[SA] gigabytes arm servers represent sea change market

[monstercameron]

source: http://semiaccurate.com/2015/03/30/gigabytes-arm-servers-represent-sea-change-market/

Gigabyte just launched three new ARM server products that don’t look like much on the surface. Under it however there is something that signifies a sea-change in the entire server market.

The three products are a board and a 1U system based on the Applied Micro X-Gene SoC and a 1U storage unit built around an quad-core A15 from Annapurna Labs. Like we said earlier it doesn’t sound all that exciting to describe, this has all been done before.

hmm so this is where amd sees potential and intel is wary [xeon-d anyone?].

 

[ShintaiDK]

X-Gene soc was tested at Anandtech. And it was...a joke really. Like previous ARM attempts on the server segment.

They wanted to compete with Haswell Xeons, but couldnt even compete with Atom.

 

[Nothingness]

X-Gene soc was tested at Anandtech. And it was...a joke really. Like previous ARM attempts on the server segment.

They wanted to compete with Haswell Xeons, but couldnt even compete with Atom.

Yeah, X-Gene 1 looks poor. It reminds me of Intel first attempts at entering the mobile market, many strong claims, short on delivery. Sorry, couldn't resist :biggrin:

 

[Arachnotronic]

Yeah, X-Gene 1 looks poor. It reminds me of Intel first attempts at entering the mobile market, many strong claims, short on delivery. Sorry, couldn't resist :biggrin:

I don't think Atom was ever hyped up as much as X-Gene was. These AMCC execs were out there pooh-pooh'ing "standard tablet" cores like the A57 while literally claiming X-Gene 1 would have "Xeon-class" performance.

The fact that it literally performs no better than an Avoton but guzzles a multiple of the power that Silvermont requires is a pretty serious fail. I don't think any of the Intel mobile SoCs starting with Medfield were quite this bad competitively.

 

[Nothingness]

I don't think Atom was ever hyped up as much as X-Gene was. These AMCC execs were out there pooh-pooh'ing "standard tablet" cores like the A57 while literally claiming X-Gene 1 would have "Xeon-class" performance.

Yeah, you're right. But I'm just tired of people claiming ARM will never materialize in the server market, they sound as wrong as those who were claiming Intel would never succeed in the mobile market. This is utterly stupid.

The fact that it literally performs no better than an Avoton but guzzles a multiple of the power that Silvermont requires is a pretty serious fail. I don't think any of the Intel mobile SoCs starting with Medfield were quite this bad competitively.

I wonder why X-Gene 1 is making its way to the market. I always thought it was just a demonstrator, and the X-Gene 2 @ 28nm would be the real consumer product.

 

[Arachnotronic]

I wonder why X-Gene 1 is making its way to the market. I always thought it was just a demonstrator, and the X-Gene 2 @ 28nm would be the real consumer product.

Yeah, that's how people started viewing it after the initial 40nm part was delayed and delayed to the point that it wasn't viable as a commercial product

 

[CHADBOGA]

I wonder why X-Gene 1 is making its way to the market. I always thought it was just a demonstrator, and the X-Gene 2 @ 28nm would be the real consumer product.

This all eerily sounds like the birth of Itanium.

 

[NTMBK]

I don't think Atom was ever hyped up as much as X-Gene was. These AMCC execs were out there pooh-pooh'ing "standard tablet" cores like the A57 while literally claiming X-Gene 1 would have "Xeon-class" performance.

The fact that it literally performs no better than an Avoton but guzzles a multiple of the power that Silvermont requires is a pretty serious fail. I don't think any of the Intel mobile SoCs starting with Medfield were quite this bad competitively.

To be fair, they never specified which Xeon. Compared to a Netburst Xeon it probably looks pretty good :awe:

 

[Arachnotronic]

To be fair, they never specified which Xeon. Compared to a Netburst Xeon it probably looks pretty good :awe:

Yes they did.

 

[ShintaiDK]

Yeah, you're right. But I'm just tired of people claiming ARM will never materialize in the server market, they sound as wrong as those who were claiming Intel would never succeed in the mobile market. This is utterly stupid.


I wonder why X-Gene 1 is making its way to the market. I always thought it was just a demonstrator, and the X-Gene 2 @ 28nm would be the real consumer product.

Because its the reality. ARM servers isnt something new. It just got more focus on forums because...well..

When 28nm X-Gene 2 comes, there are 14nm Xeons and Atoms. The cycle repeats. Not that 28nm would likely be able to compete with Atom in the first place. The X-Gene 1 uses almost 2½x the power of Atom.

 

[NTMBK]

Yes they did.

Well that's just embarassing. Can't see X-Gene lasting much longer.

 

[Burpo]

What's that fine print say at the bottom? 32 core vs 12 core? 64Gb ram vs 4Gb?

 

[Nothingness]

Because its the reality. ARM servers isnt something new. It just got more focus on forums because...well..

Intel started talking about phone chips in 2008. Perhaps we'll see something this year. And how long did it take Intel to enter significantly the server market?

During all these years Intel were considered as a joke. How things have changed

When 28nm X-Gene 2 comes, there are 14nm Xeons and Atoms. The cycle repeats. Not that 28nm would likely be able to compete with Atom in the first place. The X-Gene 1 uses almost 2½x the power of Atom.

