AMD Readies 9 Threadripper models

[leadeater]

Just now, Falconevo said:

It will come, the early opterons didn't have quad support to start with if memory serves.

Think it'll be in Zen+ which could have a different way of linking the CPUs together. They might go with a switch fabric controller like the HPE The Machine, they are in the Gen-z working group after all and the Infinity Fabric is just an early form of that.

[Falconevo]

4 minutes ago, leadeater said:

Think it'll be in Zen+ which could have a different way of linking the CPUs together. They might go with a switch fabric controller like the HPE The Machine, they are in the Gen-z working group after all and the Infinity Fabric is just an early form of that.

With CERN getting involved in the early development stages of the CPU for testing etc with their OpenStack stuff I would imagine they are really keen to see quad socket boards to improve their compute capacity 2 fold for the same rack U space.

[kuddlesworth9419]

Such a huge CPU. I really want it. I have a 5820k though so I can't give a good reason to.

[cj09beira]

8 minutes ago, Falconevo said:

With CERN getting involved in the early development stages of the CPU for testing etc with their OpenStack stuff I would imagine they are really keen to see quad socket boards to improve their compute capacity 2 fold for the same rack U space.

for that wouldn't blades be a better option?

[Space Reptile]

3 minutes ago, kuddlesworth9419 said:

Such a huge CPU. I really want it. I have a 5820k though so I can't give a good reason to.

id upgrade that GTX 680 first  

[Lurick]

3 minutes ago, kuddlesworth9419 said:

Such a huge CPU. I really want it. I have a 5820k though so I can't give a good reason to.

I can think of the perfect reason.

AMD needs your money and it's a new shiny object

[cj09beira]

Just now, Lurick said:

I can think of the perfect reason.

AMD needs your money and it's a new shiny object

its even laser etched 

[Falconevo]

1 minute ago, cj09beira said:

for that wouldn't blades be a better option?

depends on use case and already present infrastructure, blade chassis also tie you to a specific vendor.

To save money, most large enterprises will go with off the shelf hardware compatible with their existing infrastructure that is easy to replace when upgrade become available.  I know this is the case with CERN's openstack and ceph projects.

 

 

[leadeater]

9 minutes ago, Falconevo said:

With CERN getting involved in the early development stages of the CPU for testing etc with their OpenStack stuff I would imagine they are really keen to see quad socket boards to improve their compute capacity 2 fold for the same rack U space.

Right now they are solving that with 2U 4 node dual socket systems. A 2U 4 socket system doesn't make a ton of sense unless you need the I/O slots, not an issue with Naples, or have an application that can't parallelize across servers.

[Falconevo]

Just now, leadeater said:

Right now they are solving that with 2U 4 node dual socket systems. A 2U 4 socket system doesn't make a ton of sense unless you need the I/O slots, not an issue with Naples, or have an application that can't parallelize across servers.

I would say it would be more for their compute nodes for the openstack project they have running (it is huge) and they are always looking to improve the compute capability without increased U space chewed up.

As soon as the AMD stuff is available on test/evaluation from Dell/HP etc I will get hold of some and let you guys know what the deal is server side.  Having been through eval and testing with the Intel v6, its just more of the same shit.

[cj09beira]

1 minute ago, Falconevo said:

I would say it would be more for their compute nodes for the openstack project they have running (it is huge) and they are always looking to improve the compute capability without increased U space chewed up.

As soon as the AMD stuff is available on test/evaluation from Dell/HP etc I will get hold of some and let you guys know what the deal is server side.  Having been through eval and testing with the Intel v6, its just more of the same shit.

we would be very grateful for that insight

[leadeater]

5 minutes ago, Falconevo said:

I would say it would be more for their compute nodes for the openstack project they have running (it is huge) and they are always looking to improve the compute capability without increased U space chewed up.

As soon as the AMD stuff is available on test/evaluation from Dell/HP etc I will get hold of some and let you guys know what the deal is server side.  Having been through eval and testing with the Intel v6, its just more of the same shit.

We had HPE in last week covering upcoming stuff like Gen-z, Gen10 product range etc. Gen10 is getting AMD processors and it's the same old story as before i.e. DL380 vs DL385 so no great difference in configuration.

 

The AMD systems have better I/O expansion/bandwidth and support higher maximum RAM configuration. No solid details on performance comparisons as they haven't actually finished that testing yet.

 

One big warning is Intel are raising their pricing by 30% across the board so Gen10 is going to be more expensive than Gen9, due to this HPE is extending Gen9 sales life.

