Sponsors
Sponsor Products
New Xeon pricing
posted by Jean-Francois Panisset  on July 11, 2017, 6:15 p.m. (15 days ago)
9 Responses     0 Plus One's     0 Comments  
Ars has details on the new Xeon pricing: http://www.anandtech.com/show/11544/intel-skylake-ep-vs-amd-epyc-7000-cpu-battle-of-the-decade/7 First impressions: - nothing is affordable in the "Platinum" range, cheapest part is a 16 core 2.0GHz 8153 for $3K+ - in the "Gold" range, at around $2K there's the 14 core, 2.6GHz 140W 6132 for $2111 which kind of lines up with the 14 core, 2.6GHz 135W E5-2690v4 for $2090 Which seems a bit disappointing: for the last few Xeon generations the pricing seem to come out at "you get 2 more cores for the same price point", now it's "you get the same core count, same GHZ, higher power draw and slightly more expensive". Although the 6132 will TurboBoost to 3.7GHz for single core performance vs 3.5GHz for the E5-2690v4, but that's less useful for efficiently multi-threaded rendering applications. As an aside, "Platinum" / "Gold" / "Silver" / "Bronze" sound like options at the car wash or gym memberships. I wonder who decided those were good branding ideas. JF To unsubscribe from the list send a blank e-mail to mailto:studiosysadmins-discuss-request@studiosysadmins.com?subject=unsubscribe

Thread Tags:
  discuss-at-studiosysadmins 

Response from Travis Best @ July 17, 2017, 11:34 a.m.

I'm aware that Epyc is effectively just released, but has anyone got hardware in to demo?  I'm exploring possibilities to get something in to see how it works and if I can get CentOS 7.x running on it smoothly.  It appears the initial install is a bit more involved that "image and forget" as you need to disable SMT for the initial install then install kernel-ml from epel-kernel before re-enabling.  I'm quite curious to see if the increase in memory channels effects render times and/or results in different output from the Xeons.


0 Plus One's     0 Comments  
   

Response from Jean-Francois Panisset @ July 14, 2017, 2 p.m.
Something else interesting with these new parts is that quad socket configurations are no longer unattainable, for instance SuperMicro has 4 socket blades for its new "8U SuperBlade" chassis: https://www.supermicro.com/products/superblade/module/SBI-8149P-C4N.cfm You could imagine a system with 4 x Xeon Gold 6132 parts with 56 physical / 112 HT cores: if your renderer of choice scales up to that many cores, you could in theory cut your licensing costs in half over the more traditional 2 x 2P servers. Would love to see some benchmarks for 1 x 4P vs 2 x 2P using the same CPUs... JF On Wed, Jul 12, 2017 at 4:56 PM, Joseph Chonacky wrote: > > > On Wed, Jul 12, 2017 at 1:37 PM, Will Rosecrans > wrote: >> >> I think "all core boost" is a temporary thing. If renders take more than >> a few seconds, it'll have to drop back from being boosty. If it could >> sustain that speed, they'd just advertise it as the base clock rather than a >> boost. >> >> It'll still be neat for some bursty things, like a Nuke workstation that >> wants to peg all cores for a short amount of time to render a frame very >> quickly, then you wait for somebody to make an absurd comment like "make it >> look further away, but not smaller in the frame" while the cores cool off >> for a few seconds, artist makes a tweak, frame renders for a second, etc., >> 50 times until the artist kills somebody. You only accumulate a few minutes >> of boost speed over the course of the hour. >> >> > > Shared this comment with our lead Nuke compositor. He corroborates the > described workflow. > > Joe Chonacky > IT Manager > Bent Image Lab > 503.228.6206 > > > > To unsubscribe from the list send a blank e-mail to > mailto:studiosysadmins-discuss-request@studiosysadmins.com?subject=unsubscribe To unsubscribe from the list send a blank e-mail to mailto:studiosysadmins-discuss-request@studiosysadmins.com?subject=unsubscribe

0 Plus One's     0 Comments  
   

Response from Joe Chonacky @ July 12, 2017, 8 p.m.


