So, where is AMD Seattle?

[DrMrLordX]

Right, but the X2150 (mentioned in the linked hothardware article) is only a quad, as is that $199 Broadwell-D quad. There may be a niche where Seattle can fit in. Maybe.

 

[mrmt]

Right, but the X2150 (mentioned in the linked hothardware article) is only a quad, as is that $199 Broadwell-D quad. There may be a niche where Seattle can fit in. Maybe.

The score mentioned is for the 8C version:

Update: AMD is claiming a SpecInt_Rate score of ~80 for the 8-core Seattle at a 25W TDP, compared to a SpecInt score of 28.1 for the quad-core X2150. An Intel Avoton (top SKU) has a SpecInt score of 106 and a TDP of 20W. How well SpecInt maps to workloads in the microserver industry is, of course, an open question, but Intel's 22nm process gives it a non-trivial performance/watt advantage in this test.

If this is what AMD is really claiming then Seattle is DOA. It cannot beat the 2013 Avoton in terms of raw performance, and this year we should see the 14nm server Atom, and I'm not even counting Broadwell-D. Maybe they can showcase Seattle in a good light against Bulldozer servers.

 

[richaron]

If this is what AMD is really claiming then Seattle is DOA.

You're psychic powers put us mere mortals to shame. And here we thought we were in a technical forum.

I, for one, have very little frame of reference for how a multi-core ARM server handles light/moderate/heavy loads with regards to performance & power. You were even able to make a decision without knowing exact costs or total system features/power!!

 

[imported_ats]

Not in SPECint 2k6 it isn't...

http://hothardware.com/News/AMD-Ann...ore-ARM-SoC-At-Facebooks-Open-Compute-Summit/

Late, slow, and probably consumers more power than Avoton. AMD has a real winner here.

not to mention that Avaton is almost 2 years old and will likely be getting a refresh soon... Its quite possible that by the time Seattle ships in volume its competition will be a 3Ghz Avaton with similar functionality (and it wouldn't even be hard since Intel can copy/paste large chunks of Xeon-D).

 

[imported_ats]

I assume the release of ARMv8.1-a at the end of last year helped to delay Seattle.

I've been waiting for a powerful router to serve as NAS/home server but I might just go old school & build it myself. At least with something like this it'll be relatively low power (important for always on where I live), but with enough grunt for a PBX server & PVR backend. A few SATA ports and 10G networks were limiting my choices.

Then you should probably be looking at an C2750 or C2758. Much better supported than any ARM based solution (its basically the goto choice for things like pfsense, nas, et al even by the developers), just as fast, and available from multiple vendors in multiple configurations. If you really have a use case for 10g, just buy say a connetx-2 off ebay for <$80.

 

[monstercameron]

Then you should probably be looking at an C2750 or C2758. Much better supported than any ARM based solution, just as fast, and available from multiple vendors in multiple configurations. If you really have a use case for 10g, just buy say a connetx-2 off ebay for <$80.

how do you define this?

 

[imported_ats]

how do you define this?

Been on the market for almost 2 years, standard plug and play support by windows, linux, BSD, Solaris. Goto platform for most of the solutions within the target markets: freenas, pfsense, etc (aka they actually sell full on C2750/C2758 retail solutions). Support any add on cards that work in any OS that support x64 just like any other x64 part. Used in production systems by pretty much every cloud based company (C2550 power basically all the storage at Facebook for example, and C235x power all the routers at Facebook).

AKA it just works with anything that works with x64 which is basically everything.

Everything ARM based is still very much experimental and will be that way for some time.

 

[AtenRa]

not to mention that Avaton is almost 2 years old and will likely be getting a refresh soon... Its quite possible that by the time Seattle ships in volume its competition will be a 3Ghz Avaton with similar functionality (and it wouldn't even be hard since Intel can copy/paste large chunks of Xeon-D).

