There is literally no competitor - 64 core AMD EPYC 7713 found

[williamcll]

A few days ago, benchmarks for a new AMD server processor has appeared on sisbench and it's already breaking world records:

Quote

Expected to start shipping before the end of this year and with full availability planned for 2021, the next-gen EPYC Milan lineup is already showing up in online benchmark databases. Today, the time has come for the processor known as the EPYC 7713, a 64-core chip that promises to reach an impressive level of performance. The aforementioned processor for the server market is listed in the SiSoftware database with the following details: 64C 128T 2.45GHz, 64x 512kB L2, 8x 32MB L3. What appears to be an early engineering sample — due to the 100-000000344 OPN code — will surely have faster clocks in the market-ready variant. However, the EPYC 7713 is already very fast, with results for the CPU Arithmetic tests that sometimes outperform a quad Intel Xeon Platinum (Caskade Lake-SP) system. When compared to dual EPYC 7742 Rome configurations with clock speeds of over 3.3 GHz, the EPYC 7713 has absolutely no problem to come out on top. Lastly, the top ranking is currently held by four Intel Xeon Platinum 8280L processors, which barely manage a 5 percent speed increase over the Zen 3-based EPYC Milan chip.

Note that the rankings here seems to show a dual core setup on a Gigabyte motherboard rather than a single processor as the source imply.
Source: https://ranker.sisoftware.co.uk/show_run.php?q=c2ffcfe988e9d4ecd9ebdce9dafc8eb383a5c0a598a88efdc0f8&l=en
https://ranker.sisoftware.co.uk/show_device.php?q=c9a598aad2f2b3feba9adfb1d6f6a5c4a9d9b5d0eacafbcbfbd6e6d6e6d6e6d6e6d3e0cdfdc996a597b88abae5ab8dead7facbed9fa292b4dde0d5f39ba693b5cdf0c1e782e7daeaccbf82ba&l=en

https://www.notebookcheck.net/64-core-AMD-Zen-3-based-EPYC-7713-Milan-server-CPU-spotted-online-obliterates-the-Intel-Xeon-Platinum.502439.0.html

Thoughts: It's great. But I don't understand what the reasoning of the number that EPYC uses. Perhaps it means there's an ever better clocked version? Though compared the last gen the jump in performance isn't as big as I thought it would be.

[Ankh Tech]

1 minute ago, williamcll said:

A few days ago, benchmarks for a new AMD server processor has appeared on sisbench and it's already breaking world records:

Source: https://ranker.sisoftware.co.uk/show_run.php?q=c2ffcfe988e9d4ecd9ebdce9dafc8eb383a5c0a598a88efdc0f8&l=en
https://ranker.sisoftware.co.uk/show_device.php?q=c9a598aad2f2b3feba9adfb1d6f6a5c4a9d9b5d0eacafbcbfbd6e6d6e6d6e6d6e6d3e0cdfdc996a597b88abae5ab8dead7facbed9fa292b4dde0d5f39ba693b5cdf0c1e782e7daeaccbf82ba&l=en

https://www.notebookcheck.net/64-core-AMD-Zen-3-based-EPYC-7713-Milan-server-CPU-spotted-online-obliterates-the-Intel-Xeon-Platinum.502439.0.html

Thoughts: It's great. But I don't understand what the reasoning of the number that EPYC uses. Perhaps it means there's an ever better clocked version? Though compared the last gen the jump in performance isn't as big as I thought it would be.

well, another reason to pick up the dual epyc motherboard

[leadeater]

11 minutes ago, williamcll said:

Thoughts: It's great. But I don't understand what the reasoning of the number that EPYC uses. Perhaps it means there's an ever better clocked version? Though compared the last gen the jump in performance isn't as big as I thought it would be.

The last number is the generation, that's why the current EYPCs all end with a 2 as they are the second generation.

 

[Fatih19]

Ok, this is epyc!

[Master Disaster]

Quote

There is literally no competition, assuming you have $8,000 to spend on a CPU

FTFY

 

Lets be real here, less than a fraction of 1% of home computer users will ever get their hands on something this powerful.

 

It really is a blisteringly fast chip though, wonder if Linus will try to upgrade Whannok?