On Wed, Jul 12, 2017 at 1:37 PM, Will Rosecrans <wrosecrans@gmail.com> wrote:
I think "all core boost" is a temporary thing. If renders take more than a few seconds, it'll have to drop back from being boosty. If it could sustain that speed, they'd just advertise it as the base clock rather than a boost.
It'll still be neat for some bursty things, like a Nuke workstation that wants to peg all cores for a short amount of time to render a frame very quickly, then you wait for somebody to make an absurd comment like "make it look further away, but not smaller in the frame" while the cores cool off for a few seconds, artist makes a tweak, frame renders for a second, etc., 50 times until the artist kills somebody. You only accumulate a few minutes of boost speed over the course of the hour.


Shared this comment with our lead Nuke compositor. He corroborates the described workflow.
Joe ChonackyIT ManagerBent Image Lab503.228.6206

0 Plus One's     0 Comments  
   

Response from Will Rosecrans @ July 12, 2017, 4:40 p.m.
I think "all core boost" is a temporary thing. If renders take more than a few seconds, it'll have to drop back from being boosty. If it could sustain that speed, they'd just advertise it as the base clock rather than a boost.
It'll still be neat for some bursty things, like a Nuke workstation that wants to peg all cores for a short amount of time to render a frame very quickly, then you wait for somebody to make an absurd comment like "make it look further away, but not smaller in the frame" while the cores cool off for a few seconds, artist makes a tweak, frame renders for a second, etc., 50 times until the artist kills somebody. You only accumulate a few minutes of boost speed over the course of the hour.
On Wed, Jul 12, 2017 at 12:36 AM, Peter Smith <peter.smith@framestore.com> wrote:
Hmm. No details on the "all core boost" speeds.

Which is what your looking at when you're rendering, right?

On 11 July 2017 at 23:11, Jean-Francois Panisset <panisset@gmail.com> wrote:
Ars has details on the new Xeon pricing:

http://www.anandtech.com/show/11544/intel-skylake-ep-vs-amd-epyc-7000-cpu-battle-of-the-decade/7

First impressions:

- nothing is affordable in the "Platinum" range, cheapest part is a 16
core 2.0GHz 8153 for $3K+
- in the "Gold" range, at around $2K there's the 14 core, 2.6GHz 140W
6132 for $2111 which kind of lines up with the 14 core, 2.6GHz 135W
E5-2690v4 for $2090

Which seems a bit disappointing: for the last few Xeon generations the
pricing seem to come out at "you get 2 more cores for the same price
point", now it's "you get the same core count, same GHZ, higher power
draw and slightly more expensive". Although the 6132 will TurboBoost
to 3.7GHz for single core performance vs 3.5GHz for the E5-2690v4, but
that's less useful for efficiently multi-threaded rendering
applications.

As an aside, "Platinum" / "Gold" / "Silver" / "Bronze" sound like
options at the car wash or gym memberships. I wonder who decided those
were good branding ideas.

JF
To unsubscribe from the list send a blank e-mail to mailto:studiosysadmins-discuss-request@studiosysadmins.com?subject=unsubscribe



--
Framestore Peter Smith Senior Systems Engineer
London New York Los Angeles Chicago Montral
T+44 (0)20 7344 8000 M+44 (0)7816 123009
19-23 Wells Street, London W1T 3PQ
Twitter Facebook framestore.com
https://www.framestore.com/

To unsubscribe from the list send a blank e-mail to mailto:studiosysadmins-discuss-request@studiosysadmins.com?subject=unsubscribe


0 Plus One's     0 Comments  
   

Response from Ian Stewart @ July 12, 2017, 3:03 p.m.

