Why Didn't AMD Use Jaguar For Microservers?

[Arachnotronic]

Jaguar is good. Like, really good. In terms of single threaded performance, this thing is probably the fastest low-power core that'll be available on the market. Sure it doesn't have super deep active-idle states, but does this really matter for a micro-server, where the chip will probably be at full load most of the time?

Could somebody help me understand what technical reasons - if any - exist for AMD to have chosen the ARM A57 rather than the Jaguar for this new "Seattle" SoC? From a business standpoint, I suppose going under the ARM banner could drum up some hype, but I feel as though X86 is an ASSET in this space, not a detriment.

My hypothesis is that SeaMicro had this Seattle chip in development BEFORE they were bought by AMD, particularly given the time-frame of the launch.

Thoughts?

 

[Meekers]

I thought they are making arm AND jaguar microservers? I could of sword I read a press release about various jaguar models for microservers.

Edit: They do have jaguar models called kyoto.
http://seekingalpha.com/article/1490182-amd-enters-the-microserver-story-with-kyoto
http://www.techpowerup.com/184607/a...-series-family-of-microserver-processors.html

 

[Arachnotronic]

Meekers,

The Jaguar falls right off the roadmap once Seattle arrives.

 

[Meekers]

Meekers,

The Jaguar falls right off the roadmap once Seattle arrives.

I was under the impression that both would be available, but guess I was wrong.

 

[Arachnotronic]

It's a bummer, too. I think Jaguar will be AMD's most successful core, and it would have been a very sweet cherry on top of it had led the firm's micro-server efforts.

 

[NostaSeronx]

Could somebody help me understand what technical reasons - if any - exist for AMD to have chosen the ARM A57 rather than the Jaguar for this new "Seattle" SoC? From a business standpoint, I suppose going under the ARM banner could drum up some hype, but I feel as though X86 is an ASSET in this space, not a detriment.

X1150 and X2150 will be in microservers, way before Seattle comes out.

 

[Idontcare]

IMO the reason why AMD doesn't have Jaguar in their server plans at this time is because by the time they realized they needed to have something like that in their portfolio it was already far too late in the product development cycle to intersect and pull in such a platform (requires validation/certification/etc with OEMs) on a timeline for which Jaguar-based SKUs would still be relevant.

As to why AMD missed the opportunity and didn't realize it should do something there sooner instead of later...well who best to speak of their mindset at the time than Mr JFAMD himself?

Thanks for posting here, we do appreciate it--don't let some of our more "obnoxious" members discourage you from posting, just ignore list them and don't reply to them, is your best bet. I learn a lot from you, Idontcare, Aigo, and a few others. Anyway.

Few questions:
1. Can you comment on that "accelerate mode" rumor that dresdenboy has posted on his blog?
2. Are you working with Microsoft, etc, to get Windows and it's associated libraries optimized to take advantage of fused multiply add? Do you anticipate it making a noticeable difference in performance? I'm not a programmer or anything, but since it's something you have that Intel doesn't (yet), I'm curious about what support you have lined up for it. Will Intel's 3 operand FMA be able to operate as a "subset" of your 4 version, or will it require completely different stuff?
3. This is a strange one...but are there any Bobcat low power server plans?

1. No I really can't, for 2 reasons: I didn't read enough about his thoughts to know what he is referring to and the other reason is that *if* this is something in the architecture, it is not something that is disclosed yet.
2. Yes, MSFT and all of the key Linux distributions.
3. Not currently. We look at it because we need to make full use of any silicon that we have. The problem isn't the silicon, it's the market. For small servers with low utilization that just need a low power solution, virtualization does a much better job. For cloud clusters you get low power, by you seriously increase the number of physical servers, which ultimately leads to more cost and more management hassle. The market just isn't there.

^ that was in direct discussion regarding Bobcat, but this tells us the mindset regarding Jaguar as well as Jaguar was very much under development at the time.

 

[zir_blazer]

Everyone missed this?

 

[Vesku]

The AMD person making server decisions is a former Seamicro executive, he really likes ARM 64-bit. I suppose they don't have the resources to keep 3 cores in development (ARM, Jaguar, Steamroller/Excavator). Perhaps Steamroller and future cores are being reworked as efficient x86 so they won't need a gap filling core. Us outsiders will have a better view of what's going on with AMD by mid next year.

 

[beginner99]

Everyone missed this?

or this

 

[Vesku]

OP missed the Kyoto chips but there is the other aspect that people noticed, there is no Cat family x86 on the 2014 server roadmap.

ARM 64 bit -> 2 module Steamroller APU -> 6-8 module refined Vishera cores for multi-socket drop in upgrades for G34 systems

 

[NostaSeronx]

@Vesku

x64-2014:
Berlin APU
Warsaw CPU

AArch64-2014:
Seattle CPU

The 28-nm Jaguar SoC will be replaced by the 20-nm Leopard SoC. It isn't on the roadmap because of possible delays. If it can get delayed it won't appear on the roadmap.

 

[NTMBK]

IMO the reason why AMD doesn't have Jaguar in their server plans at this time is because by the time they realized they needed to have something like that in their portfolio it was already far too late in the product development cycle to intersect and pull in such a platform (requires validation/certification/etc with OEMs) on a timeline for which Jaguar-based SKUs would still be relevant.

Probably this. Not to mention their semicustom APU teams were busy working on the PS4 and XBone chips, so they probably didn't have any manpower to spare to make a proper Jaguar server chip. (Which explains why the Jaguar Opteron X parts are just rebadged Kabini.)

