By: Wilco (Wilco.Dijkstra.delete@this.ntlworld.com), September 19, 2009 3:15 pm
Room: Moderated Discussions
David Kanter (dkanter@realworldtech.com) on 9/19/09 wrote:
---------------------------
>Wilco (Wilco.Dijkstra@ntlworld.com) on 9/19/09 wrote:
>---------------------------
>>David Kanter (dkanter@realworldtech.com) on 9/18/09 wrote:
>>---------------------------
>>>To achieve the performance of a 2S Nehalem, you're going to need a ton of ARMs.
>>>And they all need to have ethernet to communicate. Those cables are not free and
>>>use space and power, as do the switches they connect to.
>>
>>The current Cortex-A9 doesn't compete with Nehalem indeed. However the gap with
>>x86 is closing fast, and I'd expect the next generation to be far closer.
>>
>>What low power enables is packing many sockets in a single rack - 8-16 sockets
>>should be possible. As long as they can talk to each other, you don't need more
>>than 1 network connection per board.
>
>So how do you want them to talk? Coherency for 16S isn't trivial (look how long
>it took intel to get there), and if it's not coherent and you are using a network
>stack...then you still need an on-board switch of some sort.
ARM doesn't have something like HyperTransport, and the market is too small to be worth adding inter-socket coherency. So you could use an on-board switch and use an ethernet connection in each socket, or run the network stack on one socket, and DMA the packets to the relevant cores via a fast serial link.
>>Stacking 2 - 4 boards in a 1U rack is feasible
>>if you solder cores and DRAM on the board - no sockets, >heatsinks or DIMMS means
>>much more vertical space (I don't know whether the failure >rates of DRAM are low
>>enough for this to be cost effective - but avoiding >sockets/DIMMS should improve reliability).
>
>I think Arun mentioned that SGI did this for a lot of their large altix systems.
OK
>>Assuming a conservative 2 boards, 8 sockets, 4 cores per >socket, you could have
>>64 cores in a 1U rack (ie. 8 times more than a typical 2S >quad core today).
>
>That's still a lot of hypothetical infrastructure. 2S x86s exist today, and they will have 12 cores per socket soon.
It's an interesting thought experiment though - all it needs is a startup making it happen. How much power does a 12 core x86 chip use? Frequency would be below 2GHz for a reasonable power budget. Not being power limited allows ARM to close the performance gap even quicker.
Wilco
---------------------------
>Wilco (Wilco.Dijkstra@ntlworld.com) on 9/19/09 wrote:
>---------------------------
>>David Kanter (dkanter@realworldtech.com) on 9/18/09 wrote:
>>---------------------------
>>>To achieve the performance of a 2S Nehalem, you're going to need a ton of ARMs.
>>>And they all need to have ethernet to communicate. Those cables are not free and
>>>use space and power, as do the switches they connect to.
>>
>>The current Cortex-A9 doesn't compete with Nehalem indeed. However the gap with
>>x86 is closing fast, and I'd expect the next generation to be far closer.
>>
>>What low power enables is packing many sockets in a single rack - 8-16 sockets
>>should be possible. As long as they can talk to each other, you don't need more
>>than 1 network connection per board.
>
>So how do you want them to talk? Coherency for 16S isn't trivial (look how long
>it took intel to get there), and if it's not coherent and you are using a network
>stack...then you still need an on-board switch of some sort.
ARM doesn't have something like HyperTransport, and the market is too small to be worth adding inter-socket coherency. So you could use an on-board switch and use an ethernet connection in each socket, or run the network stack on one socket, and DMA the packets to the relevant cores via a fast serial link.
>>Stacking 2 - 4 boards in a 1U rack is feasible
>>if you solder cores and DRAM on the board - no sockets, >heatsinks or DIMMS means
>>much more vertical space (I don't know whether the failure >rates of DRAM are low
>>enough for this to be cost effective - but avoiding >sockets/DIMMS should improve reliability).
>
>I think Arun mentioned that SGI did this for a lot of their large altix systems.
OK
>>Assuming a conservative 2 boards, 8 sockets, 4 cores per >socket, you could have
>>64 cores in a 1U rack (ie. 8 times more than a typical 2S >quad core today).
>
>That's still a lot of hypothetical infrastructure. 2S x86s exist today, and they will have 12 cores per socket soon.
It's an interesting thought experiment though - all it needs is a startup making it happen. How much power does a 12 core x86 chip use? Frequency would be below 2GHz for a reasonable power budget. Not being power limited allows ARM to close the performance gap even quicker.