I just received the pricing on all of the CPUs this morning. If anyone wants a full list of the CPUs and prices, just send me an email at ians@icc-usa.com

 

I also have pricing on all of the new Supermicro X11 models as well if you want to price out a system.

 

Cheers,

Ian

 

 


0 Plus One's     0 Comments  
   

Response from Jimmy Christensen @ July 12, 2017, 5:25 a.m.
What I found most interesting was the EPYC prices and benchmarks against the Xeons.
- Jimmy
On 12 July 2017 at 10:06, Peter Smith <peter.smith@framestore.com> wrote:
Ah yes, the next page hidden as a dropdown menu. I've not had coffee yet ...

On 12 July 2017 at 08:54, Jean-Francois Panisset <panisset@gmail.com> wrote:
The next "page" on that Anandtech article, "Intel's Turbo Modes", has some analysis on the frequency at which cores will actually run on a Platinum 8180 part (that's going to take a while to get used to) . My take away was "it depends, you'll never figure it out". It even depends on the instruction set you are using.

There's some verbiage about how "Each core can have its frequency adjusted independently, allowing VMs to take advantage of different workload types and not be hamstrung by occupants on other VMs in the same socket.". And it does seem like the real use case for these Xeon parts is multi-tenant, multi-VM workloads, we're just lucky that you can also happen to be able to run highly parallelized ray tracing code for rendering.

JF



On Wed, Jul 12, 2017 at 12:36 AM, Peter Smith <peter.smith@framestore.com> wrote:
Hmm. No details on the "all core boost" speeds.

Which is what your looking at when you're rendering, right?

On 11 July 2017 at 23:11, Jean-Francois Panisset <panisset@gmail.com> wrote:
Ars has details on the new Xeon pricing:

http://www.anandtech.com/show/11544/intel-skylake-ep-vs-amd-epyc-7000-cpu-battle-of-the-decade/7

First impressions:

- nothing is affordable in the "Platinum" range, cheapest part is a 16
core 2.0GHz 8153 for $3K+
- in the "Gold" range, at around $2K there's the 14 core, 2.6GHz 140W
6132 for $2111 which kind of lines up with the 14 core, 2.6GHz 135W
E5-2690v4 for $2090

Which seems a bit disappointing: for the last few Xeon generations the
pricing seem to come out at "you get 2 more cores for the same price
point", now it's "you get the same core count, same GHZ, higher power
draw and slightly more expensive". Although the 6132 will TurboBoost
to 3.7GHz for single core performance vs 3.5GHz for the E5-2690v4, but
that's less useful for efficiently multi-threaded rendering
applications.

As an aside, "Platinum" / "Gold" / "Silver" / "Bronze" sound like
options at the car wash or gym memberships. I wonder who decided those
were good branding ideas.

JF
To unsubscribe from the list send a blank e-mail to mailto:studiosysadmins-discuss-request@studiosysadmins.com?subject=unsubscribe



--
Framestore Peter Smith Senior Systems Engineer
London New York Los Angeles Chicago Montral
T+44 (0)20 7344 8000 M+44 (0)7816 123009
19-23 Wells Street, London W1T 3PQ
Twitter Facebook framestore.com
https://www.framestore.com/

To unsubscribe from the list send a blank e-mail to mailto:studiosysadmins-discuss-request@studiosysadmins.com?subject=unsubscribe


To unsubscribe from the list send a blank e-mail to mailto:studiosysadmins-discuss-request@studiosysadmins.com?subject=unsubscribe



--
Framestore Peter Smith Senior Systems Engineer
London New York Los Angeles Chicago Montral
T+44 (0)20 7344 8000 M+44 (0)7816 123009
19-23 Wells Street, London W1T 3PQ
Twitter Facebook framestore.com
https://www.framestore.com/

To unsubscribe from the list send a blank e-mail to mailto:studiosysadmins-discuss-request@studiosysadmins.com?subject=unsubscribe


0 Plus One's     0 Comments  
   

Response from Anonymous @ July 12, 2017, 4:10 a.m.
Ah yes, the next page hidden as a dropdown menu. I've not had coffee yet ...

