AMD announce EPYC

[Deleted member 209350]

Milan? Nope, literally nothing in the channel to suggests that it is close. Also I will probably be able to procure some pre production test chips when the time comes so will possibly be able to source them early again. There is certainly room in my estate for a couple of highly clocked bits of Milan Silicon

Thanks! Was just wondering as they are going to be the first of the Zen3 architecture to release so was hoping to have heard some news on it at least but I guess not!

I very much look forward to seeing some high clocked Milan silicon

 

[Journey]

Milan is due into production in Q3 of this year, according to AMD's own road map, tapeout completed Q2/3' 19. Genoa is going to be the biggest change though, 2021-22 for that.

as they are going to be the first of the Zen3 architecture to release

Nope.

 

[Vince]

Milan is due into production in Q3 of this year, according to AMD's own road map, tapeout completed Q2/3' 19. Genoa is going to be the biggest change though, 2021-22 for that.



Nope.

I imagine the guys at HP will start talking about it somewhere in Q2 with some silicon flying around, around then. TBH im not sure why I need more than Rome but in for a penny and all that

 

[Armageus]

I imagine the guys at HP will start talking about it somewhere in Q2 with some silicon flying around, around then. TBH im not sure why I need more than Rome but in for a penny and all that

Be a little while yet I expect - G10 Plus Servers have only just started hitting retail. Whether these (Milan) form G11 or not I have no idea, rumour has it that G10 Plus was meant to be G11, but Intel released no new chips/platform in time to align for a full Generation referesh.

 

[Vince]

Be a little while yet I expect - G10 Plus Servers have only just started hitting retail. Whether these (Milan) form G11 or not I have no idea, rumour has it that G10 Plus was meant to be G11, but Intel released no new chips/platform in time to align for a full Generation referesh.

That does sound like it could well be the case. My understanding was that G11 was basically ready so renaming that to G10 plus and then going for a full new stack on Milan would make a lot of sense.

 

[blaineoliver]

I imagine the guys at HP will start talking about it somewhere in Q2 with some silicon flying around, around then. TBH im not sure why I need more than Rome but in for a penny and all that

You can bet that at HPE we get samples much earlier than a quarter before realise. Writing firmware takes a while.

 

[Vince]

You can bet that at HPE we get samples much earlier than a quarter before realise. Writing firmware takes a while.

Id be concerned if you didn't assuming you work for HPE. What I was trying to say is that I don't tend to hear or get to play until perhaps a month or two before they hit the shelves.

 

[blaineoliver]

Id be concerned if you didn't assuming you work for HPE. What I was trying to say is that I don't tend to hear or get to play until perhaps a month or two before they hit the shelves.

Yeah that sounds about right

 

[ChrisD.]

This may be of interest to those thinking about getting an EPYC based server with more than 32 cores on one socket.

https://www.vmware.com/company/news/updates/cpu-pricing-model-update-feb-2020.html

 

[Hotwired]

This may be of interest to those thinking about getting an EPYC based server with more than 32 cores on one socket.

https://www.vmware.com/company/news/updates/cpu-pricing-model-update-feb-2020.html

They would like to counter savings available by having more cores per licence by insisting each licence only covers 32 cores but you can buy more licences

There may be a rush to get greater than 32 core cpus in before the 30th April as you'll dodge the additional licence costs which will apply after that.

Also great incentive to find or create alternatives...

 

[Vince]

They would like to counter savings available by having more cores per licence by insisting each licence only covers 32 cores but you can buy more licences

There may be a rush to get greater than 32 core cpus in before the 30th April as you'll dodge the additional licence costs which will apply after that.

Also great incentive to find or create alternatives...

I don't like their licensing, particularly the backdating support over missed years to renew support. Don't like that, it's also because of that I now have about 4 different essentials packages when really some flexibility in the licensing could have avoided the confusion. You can't blame them for the way they are handling core count though, I like it a lot more than Microsoft and the per core model.

 

[Journey]

Interesting article about CERN and AMD with Epyc severe.

https://www.fudzilla.com/news/51056-epyc-processors-determine-a-third-of-cern-servers-are-dead

 

[humbug]

Interesting article about CERN and AMD with Epyc severe.

https://www.fudzilla.com/news/51056-epyc-processors-determine-a-third-of-cern-servers-are-dead

Is it just me or is the title of that article a little too easy to misunderstand?

 

[Vince]

Is it just me or is the title of that article a little too easy to misunderstand?

Yea it's not great is it. Article was decent though, 40tb a second throughput is significant!

 

[humbug]

Yea it's not great is it. Article was decent though, 40tb a second throughput is significant!

Mental.... that's what it is. Mental Throughput.

 

[Vince]

Mental.... that's what it is. Mental Throughput.

I can't imagine what 40tb of data a second looks like.. How do you process that kind of volume?

 

[BongoHunter]

I guess a lot of that data could never be written to disk (and doesn't all have to be), the number of drives you'd need to write that much data a second would be way to high (never mind capacity needed!)

 

[Journey]

I can't imagine what 40tb of data a second looks like.. How do you process that kind of volume?

Most I've had to deal with is shoving 14,000 MB/s so ~112Gb/s into a single system and that was only in bursts of 15 secs to 10 minutes max. Although I am hoping to get some of the Mellanox 200Gbs cards to test out over the next year, along with some switch gear (obviously.

I wonder what sort of storage solution they are using.

 

[SupraWez]

Some serious numbers there

 

[Vince]

Most I've had to deal with is shoving 14,000 MB/s so ~112Gb/s into a single system and that was only in bursts of 15 secs to 10 minutes max. Although I am hoping to get some of the Mellanox 200Gbs cards to test out over the next year, along with some switch gear (obviously.

I wonder what sort of storage solution they are using.

The maximum throughput of the storage solution we use is about 1700MB/s aggregated thats effectively 48Gb/s of throughput. The 40tb/s number is mental, I don't even know what sort of solution could cope with that sort of data rate. Sounds awesome though.