / 128	2.25 / 3.4 GHz	256MB	225 W	204.8 GB/s	€7.500

Two EPYC 7742s are equipped with 64 cores each and work with an eight-channel interface. We get a bandwidth of 204. 8 GB / s per socket, which is significantly less, which will also be seen from the tests. There are also significant differences in the thermal package. In case of EPYC 7742 TDP is 225W, in Xeon Platinum 9282 we get 400 watts.

EPYC 7742 processors can be purchased at retail and installed on a compatible motherboard, but finding the Xeon Platinum 9282 is much more difficult, if not impossible. Intel only sells processors in off-the-shelf server systems such as the S9200WK. The processor is not installed in the socket, but soldered to the motherboard.

Tests show that two Xeon Platinum 9282s are 8-84% faster than two EPYC 7742s. Depending on the scenario, the difference is not that big. At the same time, it does not interfere with taking into account the price and energy consumption.

AVX-512 extensions play an important role. If an application knows how to use them, then it gets a significant performance boost. Modern Intel Xeon processors support AVX-512, which cannot be said about EPYC processors.

The table shows the tests used. Moreover, it is noted whether the application receives an increase from the presence of AVX-512. It also indicates whether the application is limited by compute performance or memory bandwidth.

Intel confidently claims leadership in HPC. The performance is significantly higher, which is due to the high performance of individual cores with and without AVX-512, as well as a wider memory interface and greater memory bandwidth.

The prices of the Xeon Platinum line are unknown to us. Intel here refers to partners Atos, HPE/Cray, Lenovo, Inspur, Sugon, h4C, and Penguin Computing. But we can assume that the Xeon Platinum 9282 costs significantly more than €7.500 (530.000 ₽) of the EPYC 7742 processor. In any case, the total cost of ownership TCO should be taken into account. The computational density of Intel is higher, the cost of interconnection between clusters is lower. The result is:

«More processor cores do not always translate to higher performance, and nor do more processor cores always translate to better TCO.» Translation: More processor cores don’t always translate into better performance, not to mention that more cores don’t always lead to better TCO.

At the upcoming Supercomputing 19 conference in mid-November, there will be new announcements about the Xeon Platinum 9200 line.

Unoptimized benchmarks and controversial hardware solutions

However, the experts took a closer look at the composition of the tests and notes published by Intel. In particular, an old version of GROMACS, a molecular dynamics simulation software package, was used. Intel chose the GROMACS 2019.3 version, but GROMACS 2019.4 has been available for a relatively long time, which is optimized for the new EPYC processors. The latest version uses the AVX2 function blocks of the Zen 2 architecture. The GROMACS developer mentioned the following note in the Release Notes of the latest version.

«The AMD Zen 2 architecture is now detected as different from Zen 1 and uses 256-bit wide AVX2 SIMD instructions (GMX_SIMD=AVX2_256) by default. Also the non-bonded kernel parameters have been tuned for Zen 2. This has a significant impact on performance.»

Of course, it would be interesting to look at the comparative results of the two EPYC 7742s in an optimized version of the application. Our ServeTheHome colleagues did a deeper performance analysis, but they didn’t have two Xeon Platinum 9 processors282 to run your own tests.

Also, Intel seems to have enabled only two threads per core in the case of a system with two Xeon Platinum 9282s, but only one thread per core for two EPYC 7742s. : Intel Compiler 2019u4, Intel® Math Kernel Library (Intel MKL) 2019u4, Intel MPI 2019u4, AVX-512 build, BIOS: HT ON, Turbo OFF, SNC OFF, 2 threads per core;

AMD EPYC 7742: Intel Compiler 2019u4, Intel MKL 2019u4, Intel MPI 2019u4, AVX2 build, BIOS: SMT ON, Boost ON, NPS 4, 1 threads per core.

That is, we are comparing 224 streams against 128 streams, which does not fully exploit the potential of the 2S EPYC system with a possible 256 streams. Other settings in the GROMACS package are also controversial, as ServeTheHome points out. This alone already puts Intel’s results in a bad light.

Of course, AMD is also not without sin. At a preliminary demonstration of the new EPYC processors, the results of NAMD (Nanoscale Molecular Dynamics Program) tests were published. And there, in the case of Intel processors, a version was used that does not use the AVX 512 instruction sets, that is, it is not optimized for Xeon processors.