Wilco
Topic | Posted By | Date |
---|---|---|
ARM announces 2GHz netbook CPU | Wilco | 2009/09/16 11:00 AM |
ARM announces 2GHz netbook CPU | Richard Cownie | 2009/09/16 03:51 PM |
ARM announces 2GHz netbook CPU | Potatoswatter | 2009/09/16 04:12 PM |
ARM announces 2GHz netbook CPU | Megol | 2009/09/16 04:21 PM |
ARM announces 2GHz netbook CPU | Wilco | 2009/09/18 01:13 AM |
ARM announces 2GHz netbook CPU | Megol | 2009/09/18 11:54 AM |
ARM announces 2GHz netbook CPU | someone | 2009/09/16 06:01 PM |
ARM announces 2GHz netbook CPU | Seni | 2009/09/16 06:41 PM |
ARM announces 2GHz netbook CPU | someone | 2009/09/16 07:32 PM |
ARM announces 2GHz netbook CPU | anon | 2009/09/16 08:58 PM |
ARM announces 2GHz netbook CPU | David Kanter | 2009/09/17 04:53 AM |
ARM announces 2GHz netbook CPU | someone | 2009/09/17 06:10 AM |
ARM announces 2GHz netbook CPU | David Kanter | 2009/09/17 03:02 PM |
ARM announces 2GHz netbook CPU | anon | 2009/09/20 05:34 PM |
ARM announces 2GHz netbook CPU | Seni | 2009/09/17 02:05 PM |
ARM announces 2GHz netbook CPU | Wilco | 2009/09/17 03:29 PM |
ARM announces 2GHz netbook CPU | Seni | 2009/09/17 03:53 PM |
ARM announces 2GHz netbook CPU | Wilco | 2009/09/17 11:31 PM |
ARM announces 2GHz netbook CPU | David Kanter | 2009/09/18 04:17 PM |
ARM announces 2GHz netbook CPU | Jouni Osmala | 2009/09/16 09:40 PM |
ARM announces 2GHz netbook CPU | Martin Høyer Kristiansen | 2009/09/17 01:15 AM |
ARM announces 2GHz netbook CPU | Wilco | 2009/09/18 01:56 AM |
ARM announces 2GHz netbook CPU | Jouni Osmala | 2009/09/18 04:46 AM |
ARM announces 2GHz netbook CPU | Michael S | 2009/09/18 06:22 AM |
ARM announces 2GHz netbook CPU | Wilco | 2009/09/18 11:36 AM |
ARM announces 2GHz netbook CPU | rcf | 2009/09/19 08:49 AM |
The other sides of the coin | ? | 2009/09/19 11:09 AM |
The other sides of the coin | Paul | 2009/09/20 02:30 AM |
The other sides of the coin | Jukka Larja | 2009/09/20 10:20 PM |
ARM announces 2GHz netbook CPU | AM | 2009/09/19 03:41 AM |
ARM announces 2GHz netbook CPU | Michael S | 2009/09/18 05:15 AM |
ARM announces 2GHz netbook CPU | Michael S | 2009/09/18 05:48 AM |
ARM announces 2GHz netbook CPU | Wilco | 2009/09/18 01:15 PM |
ARM announces 2GHz netbook CPU | Michael S | 2009/09/20 09:00 AM |
ARM announces 2GHz netbook CPU | Linus Torvalds | 2009/09/20 09:59 AM |
ARM announces 2GHz netbook CPU | Michael S | 2009/09/20 10:42 AM |
ARM announces 2GHz netbook CPU | Linus Torvalds | 2009/09/20 02:19 PM |
ARM announces 2GHz netbook CPU | Michael S | 2009/09/20 03:12 PM |
ARM announces 2GHz netbook CPU | anonymous | 2009/09/20 04:58 PM |
ARM announces 2GHz netbook CPU | Michael S | 2009/09/21 02:49 AM |
ARM announces 2GHz netbook CPU | Wilco | 2009/09/21 03:38 AM |
ARM announces 2GHz netbook CPU | Linus Torvalds | 2009/09/21 07:05 AM |
ARM announces 2GHz netbook CPU | none | 2009/09/21 07:10 AM |
ARM announces 2GHz netbook CPU | Linus Torvalds | 2009/09/21 08:24 AM |
ARM announces 2GHz netbook CPU | IntelUser2000 | 2009/09/21 06:59 PM |
ARM announces 2GHz netbook CPU | Linus Torvalds | 2009/09/21 07:15 PM |
Atom-based Octane III | David Wragg | 2009/09/22 04:11 AM |
Atom-based Octane III | Arun Ramakrishnan | 2009/09/22 10:02 PM |
Many Atoms -> Molecule ? :-) (NT) | Anon4 | 2009/09/23 03:49 AM |
Many Atoms -> Molecule ? :-) (NT) | Arun Ramakrishnan | 2009/09/23 04:11 AM |
Scalar DFlops/Hz | Michael S | 2009/09/20 12:00 PM |