On 12 July 2017 at 08:54, Jean-Francois Panisset <panisset@gmail.com> wrote:
The next "page" on that Anandtech article, "Intel's Turbo Modes", has some analysis on the frequency at which cores will actually run on a Platinum 8180 part (that's going to take a while to get used to) . My take away was "it depends, you'll never figure it out". It even depends on the instruction set you are using.

There's some verbiage about how "Each core can have its frequency adjusted independently, allowing VMs to take advantage of different workload types and not be hamstrung by occupants on other VMs in the same socket.". And it does seem like the real use case for these Xeon parts is multi-tenant, multi-VM workloads, we're just lucky that you can also happen to be able to run highly parallelized ray tracing code for rendering.

JF



On Wed, Jul 12, 2017 at 12:36 AM, Peter Smith <peter.smith@framestore.com> wrote:
Hmm. No details on the "all core boost" speeds.

Which is what your looking at when you're rendering, right?

On 11 July 2017 at 23:11, Jean-Francois Panisset <panisset@gmail.com> wrote:
Ars has details on the new Xeon pricing:

http://www.anandtech.com/show/11544/intel-skylake-ep-vs-amd-epyc-7000-cpu-battle-of-the-decade/7

First impressions:

- nothing is affordable in the "Platinum" range, cheapest part is a 16
core 2.0GHz 8153 for $3K+
- in the "Gold" range, at around $2K there's the 14 core, 2.6GHz 140W
6132 for $2111 which kind of lines up with the 14 core, 2.6GHz 135W
E5-2690v4 for $2090

Which seems a bit disappointing: for the last few Xeon generations the
pricing seem to come out at "you get 2 more cores for the same price
point", now it's "you get the same core count, same GHZ, higher power
draw and slightly more expensive". Although the 6132 will TurboBoost
to 3.7GHz for single core performance vs 3.5GHz for the E5-2690v4, but
that's less useful for efficiently multi-threaded rendering
applications.

As an aside, "Platinum" / "Gold" / "Silver" / "Bronze" sound like
options at the car wash or gym memberships. I wonder who decided those
were good branding ideas.

JF
To unsubscribe from the list send a blank e-mail to mailto:studiosysadmins-discuss-request@studiosysadmins.com?subject=unsubscribe



--
Framestore Peter Smith Senior Systems Engineer
London New York Los Angeles Chicago Montral
T+44 (0)20 7344 8000 M+44 (0)7816 123009
19-23 Wells Street, London W1T 3PQ
Twitter Facebook framestore.com
https://www.framestore.com/

To unsubscribe from the list send a blank e-mail to mailto:studiosysadmins-discuss-request@studiosysadmins.com?subject=unsubscribe


To unsubscribe from the list send a blank e-mail to mailto:studiosysadmins-discuss-request@studiosysadmins.com?subject=unsubscribe



--
Framestore Peter Smith Senior Systems Engineer
London New York Los Angeles Chicago Montral
T+44 (0)20 7344 8000 M+44 (0)7816 123009
19-23 Wells Street, London W1T 3PQ
Twitter Facebook framestore.com
https://www.framestore.com/

0 Plus One's     0 Comments  
   

Response from Jean-Francois Panisset @ July 12, 2017, 3:55 a.m.
The next "page" on that Anandtech article, "Intel's Turbo Modes", has some analysis on the frequency at which cores will actually run on a Platinum 8180 part (that's going to take a while to get used to) . My take away was "it depends, you'll never figure it out". It even depends on the instruction set you are using.

There's some verbiage about how "Each core can have its frequency adjusted independently, allowing VMs to take advantage of different workload types and not be hamstrung by occupants on other VMs in the same socket.". And it does seem like the real use case for these Xeon parts is multi-tenant, multi-VM workloads, we're just lucky that you can also happen to be able to run highly parallelized ray tracing code for rendering.