 

[Vesku]

That is a good point, an 8 core Jaguar model would be interesting in their Kyoto Jaguar server line up. However, it would probably need dual channel support to have wide appeal.

 

[Dark Shroud]

Neither of the designs for PS34 or Xbone would work in a server to the point of needed almost a whole redesign.

Eventually I think we'll see 8-12-16 core Leopard clusters.

These could do well for AMD. In the past both Microsoft & Facebook were asking Intel for Atom clusters for use in their massive data centers. I do not believe Intel ever delivered on those.

 

[Idontcare]

Probably this. Not to mention their semicustom APU teams were busy working on the PS4 and XBone chips, so they probably didn't have any manpower to spare to make a proper Jaguar server chip. (Which explains why the Jaguar Opteron X parts are just rebadged Kabini.)

Yep, AMD's problems are that they have more great ideas than they have cash to fund.

 

[ShintaiDK]

The 28-nm Jaguar SoC will be replaced by the 20-nm Leopard SoC. It isn't on the roadmap because of possible delays. If it can get delayed it won't appear on the roadmap.

So its currently just a pure fantasy product? Perhaps one you made up?

 

[NTMBK]

So its currently just a pure fantasy product? Perhaps one you made up?

It wouldn't be the first time...

 

[ShintaiDK]

It wouldn't be the first time...

Yep...

This one should be pretty clear what happens in the server space. I wonder if the future potential of the cat cores was simply too low from AMDs point of view to continue. Jaguar is already a product that falls between 2 chairs.

 

[galego]

Jaguar is good. Like, really good. In terms of single threaded performance, this thing is probably the fastest low-power core that'll be available on the market. Sure it doesn't have super deep active-idle states, but does this really matter for a micro-server, where the chip will probably be at full load most of the time?

Could somebody help me understand what technical reasons - if any - exist for AMD to have chosen the ARM A57 rather than the Jaguar for this new "Seattle" SoC? From a business standpoint, I suppose going under the ARM banner could drum up some hype, but I feel as though X86 is an ASSET in this space, not a detriment.

My hypothesis is that SeaMicro had this Seattle chip in development BEFORE they were bought by AMD, particularly given the time-frame of the launch.

Thoughts?

First, AMD is using jaguar in servers

Second, AMD is replacing jaguar servers by ARM servers because:

Seattle is an 8- and then 16-core CPU based on the ARM Cortex-A57 core and is expected to run at or greater than 2GHz. The Seattle processor is expected to offer 2-4 times the performance of AMD’s recently announced AMD Opteron X-Series processor with significant improvement in compute-per-watt.

 

[mrmt]

Jaguar is good. Like, really good. In terms of single threaded performance, this thing is probably the fastest low-power core that'll be available on the market. Sure it doesn't have super deep active-idle states, but does this really matter for a micro-server, where the chip will probably be at full load most of the time?

It's a different business model. With x86 they were alone for most of the time. Not only they had to support Intel instructions, they also had to bring new ideas of their own to their chips. Both Brazos and Kabini also had a huge advantage over their big core siblings: Intel was looking elsewhere.

Without disregarding how efficient these chips are, Intel was fielding a very old core in a lagging edge node to compete with them. This is changing with Silvermont, as Intel poured a lot of resources in designing the chip and is bringing it to bleeding edge nodes. On top of that Airmont is also coming with 14nm and Core is also in the ballpark of Kabini power consumption. Whatever advantage AMD had in the small core market, it will be hard to maintain in the next two years.

When looking at their big core line, and being tied to a subpar foundry partner for the next 11 years, it's easy to see that the cat line may suffer as much as the big core line suffered, so here the ARM options make sense:

- The playfield will more or less leveled, as custom implementations won't have significantly higher performance than the vanilla ARM core of the time

- AMD IP can make a difference there

- Some of the R&D will be funded by ARM itself, further easing cash requirements for AMD.

- There aren't any big established vendors there yet

- In the ARM world they won't be as handicaped as they are now regarding the foundry partner.

To sum up everything, it's not a limitation of the design, but a different business direction. Let's see where this will take them.

 

[Vesku]

AMD could be keeping its ARM cards close to the chest but their public statements about it are that their first 64 bit ARM will be the standard design ARM provides.

 

[iAMunderDog]

Jaguar is being used for micro-server, we heard about "Kyoto" that is renamed into Opteron-X and those are re-badged "Temas/Kabini" APU's that are basically unlocked to a certain things eg frequency of CPU and GPU, but no overclocking.

"Seattle" is replacing "Kyoto" because of number of cores and raw performance, single thread performance won't be better compared to multi threaded performance while the main reason is better power saving features and easier implementation. "Seattle" will incorporate "Freedom Fabric" from SeaMicro thus you will see micro-server clusters.

It will give a further edge over Intel and the competition.

 

[Ancalagon44]

First, AMD is using jaguar in servers



Second, AMD is replacing jaguar servers by ARM servers because:

Ah, but how many cores does each have?

Jaguar: 2-4
Seattle: 8-16

Notice that they did not say performance per core or performance per watt, they simply said performance. 4 times the cores giving you 4 times the performance - I certainly hope so!

 

[galego]

Ah, but how many cores does each have?

Jaguar: 2-4
Seattle: 8-16

Notice that they did not say performance per core or performance per watt, they simply said performance. 4 times the cores giving you 4 times the performance - I certainly hope so!

What about this part?

The Seattle processor is expected to offer 2-4 times the performance of AMD’s recently announced AMD Opteron X-Series processor with significant improvement in compute-per-watt.