On the one hand, it is correct to use the same software packages and tests for both systems. On the other hand, server applications are initially developed with platform optimization in mind. HPC software developers are constantly working to maximize the full potential of hardware platforms. This is true for both Intel and AMD processors. Accordingly, it is correct to use optimized software packages for each platform. Moreover, Intel is unlikely to receive any disadvantages due to the transition to the new version of GROMACS 2019..4 because it also supports the AVX-512 instruction set.

Finally, Intel chose the Supermicro AS-2023-TR4 server with HD11DSU-iN motherboard for two EPYC processors. It was used in version 2.0 with support for the new EPYC processors, but the board still does not support PCI Express 4.0. We didn’t use any computing accelerators or drives that would benefit from PCI Express 4.0 support, but you should still rely on a modern hardware platform to avoid attacks from critics from this side.

The question then becomes, is Intel deliberately portraying a competitor’s product in this light, or are we just dealing with a bug? In the case of the GROMACS software package, the error is unlikely to occur, since the latest version was submitted on October 2, 2019. Even if the test results were obtained earlier, they should have been redone. There is a feeling that Intel decided to use the old version of the package on purpose.

We’ve previously published a similar study from Intel on how games don’t need more than eight cores. As you can see, the products of the Intel marketing department should be approached scrupulously, analyzing every detail. There are many questions.

Update:

Intel has corrected the results and provided some clarifications.

The GROMACS tests were repeated in version 2019.4, with no significant change in the results. However, the number of threads per core for EPYC processors in the test is a typo. The Xeon and EPYC processors were tested with two threads per core.

Intel Xeon Platinum 9282 — processor overview. Tests and Specs

The Intel Xeon Platinum 9282 processor is based on a 14 nm process and Cascade Lake architecture. The base clock is 2.60 GHz and the maximum turbo clock is 3.80 GHz. Intel Xeon Platinum 9282 contains 56 processor cores. To make the right choice for upgrading your computer, check out the detailed specifications and test results. Please check your motherboard socket compatibility before choosing.

2.5
From 2
Hitesti Score

TOP 10 processors

General information

Intel Xeon Platinum 9282 processor base and maximum clock speed, number of processor cores and threads. The more the better (marked in green).

CPU cores, base and turbo speeds

The overall performance of a processor can be easily determined based on its number of cores and threads, as well as its standard and turbo clock speeds. The more GHz, cores and L2-L3 cache a processor has, the better. Please note that high specs require a powerful cooling system and a quality chipset (check the VRM on the motherboard).

Clock frequency:	2.60 GHz	Number of cores:	56
Turbo (1 core):	3. 80 GHz	Number of threads:	112
Hyper trading:	Yes	Acceleration:	No.
Architecture:	normal	A core:	0x
B core:	0x	Turbo (56 Cores):	2. 80 GHz

Internal graphics

Some manufacturers (most often Intel) supplement processors with graphics chips, such a solution is especially popular in laptops, but is ineffective in workstations and gaming stations. The higher the clock speed of the video card and the more memory on board, the better.

GPU (Turbo):

No turbo

Maximum memory:

—

Hardware codec support

Here we are dealing with specifications that are used by some processor manufacturers. These figures are mostly technical and can be disregarded for comparative purposes.

h364:		No
JPEG:		No.
VP8:		No
VP9:		No.
VC-1:		No
AVC:		No.
h365 / HEVC (8 bit):		No
h365 / HEVC (10 bit):		No.
AV1:		No

RAM and PCIe

These are the memory standards supported by processors. The higher the standard and clock frequency with the amount of RAM, the better the performance and speed of the processor. It is also worth considering the possibility of memory operation in multichannel mode.

Memory type:	DDR4-2933	Memory channels:	12
ECC:	Yes	PCIe version:	3. 0
PCIe lanes:	48

Encryption

Support for data encryption

AES-NI:

Yes

Thermal management and TDP

Maximum temperature:	—	Max TDP:	—
TDP down:	—	TDP (PL1):	400W
TDP (PL2):	—

Technical details

These are key parameters that will help you determine which processor is best. Pay special attention to the release date, the technological aspects of the manufacturing process (measured in nanometers) and the third level (L3) cache.

L3-Cache:	77.00 MB	Technology:	14 nm
Architecture:	Cascade Lake	Virtualization:	VT-x, VT-x EPT, VT-d, vPro
Socket:	LGA 3647	Release date:	Q3/2019
Instruction set (ISA):	x86-64 (64 bit)	L2-Cache:	—

Devices compatible with this processor

You probably already know which devices use processors.