JF



On Wed, Jul 12, 2017 at 12:36 AM, Peter Smith <peter.smith@framestore.com> wrote:
Hmm. No details on the "all core boost" speeds.

Which is what your looking at when you're rendering, right?

On 11 July 2017 at 23:11, Jean-Francois Panisset <panisset@gmail.com> wrote:
Ars has details on the new Xeon pricing:

http://www.anandtech.com/show/11544/intel-skylake-ep-vs-amd-epyc-7000-cpu-battle-of-the-decade/7

First impressions:

- nothing is affordable in the "Platinum" range, cheapest part is a 16
core 2.0GHz 8153 for $3K+
- in the "Gold" range, at around $2K there's the 14 core, 2.6GHz 140W
6132 for $2111 which kind of lines up with the 14 core, 2.6GHz 135W
E5-2690v4 for $2090

Which seems a bit disappointing: for the last few Xeon generations the
pricing seem to come out at "you get 2 more cores for the same price
point", now it's "you get the same core count, same GHZ, higher power
draw and slightly more expensive". Although the 6132 will TurboBoost
to 3.7GHz for single core performance vs 3.5GHz for the E5-2690v4, but
that's less useful for efficiently multi-threaded rendering
applications.

As an aside, "Platinum" / "Gold" / "Silver" / "Bronze" sound like
options at the car wash or gym memberships. I wonder who decided those
were good branding ideas.

JF
To unsubscribe from the list send a blank e-mail to mailto:studiosysadmins-discuss-request@studiosysadmins.com?subject=unsubscribe



--
Framestore Peter Smith Senior Systems Engineer
London New York Los Angeles Chicago Montral
T+44 (0)20 7344 8000 M+44 (0)7816 123009
19-23 Wells Street, London W1T 3PQ
Twitter Facebook framestore.com
https://www.framestore.com/

To unsubscribe from the list send a blank e-mail to mailto:studiosysadmins-discuss-request@studiosysadmins.com?subject=unsubscribe


0 Plus One's     0 Comments  
   

Response from Anonymous @ July 12, 2017, 3:40 a.m.
Hmm. No details on the "all core boost" speeds.

Which is what your looking at when you're rendering, right?

On 11 July 2017 at 23:11, Jean-Francois Panisset <panisset@gmail.com> wrote:
Ars has details on the new Xeon pricing:

http://www.anandtech.com/show/11544/intel-skylake-ep-vs-amd-epyc-7000-cpu-battle-of-the-decade/7

First impressions:

- nothing is affordable in the "Platinum" range, cheapest part is a 16
core 2.0GHz 8153 for $3K+
- in the "Gold" range, at around $2K there's the 14 core, 2.6GHz 140W
6132 for $2111 which kind of lines up with the 14 core, 2.6GHz 135W
E5-2690v4 for $2090

Which seems a bit disappointing: for the last few Xeon generations the
pricing seem to come out at "you get 2 more cores for the same price
point", now it's "you get the same core count, same GHZ, higher power
draw and slightly more expensive". Although the 6132 will TurboBoost
to 3.7GHz for single core performance vs 3.5GHz for the E5-2690v4, but
that's less useful for efficiently multi-threaded rendering
applications.

As an aside, "Platinum" / "Gold" / "Silver" / "Bronze" sound like
options at the car wash or gym memberships. I wonder who decided those
were good branding ideas.

JF
To unsubscribe from the list send a blank e-mail to mailto:studiosysadmins-discuss-request@studiosysadmins.com?subject=unsubscribe



--
Framestore Peter Smith Senior Systems Engineer
London New York Los Angeles Chicago Montral
T+44 (0)20 7344 8000 M+44 (0)7816 123009
19-23 Wells Street, London W1T 3PQ
Twitter Facebook framestore.com
https://www.framestore.com/

0 Plus One's     0 Comments