Time will tell. All we know for sure is that X-Gene 1 is not efficient and makes APM look like fools when one reads their claims.

 

[Enigmoid]

What's that fine print say at the bottom? 32 core vs 12 core? 64Gb ram vs 4Gb?

Yep.

Marketing at its finest.

 

[tynopik]

What's that fine print say at the bottom? 32 core vs 12 core? 64Gb ram vs 4Gb?

but they were all 1U half-width units

i guess they're going for rack density comparisons

'yes our chips are slower, but you can pack more of them into same space'

 

[erunion]

Well that's just embarassing. Can't see X-Gene lasting much longer.

Calxeda stopped operations in 2014. Applied micro will be gone in a year or two. By the time AMD brings K12 to market the ARMy could be an army of one.

 

[Headfoot]

Dense ARM servers are far from dead -- just premature, the smart players decided to not enter at all rather than enter with an uncompetitive product. No way Qualcomm would be sitting around on its laurels if they legitimately thought the current Snapdragon iteration was suitable for the dense microserver market. It's telling that AMD, Qualcomm, etc have all waited until ARMv8 officially came out and the 64bit ARM ecosystem evolves on its own natural path without forced (read: expensive) acceleration.

Now, whether they'll be competitive versus a microserver Knight's Landing variant is an entirely different question...

 

[mrmt]

Calxeda stopped operations in 2014. Applied micro will be gone in a year or two. By the time AMD brings K12 to market the ARMy could be an army of one.

When was the last time AMD correctly read a profitable market trend? The fact that AMD is pursuing something is a strong indicator that this thing is bad business.

 

[monstercameron]

When was the last time AMD correctly read a profitable market trend? The fact that AMD is pursuing something is a strong indicator that this thing is bad business.

im sure 1st gen cat cores were a relative success. VIA has been doing it for a while and so was intel.

 

[erunion]

When was the last time AMD correctly read a profitable market trend? The fact that AMD is pursuing something is a strong indicator that this thing is bad business.

Being the only participant left isn't necessarily a good thing for AMD. The existence of multiple companies gives an extra air of legitimacy.

Though if AMD does demonstrate profitability Im sure we'll see others try to get a piece of that pie. But if not, well, what's 100% of nothing?

 

[mrmt]

im sure 1st gen cat cores were a relative success. VIA has been doing it for a while and so was intel.

I'm talking about something a little deeper than that:

- AMD invested in fireball CPU designs at the time the CPU market was shifting to power efficiency.

- AMD invested in compute performance in a time the GPU market was shifting towards graphics and efficiency.

- AMD invested in the semi-custom market in a time where off-the-shelf solutions are gaining even more share.

- AMD sold its mobile GPU business in a moment the world was gearing towards mobile computing.

AMD BoD as a rule of thumb always fail to correctly point the direction the market will follow. Bobcat was a success but that market wasn't AMD invention, it was Intel's. AMD to its merit could place a very good product in the huge gap between Intel's two product lines, and Atom was really bad before Silvermont, but again they just followed Intel's lead on the subject (which isn't a bad strategy at all). Whenever they try to develop something, they screw it up.

 

[monstercameron]

I'm talking about something a little deeper than that:



- AMD invested in fireball CPU designs at the time the CPU market was shifting to power efficiency.



- AMD invested in compute performance in a time the GPU market was shifting towards graphics and efficiency.



- AMD invested in the semi-custom market in a time where off-the-shelf solutions are gaining even more share.



- AMD sold its mobile GPU business in a moment the world was gearing towards mobile computing.



AMD BoD as a rule of thumb always fail to correctly point the direction the market will follow. Bobcat was a success but that market wasn't AMD invention, it was Intel's. AMD to its merit could place a very good product in the huge gap between Intel's two product lines, and Atom was really bad before Silvermont, but again they just followed Intel's lead on the subject (which isn't a bad strategy at all). Whenever they try to develop something, they screw it up.

I think the first point is valid but the second is just bs and marketing from an competitor trying to show their gimped products in a good light. The others are debatable...

 

[NTMBK]

Now, whether they'll be competitive versus a microserver Knight's Landing variant is an entirely different question...

Nah, Knight's Landing isn't aimed at microservers- it's aimed at floating point throughput. It's the Xeon D which is keeping Applied Micro up at night.

 

[podspi]

Intel has made a few missteps as well... Can you imagine how hard they are kicking themselves at not bending over backwards for Steve Jobs to get into that first iPhone?

I think a lot of people have it right, a lot of the first movers for ARM in servers moved too early. 64-bit is basically mandatory.

 

[DrMrLordX]

but they were all 1U half-width units

i guess they're going for rack density comparisons

'yes our chips are slower, but you can pack more of them into same space'

Also note the big letters and numbers denoting the alleged power envelope. The X-Gene systems are supposed to be able to operate moar cores within the same power envelope as compared to the listed Xeon competitors.

Nah, Knight's Landing isn't aimed at microservers- it's aimed at floating point throughput. It's the Xeon D which is keeping Applied Micro up at night.

Theoretically, Knight's Landing could be used to replace a lot of microservers. The power envelope is definitely right. It really depends on the per-thread performance that's needed as to whether Xeon-D or Knight's Landing would be the better choice. But yeah Knight's Landing is more aimed at HPC throughput.