AMD launches EPYC 4 processors with 96 cores and 192 threads!

 

Today are the official unveiling of the 4th generation EPYC 4 processors of the popular EPYC series, serving giant data centers. Which was part of the "together we advance data centers" event that mainly talks about the progress in the world of data centers.

 

And since progress is tied to who offers processors designed for this functionality, it is tied to AMD which has always been delivering the best possible to the market with each generation of the mighty EPYC processors, and today it offers the fourth generation of these processors that uses the most powerful AMD architecture today.

 

This architecture is Zen 4, which focuses on delivering the best possible performance with the best possible energy efficiency in order to speed up processing processes within data centers to set the limits for work in such plants that need to balance power with efficiency.

 

The processors we are talking about today have up to 96 cores per chip and 192 processing paths in the same chip. There are no higher numbers in this market now, but what does AMD offer behind these numbers in terms of advantages?

 

Let's start with performance and efficiency!

 

This year, the AMD Epyc platform will offer processors with up to 96 cores, as we mentioned from the Zen 4 cores with a 5 nm manufacturing accuracy. We should also mention that it will of course deal with the latest data transmission lines in the market, and this required it to add the number "5" in all of this.

 

Number 5 appears first in DDR5 memory that operates on 12 random memory channels, and appears again in PCIe 5.0 lines, which will provide us with a maximum of 160 lines in these processors with the use of CXL technology to allow access to storage units remote from the processor itself.

 

All this while maintaining the highest security standards and adding more standards, including the AMD SEV-SNP standard, which prevents modification of data by malicious programs on the device's memory, and which forces the caller to read the encrypted pages from the memory and make sure that they are correct. While reading the last value sent.

 

Do not forget also the encryptions for the memory itself and for the memories that will be accessed through the CXL.

 

All this allowed AMD to deliver 14% more "raw" performance than the previous Zen 3 generation. Also compared to AMD EPYC 3, performance on cloud processing is improved by 107%, processing with heavy jobs on the server is improved by 123% and enterprise applications are improved by up to 94%.

 

GENOA processors

 

These processors will be based on the new Zen 4 architecture with the design of standard cores dedicated to dealing with all kinds of applications if you will. Yes, we will undoubtedly get various improvements in terms of this architecture, such as IPC improvements, which means the number of instructions or commands that the processor is able to process during the pulse The number of cores with this type of processor will reach up to 96 cores based on the new Zen 4 architecture cores.

 

The pallets come with the name EPYC 9004

 

Here, AMD presents us with EPYC 9004 platforms on a chip in the EPYC 4 series. All platforms run on the architecture we are talking about as we mentioned to include 96 cores in 12 core groups with 192 processing paths.

 

These platforms run on up to 400 watts of thermal power. This is at the same time that you are using 12 DDR5 channels at frequencies up to 4800MHz, which is what is actually expected from DDR5 memory.

 

Two memories can be placed in each of these channels with a capacity of up to 12 TB of memory in a device that works with two chips of the same platform.

 

This is while providing the same number of PCIe lanes that we mentioned above, as well as 12 third-generation PCIe lines and 32 lines for SATA. These are of course maximum numbers and not for all combinations.

 

Memory power is something else!

 

The memories we are talking about here can work with 12 channels as we mentioned, and it is supposed to provide us with 460 gigabytes per second of performance if we run all 12 channels on the 4800MHz memory, as we also mentioned above.

 

It will support RDIMMs and 3DS RDIMMs as well, and you can install one memory per channel or two if you like, to make the full 6TB capacity on the single socket of your AMD EPYC 9004 processor.

 

All this with techniques for correcting and reviewing data while working to ensure it works as smoothly as possible, but we also do not forget that DDR5 memory is slower than DDR4 at the moment, but AMD has improved this part in its devices so that performance becomes close - in terms of response times - than last generation as possible.

 

Compared to what Intel provided with the last generation

 

Compared to Intel's 3rd generation Xeon Platinum processors, it is clearly superior to it.

 

We can see up to 96 cores on a single chip compared to 40 from Intel's 3rd generation, which brings the final thread count in a single RAC to 1,790 compared to just 602 with a processor like the Intel Xeon 8380.

 

Efficiency is also more than double that of the AMD EPYC 4, and this is evident in the comparison between the EPYC 9654 and the Xeon Platinum 8380 that delivers 2.6 times more power in terms of performance than the chip as a whole.

 

As for 2P Float processes, the new generation platform offers 2.5 times the performance compared to Intel with the same processor we talked about above, which is useful for handling tough jobs on platforms as mentioned, and even the third generation EPYC 7763 processor was almost stronger than 10%.

 

For computes per second with Java Server, it can do 815,459 processes per second compared to the 286125 provided by Intel, 2.8 times more.

 

To put an end to AMD's third-generation Xeon Platinum, the company pitted a 64-core AMD single-platform machine costing $7,104 against an 80-core dual-core Xeon Platinum 8380 machine for $18,000, which outperformed AMD with 5 percent higher performance. % consumes 25% less energy and costs less than half.

 

In the end, what AMD is offering this time around is…

 

The highest possible performance at the present time is from processors designed for data centers, and this is also with the best possible power consumption to reach the desired result without consuming an extra watt, and this is also while maintaining the confidentiality of data and preventing access to it in any way.