Trends in Server Compute 2017 – Qualcomm Centriq 2400

Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.
Patrick

Patrick

Administrator
Staff member
Dec 21, 2010
12,513
5,804
113
  • Like
Reactions: gigatexal
( )

( )

New Member
Jul 8, 2017
27
4
3
Glad you caught my post. I looked up the distance and found it's only 10 minutes away, convenient.

They say they're only sampling to "select clients", have your Elevator Summary ready, we're rooting for you.

At the very least ask if they'll say how much power the 48 Core version uses, my guess is a bit over 18W (CPU only, or 10x that on a MB with Memory).

Nice to be where the action is.
 
E

Evan

Well-Known Member
Jan 6, 2016
3,346
598
113
IPC better than atom c2000 for sure, c3000 probably depends on the workload really.

I am sure they don't make a chip and then ask where is the market ? I would think they have a very specific market in mind that they have worked with for the development of the product.
 
( )

( )

New Member
Jul 8, 2017
27
4
3
Patrick said:
Wattage will be higher. ...
Click to expand...
That's the kind of numbers that Maxcom (formerly APM) is coming out with for the X-Gene 3 (32 Cores of 835 @ 125W).

The Linley Report "Performance Arms X-Gene 3 for Cloud": https://www.linleygroup.com/uploads/x-gene-3-for-cloud.pdf , on page 5, claims that the "X-Gene 3 far outclasses current ARM server processors, including the initial X-Gene products, AMD’s A1100, and Cavium’s ThunderX" (and later compares the Centriq 2400 with the ThunderX2).

While OTOH Qualcomm claims that they've stripped down the ARM 835 (even removing the capability to execute 32-bit code) to maximize it's Bang/Watt.

The Linley Report also contains this Slide (while the price is missing from the X-Gene 3 column it is said to be 1/3 of Intel's offering).

Screenshot_2017-08-21-07-26-43.png

So anyone planning on being competitive has got to be the same or better ballpark (with Intel @ 32W).

Time will tell, and good times these are.

Looking forward to seeing ARM (even MIPS) Server Reviews (as time permits), because we don't want to detract you from time available to acquire a POWER9 Server.

Thanks for your reply, enjoy the HotChips. Rob
 
E

Evan

Well-Known Member
Jan 6, 2016
3,346
598
113
Getting off topic but 14nm FinFET for power9 we may see some TDP reductions but IBM don't really care about idle usage etc and those machines intended to be run hard. I Guess same for the ARM server versions, they are inherently energy efficient but they won't be designed for lowest power usage like we know in mobile phones etc.
 
gigatexal

gigatexal

I'm here to learn
Nov 25, 2012
2,913
607
113
Portland, Oregon
alexandarnarayan.com
( ) said:
That's the kind of numbers that Maxcom (formerly APM) is coming out with for the X-Gene 3 (32 Cores of 835 @ 125W).

The Linley Report "Performance Arms X-Gene 3 for Cloud": https://www.linleygroup.com/uploads/x-gene-3-for-cloud.pdf , on page 5, claims that the "X-Gene 3 far outclasses current ARM server processors, including the initial X-Gene products, AMD’s A1100, and Cavium’s ThunderX" (and later compares the Centriq 2400 with the ThunderX2).

While OTOH Qualcomm claims that they've stripped down the ARM 835 (even removing the capability to execute 32-bit code) to maximize it's Bang/Watt.

The Linley Report also contains this Slide (while the price is missing from the X-Gene 3 column it is said to be 1/3 of Intel's offering).

View attachment 6407

So anyone planning on being competitive has got to be the same or better ballpark (with Intel @ 32W).

Time will tell, and good times these are.

Looking forward to seeing ARM (even MIPS) Server Reviews (as time permits), because we don't want to detract you from time available to acquire a POWER9 Server.

Thanks for your reply, enjoy the HotChips. Rob
Click to expand...
those specs sound awesome, good competition here i am liking that
 
( )

( )

New Member
Jul 8, 2017
27
4
3
Patrick said:
Wattage will be higher. We saw these systems running at OCP months ago
Click to expand...
Some hard numbers I have been able to come up with ...

On Webpage: Server/SpecsAndDesigns - OpenCompute the File: "OCS Open CloudServer Power Supply v2.0" says that the Server has a 1600W Power Supply that services 6 Trays, so 1600/6=266.6, let's call that 250W peak.

There is a File called "OCS_Open_CloudServer_PowerSupply_v2.0.pdf" that on page 18 says: "When set to extended battery operation, the power supply shall look at output power at 37 +/- 1 seconds after AC loss and if output power is less than 75W, the power supply shall continue to run off of battery for 200 seconds".

So a fully functional Server Tray can't peak above 250W (or more exactly, 6 of them over 1600W) and a Tray attempting to shutdown during a power failure is going to be expected to run on less than 75W.

I read an estimate for the similar and also completely different Qualcomm 835. It takes between 2.5 and 5W (it has 8 Cores), so (ballpark Math derived from fairydust) 48/8=6, 6*2.5=15W.

There's 15W, but I'm happy to concede it will be more, it ought to be less than 32W (Intel's) divided by the difference in processing power to make it have better numbers than Intel; otherwise it's dead in the water.

If it took a third of the 75W that the Battery Backup was expected to supply it would be 25W which while not way less than Intel (and probably not enough less) at least it gives their Marketing Dept. a number to throw out.

Patrick, since they didn't give you a Board or a number (so no NDA) do you have an educated guess that you'd care to share; and possibly a performance estimate (IE: Their 48 Cores is approximately equivalent to X Intel Cores)?

Thanks for the Article you wrote and mentioning within the first few paragraphs that they wouldn't be providing those numbers, yet.
 
Last edited:
You must log in or register to reply here.
Top Bottom