You will not see a 3GHz 14nm Avaton because it will destroy the 4Core XEON-D and Intel doesnt want that.

 

[imported_ats]

You will not see a 3GHz 14nm Avaton because it will destroy the 4Core XEON-D and Intel doesnt want that.

Eh? 2.4Ghz Avaton comes no where close to impacting 4 core Xeon-D. 3Ghz won't either. They play in very different market segments in both power and performance. Avaton is a <=20W market and has performance that can't out perform 4 Core cores even at 2Ghz vs 3Ghz for Avaton. Really it doesn't even come close to being a performance impact.

 

[AtenRa]

Eh? 2.4Ghz Avaton comes no where close to impacting 4 core Xeon-D. 3Ghz won't either. They play in very different market segments in both power and performance. Avaton is a <=20W market and has performance that can't out perform 4 Core cores even at 2Ghz vs 3Ghz for Avaton. Really it doesn't even come close to being a performance impact.

Do you have data for a 20W TDP 8 Core Avoton vs 45W TDP 4 Core XEON-D ??

 

[ShintaiDK]

I, for one, have very little frame of reference for how a multi-core ARM server handles light/moderate/heavy loads with regards to performance & power. You were even able to make a decision without knowing exact costs or total system features/power!!

Anandtech have tested ARM servers.

 

[imported_ats]

Do you have data for a 20W TDP 8 Core Avoton vs 45W TDP 4 Core XEON-D ??

People have only benched the 8c Xeon D but its pretty easy to extrapolate performance to the 4 core version. The 4 core should be significantly faster in most multi-threaded workloads and absolutely smoke the C2750 in single-thread workloads.

 

[Nothingness]

Do you have data for a 20W TDP 8 Core Avoton vs 45W TDP 4 Core XEON-D ??

There are some benchmarks here for an 8-core Xeon D (pre-prod 1540 with slightly slower frequencies than the released one).

http://www.servethehome.com/intel-xeon-d-1540-power-performance-preview/

It definitely smokes the C2750 (even if you halve the MT result of c-ray).

 

[richaron]

Then you should probably be looking at an C2750 or C2758. Much better supported than any ARM based solution (its basically the goto choice for things like pfsense, nas, et al even by the developers), just as fast, and available from multiple vendors in multiple configurations. If you really have a use case for 10g, just buy say a connetx-2 off ebay for <$80.

I use open source software... So if I compile it for ARM then your suggestion have much worse support

And whilst I personally don't need much of "vendors" or "configurations"(because in this class of components I'd use my current hotswap case & drives), I am surprised you think they're worth mentioning as advantages over an unreleased product...

Lastly I see little point having a multi-drive NAS limited by Gb networks, so if I'm going to buy the hardware it will allow me to expand into 10Gb. No, I wouldn't consider adding a card which will completely negate any power advantages and and blow the cost out (unless I go uber leet with 4/5 port 10Gb card and make my own switch also).

 

[mrmt]

You're psychic powers put us mere mortals to shame. And here we thought we were in a technical forum.

I, for one, have very little frame of reference for how a multi-core ARM server handles light/moderate/heavy loads with regards to performance & power. You were even able to make a decision without knowing exact costs or total system features/power!!

25% higher TDP and 20% less performance on SPECint at 30% less price doesn't leave much room for doubt that the product is DOA. And given that it was AMD itself that gave the SPECint number, this should be an optimum case for their server chip.

 

[jpiniero]

There are some benchmarks here for an 8-core Xeon D (pre-prod 1540 with slightly slower frequencies than the released one).

http://www.servethehome.com/intel-xeon-d-1540-power-performance-preview/

It definitely smokes the C2750 (even if you halve the MT result of c-ray).

88.9 W full system load is pretty high for a 45 W TDP chip. That's some very power hungry RAM they got there.

 

[ShintaiDK]

88.9 W full system load is pretty high for a 45 W TDP chip. That's some very power hungry RAM they got there.

Its 88.2W isnt it?