Scalar DFlops/Hz | Wilco | 2009/09/20 12:45 PM |
VMLA.f64 | Michael S | 2009/09/20 01:46 PM |
VMLA.f64 | Paul | 2009/09/20 02:16 PM |
VMLA.f64 | Wilco | 2009/09/20 04:45 PM |
ARM announces 2GHz netbook CPU | David Kanter | 2009/09/18 04:19 PM |
Who wants an ARM netbook? | slacker | 2009/09/17 12:05 AM |
Who wants an ARM netbook? | slacker | 2009/09/17 12:21 AM |
Who wants an ARM netbook? | Wilco | 2009/09/17 02:10 AM |
Who wants an ARM netbook? | xxx | 2009/09/17 02:31 AM |
Who wants an ARM netbook? | Wilco | 2009/09/17 03:04 AM |
Who wants an ARM netbook? | Hello | 2009/09/21 06:47 AM |
Who wants an ARM netbook? | none | 2009/09/21 09:04 AM |
Who wants an ARM netbook? | Wilco | 2009/09/21 02:12 PM |
Who wants an ARM netbook? | Hello | 2009/09/22 02:46 AM |
Who wants an ARM netbook? | Gian-Carlo Pascutto | 2009/09/17 02:24 AM |
Who wants an ARM netbook? | slacker | 2009/09/17 04:28 AM |
Who wants an ARM netbook? | Gian-Carlo Pascutto | 2009/09/17 05:23 AM |
Who wants an ARM netbook? | rcf | 2009/09/17 06:58 AM |
Who wants an ARM netbook? | Cooper | 2009/09/17 03:33 PM |
Who wants an ARM netbook? | anon | 2009/09/17 05:32 PM |
Who wants an ARM netbook? | anon | 2009/09/17 10:50 PM |
Who wants an ARM netbook? | Howard Chu | 2009/09/17 08:48 PM |
Who wants an ARM netbook? | Foo_ | 2009/09/21 03:44 AM |
ARM for server | KISS | 2009/09/17 06:13 AM |
ARM for server | Linus Torvalds | 2009/09/17 06:57 AM |
ARM for server | Adrian | 2009/09/17 08:04 AM |
ARM for server | Richard Cownie | 2009/09/17 08:19 AM |
ARM for server | Adrian | 2009/09/17 08:49 AM |
Let's talk about an ARM-inside big machine | KISS | 2009/09/17 09:08 AM |
Let's talk about an ARM-inside big machine | anon | 2009/09/17 09:44 AM |
Let's talk about an ARM-inside big machine | Richard Cownie | 2009/09/17 11:00 AM |
article about google server | Richard Cownie | 2009/09/17 11:04 AM |
Let's talk about an ARM-inside big machine | anon | 2009/09/17 11:28 AM |
Let's talk about an ARM-inside big machine | Richard Cownie | 2009/09/17 11:47 AM |
Let's talk about an ARM-inside big machine | anon | 2009/09/17 10:31 PM |
Networking | David Kanter | 2009/09/18 05:13 PM |
Networking | Wilco | 2009/09/19 01:35 AM |
Networking | Paul | 2009/09/19 05:47 AM |
Networking | anon | 2009/09/19 06:03 AM |
Networking | David Kanter | 2009/09/19 02:04 PM |
Networking | Wilco | 2009/09/19 03:15 PM |
Networking | David Kanter | 2009/09/19 03:37 PM |
Networking | Wilco | 2009/09/20 05:11 AM |
Let's talk about an ARM-inside big machine | a reader | 2009/09/17 01:28 PM |
Let's talk about an ARM-inside big machine | A. No. Nymous | 2009/09/24 04:48 PM |
Let's talk about an ARM-inside big machine | Howard Chu | 2009/09/24 10:19 PM |
Let's talk about an ARM-inside big machine | KISS | 2009/09/26 12:06 AM |
Screen resolutions | Rob Thorpe | 2009/09/18 10:09 AM |
Screen resolutions | Howard Chu | 2009/09/18 12:32 PM |
ARM for server | Groo | 2009/09/17 12:15 PM |
ARM for server | Richard Stacpoole | 2009/09/17 02:49 PM |
ARM for server | Richard Cownie | 2009/09/17 06:35 PM |
ARM for server | Richard Stacpoole | 2009/09/17 09:21 PM |
ARM for server | Linus Torvalds | 2009/09/18 06:33 AM |
ARM for server | Michael S | 2009/09/18 07:15 AM |
ARM for server | Richard Cownie | 2009/09/18 07:56 AM |
ARM for server | slacker | 2009/09/18 03:50 PM |
causes of mechanical stress | Richard Cownie | 2009/09/18 04:27 PM |