 

Some of that is NDA information but I forget which is,  it's going to be announced soon anyway so can't get in to that big trouble 

[SpaceGhostC2C]

14 hours ago, Space Reptile said:

call now and get 32 cores

 

BUT WAIT THERE IS MORE

 

 

64 cores and 128 threads if you call now!!!!!

 

RIP Xeon PHI?  

 

*sigh* there's no way I'm getting one... But if I could convince my employer somehow that it is a worthy investment despite its fast depreciation...

[i_build_nanosuits]

Who gives a fuck about a 16 core ''threadripper'' chip...AMD...what we want and need is the ''SINGLE-threadripper'' we need a bloody fast hyper-threaded quad-core or maybe a six core but something that has the IPC of a CPU from another world...not ''more of these ''weaker than intel's'' cores'' we need FASTER...not MORE! (at least i do ) so bring your A game and push a single-threadripper my way FFS!

[Jito463]

23 minutes ago, i_build_nanosuits said:

Who gives a fuck about a 16 core ''threadripper'' chip...AMD...what we want and need is the ''SINGLE-threadripper'' we need a bloody fast hyper-threaded quad-core or maybe a six core but something that has the IPC of a CPU from another world...not ''more of these ''weaker than intel's'' cores'' we need FASTER...not MORE! (at least i do ) so bring your A game and push a single-threadripper my way FFS!

AMD releases single-core TR at 10GHz.

 

There you go, a single ThreadRipper. 

[TheRandomness]

9 hours ago, FloRolf said:

Wait, what exactly does that mean? 

Does that mean each of these cpus consists of 4 cpus?

9 hours ago, ARikozuM said:

Four dies each with X CCX's, I think.

No, it's two dies, but each die contains 2 CCXs, each containing 4 cores. So 4+3+4+3 should be written as 4+3,4+3 where 1 core in 2 CCXs is disabled.

[suicidalfranco]

WANT!

 

But then i'll be so spoiled by the amount of VM u0kk be running that no laptop will ever feel the same with the measly one VM i have on them  just like my desktop. Poor people one VM machines 

[Jito463]

1 hour ago, TheRandomness said:

No, it's two dies, but each die contains 2 CCXs, each containing 4 cores. So 4+3+4+3 should be written as 4+3,4+3 where 1 core in 2 CCXs is disabled.

I would imagine it's more like 4+4, 3+3.  I can't see them mis-matching the CCX cores in each die.

[TheRandomness]

1 minute ago, Jito463 said:

I would imagine it's more like 4+4, 3+3.  I can't see them mis-matching the CCX cores in each die.

Maybe they like symmetry and it's actually 4+3,3+4  And I know I'm missing the space after the comma, that's intentional.

[Drak3]

10 hours ago, JurunceNK said:

Uhh the Threadripper will also have 4 dies, each having up to 4 cores each, like the Ryzen 7 and Ryzen 5 having two dies with up to 4 cores each.

A single die, for both Ryzen and Threadripper, consists of two CCXs that cannot fully function independently of each other, even if you completely disable the cores of one CCX.

The APUs will probably be different, with a single CCX die and iGPU, but that's a different story from AMD's performance mainstream and enthusiast lineups.

[PlayStation 2]

CPU's sizable, but whatever works. Those ten-core models got my interest.

[Doobeedoo]

Nice! Maybe could've been less models but whatever. CPU looks awesome, specs are looking great and number of lanes is neat and also across the entire range is very nice!

[2bitmarksman]

Anyone else want one to underclock and undervolt for a cheap and smaller server build?

[MoonSpot]

Has anyone found out if ryzen laser snipped to disable cores on their lower end CCXs?  I can totally see people going batshit crazy for low end threadripper cpus if they're able to hack them into higherend cpus.  Would be pretty epyc to be that 1 asshole who's got a 15c/30t cpu

[SpaceGhostC2C]

I've just realized something: AMD may have a trouble in some market segments unless Intel goes all-in with high prices again. I'm referring to those going with the workstation platform for the PCIe lanes and quad-channel RAM, who may be well served by 8-core or even 6-core CPUs. Consumer Ryzen has the cores, but not the PCIe etc, and Threadripper starts at 10 (according to this rumor: they could still do a 2+2+2+2 CPU for X399). Hence, even if Intel's 10-core and above are more expensive, their 6 or maybe even 8-core could be cheaper than the entry-level Threadripper, while having all the X299 features. Then again, I could see a problem for AMD as well if they had to price X399 versions of Ryzen 5 and 7...

Needless to say, if Intel's 6-core parts are priced the same as the cheapest Threadripper, then they don't need to worry