[Arika]

Quote

There is literally no competitor

[leadeater]

8 minutes ago, Master Disaster said:

Lets be real here, less than a fraction of 1% of home computer users will ever get their hands on something this powerful.

#TeamOnePercent

 

 

Spoiler

 

[Taf the Ghost]

@leadeaterwith the hard flex

[Taf the Ghost]

On the topic at hand, even for a same core counts, Milan is an even bigger monster than Rome. What I'm really looking forward to is the weird core counter numbers AMD can offer. The 8 c / 8 CCD custom-order models are going to be funny.

[Arika]

3 minutes ago, leadeater said:

-hidden snip-

why are all your CPUs rusty?

[leadeater]

1 minute ago, Taf the Ghost said:

@leadeaterwith the hard flex

I mean I do have more...... so only half flex?

 

Spoiler

Like another 12 more lol

 

[Taf the Ghost]

Just now, leadeater said:

I mean I do have more...... so only half flex?

 

  Hide contents

Like another 12 more lol

 

Do you want a "Mega Hard" or "Giga Hard" flex?

[leadeater]

3 minutes ago, Arika S said:

why are all your CPUs rusty?

Rust, where? The heat spreaders are aluminum, or plated copper I forget. Rust either way isn't possible. They are dirty AF though.

[leadeater]

4 minutes ago, Taf the Ghost said:

On the topic at hand, even for a same core counts, Milan is an even bigger monster than Rome. What I'm really looking forward to is the weird core counter numbers AMD can offer. The 8 c / 8 CCD custom-order models are going to be funny.

Next large server replacement round if Zen 3 EPYC is out by then I'm going to try and get my work to make the switch. They should be just that much better and the cache layout improvements mean I won't have to worry about odd performance for larger VMs, which we have a lot of.

[paddy-stone]

 

Had to do it

 

[Taf the Ghost]

2 minutes ago, leadeater said:

Next large server replacement round if Zen 3 EPYC is out by then I'm going to try and get my work to make the switch. They should be just that much better and the cache layout improvements mean I won't have to worry about odd performance for larger VMs, which we have a lot of.

SIs have had a long time to work on Mlian, along with building off socket compatibility with Rome. Those full lane PCIe 4.0 designs took a while.

[leadeater]

15 minutes ago, Taf the Ghost said:

SIs have had a long time to work on Mlian, along with building off socket compatibility with Rome. Those full lane PCIe 4.0 designs took a while.

Honestly for the most part our servers have no use for PCIe lanes, 2 25GB NICs and HBA where one of the NICs and the HBA are imbedded so easy to connect to CPU PCIe lanes over short distance.

 

Basically we don't use many server class NVMe SSDs because tbh you really don't need them.  

[Taf the Ghost]

3 minutes ago, leadeater said:

Honestly for the most part our servers have no use for PCIe lanes, 2 25GB NICs and HBA where one of the NICs and the HBA are imbedded so easy to connect to CPU PCIe lanes over short distance.

 

Basically we don't use many server class NVMe SSDs because tbh you really don't need them.  

Not surprising. All of those PCIe 4.0 lanes aren't needed for most server setups. Still, getting those out took a while for the SIs.

[Lurick]

1 hour ago, leadeater said:

#TeamOnePercent

 

Spoiler

[Taf the Ghost]

1 minute ago, Lurick said:

  Reveal hidden contents

Tech Flex Thread, lol

[AlexGoesHigh]

1 hour ago, Lurick said:

  Hide contents

IIRC left are QSFP modules (not sure though) but what are the big chunguses on the right?

[Lurick]

1 minute ago, AlexGoesHigh said:

IIRC left are QSFP modules (not sure though) but what are the big chunguses on the right?

Those are CPAK modules (the big ones).

Yes, the QSFPs are the smaller ones

[bcredeur97]

this is gonna be good but imagine threadripper, where they usually make the clockspeeds higher. Lol

[Doobeedoo]

Can't wait to see the new TR though

[sub68]

5 hours ago, Arika S said:

why are all your CPUs rusty?

looks like the top layer started to come off leaving the copper exposed

or they are dirty