Well its wall power using some unknown PSU brick. 64GB memory+SSD and mobo.

It already uses 5W just with BMC only. And if we say a 80% PSU efficiency its already down to 70W.

So 70W on the Pico+SSD+VRM+mobo features+CPU. If the CPU is 45W, it only leaves 25W for the rest. Sounds quite normal.

 

[richaron]

25% higher TDP and 20% less performance on SPECint at 30% less price doesn't leave much room for doubt that the product is DOA. And given that it was AMD itself that gave the SPECint number, this should be an optimum case for their server chip.

True I shouldn't have been sarcastic before. Obv's they're wrong to price a (apparently) lower performing chip at a lower level?

 

[mrmt]

True I shouldn't have been sarcastic before. Obv's they're wrong to price a (apparently) lower performing chip at a lower level?

They are wrong in not pricing lower enough. 30% less price isn't enough to offset the lower 20% performance with 25% more power consumption compared to Atom C. If you will run this server in your garage or in a corner of your office, then it's a fine deal, but if you are going to a Datacenter, then Seattle TCO will be bad.

 

[NTMBK]

They are wrong in not pricing lower enough. 30% less price isn't enough to offset the lower 20% performance with 25% more power consumption compared to Atom C. If you will run this server in your garage or in a corner of your office, then it's a fine deal, but if you are going to a Datacenter, then Seattle TCO will be bad.

Oh, have you seen some measurements of J/task on final released hardware? Cool, link plz!

 

[richaron]

They are wrong in not pricing lower enough. 30% less price isn't enough to offset the lower 20% performance with 25% more power consumption compared to Atom C. If you will run this server in your garage or in a corner of your office, then it's a fine deal, but if you are going to a Datacenter, then Seattle TCO will be bad.

But it's not 25% more power consumption. It's rated TDP has been (rumoured?) to be 25% higher and you're just making up performance numbers based on a single data point (or was I on to something when I thought you were psychic??).

All lols aside you simply don't have enough data to say it's DOA, and what data you do have is being misrepresented. Eventually someone like me (who's occupation is based on logic) will call you out. There are plenty of more arguments to be made, so continue if you wish.

 

[mrmt]

Oh, have you seen some measurements of J/task on final released hardware? Cool, link plz!

Oh, come on. You have been here long enough to know that these pre-release benchmarks are to shown the product under the best possible light, and if that's what AMD is showing then we can have much, much lower expectations to what will be released. When was the last time you saw an AMD product perform better than AMD benchmarks?

That said, with what they have been released so far I wouldn't even bother to test a system with the chip.

 

[mrmt]

All lols aside you simply don't have enough data to say it's DOA, and what data you do have is being misrepresented. Eventually someone like me (who's occupation is based on logic) will call you out. There are plenty of more arguments to be made, so continue if you wish.

I do have data. Wake me up in Q4, when we should be seeing plenty of OEMs lining up for Seattle chips.

 

[richaron]

They are wrong in not pricing lower enough. 30% less price isn't enough to offset the lower 20% performance with 25% more power consumption compared to Atom C. If you will run this server in your garage or in a corner of your office, then it's a fine deal, but if you are going to a Datacenter, then Seattle TCO will be bad.

These are the data points you gave. Factor in power for PSU/RAM/disks/etc and we have: 30% less price for 20% less performance and slightly more power (~5-10% at worst?).

You said this was "DOA". Even with your own numbers you come to the wrong conclusion o_0

 

[dahorns]

These are the data points you gave. Factor in power for PSU/RAM/disks/etc and we have: 30% less price for 20% less performance and slightly more power (~5-10% at worst?).

You said this was "DOA". Even with your own numbers you come to the wrong conclusion o_0

Those numbers sound like a product that will find a niche use among nonprofessional users looking to save upfront costs, not a product that would compete for use with the high volume customers.