causes of mechanical stress | slacker | 2009/09/18 05:42 PM |
causes of mechanical stress | Richard Cownie | 2009/09/18 06:06 PM |
causes of mechanical stress | Howard Chu | 2009/09/18 06:14 PM |
causes of mechanical stress | Richard Cownie | 2009/09/18 07:33 PM |
causes of mechanical stress | Howard Chu | 2009/09/18 09:31 PM |
causes of mechanical stress | Anin | 2009/09/19 10:28 PM |
causes of mechanical stress | Megol | 2009/09/22 06:58 AM |
causes of mechanical stress | slacker | 2009/09/18 08:13 PM |
causes of mechanical stress | Richard Cownie | 2009/09/18 08:52 PM |
causes of mechanical stress | David Kanter | 2009/09/18 10:31 PM |
causes of mechanical stress | Ungo | 2009/09/22 04:34 PM |
causes of mechanical stress | Richard Cownie | 2009/09/23 10:00 AM |
causes of mechanical stress | Linus Torvalds | 2009/09/19 08:44 AM |
causes of mechanical stress | Richard Cownie | 2009/09/19 11:24 AM |
causes of mechanical stress | slacker | 2009/09/19 04:33 PM |
anecdotal lack of reliability of Macs | kdg | 2009/09/21 12:08 PM |
anecdotal lack of reliability of Macs | anon | 2009/09/22 09:30 AM |
anecdotal lack of reliability of Macs | kdg | 2009/09/22 11:33 AM |
ARM for server | Anon | 2009/09/18 07:35 PM |
ARM for server | anon | 2009/09/18 08:20 PM |
ARM for server | RagingDragon | 2009/09/19 12:33 PM |
ARM for server | anon | 2009/09/20 11:29 AM |
ARM for server | Anon | 2009/09/19 02:29 PM |
ARM for server | anon | 2009/09/20 11:24 AM |
ARM for server | RagingDragon | 2009/09/20 02:16 PM |
[O.T.] "carbon steel" | Michael S | 2009/09/20 03:45 PM |
[O.T.] "carbon steel" | slacker | 2009/09/20 05:04 PM |
[O.T.] "carbon steel" | RagingDragon | 2009/09/21 09:42 PM |
[O.T.] "carbon steel" | Michael S | 2009/09/22 12:36 AM |
ARM for server | Anon | 2009/09/20 06:52 PM |
ARM for server | anon | 2009/09/26 07:56 AM |
ARM for server | Ungo | 2009/09/22 04:25 PM |
ARM for server | RagingDragon | 2009/09/23 11:13 AM |
ARM for server | Carlie Coats | 2009/09/24 09:28 AM |
ARM for server | slacker | 2009/09/18 08:24 PM |
ARM for server | Richard Stacpoole | 2009/09/18 05:34 PM |
ARM for server | Jukka Larja | 2009/09/18 09:17 PM |
ARM for server | Megol | 2009/09/19 02:18 AM |
ARM for server | Richard Cownie | 2009/09/18 07:31 AM |
ARM for server | Richard Stacpoole | 2009/09/18 05:18 PM |
ARM for server | Richard Cownie | 2009/09/18 09:20 PM |
ARM for server | Richard Stacpoole | 2009/09/19 12:02 AM |
ARM for server | Richard Cownie | 2009/09/19 03:57 AM |
ARM for server | Richard Stacpoole | 2009/09/19 04:24 AM |
ARM for server | Richard Cownie | 2009/09/19 07:12 AM |
ARM for server | Jukka Larja | 2009/09/18 09:57 PM |
ARM for server | Mark Christiansen | 2009/09/18 03:17 PM |
Cortex performance | Tom W | 2009/09/19 08:52 AM |
Cortex performance | anon | 2009/09/19 09:50 AM |
Cortex performance | Howard Chu | 2009/09/19 11:57 AM |
Cortex performance | Howard Chu | 2009/09/19 12:05 PM |
Cortex performance | Wilco | 2009/09/19 01:38 PM |
Cortex performance | Howard Chu | 2009/09/20 10:59 PM |
No silicon yet? | someone | 2009/09/17 12:18 PM |
No silicon yet? | anon | 2009/09/18 12:06 AM |
No silicon yet? | anon | 2009/09/18 07:30 AM |
No silicon yet? | Hello | 2009/09/21 07:38 AM |
No silicon yet? | anon | 2009/09/21 12:44 PM |
No silicon yet? | none | 2009/09/21 11:54 PM |
No silicon yet? | Wilco | 2009/09/22 01:24 AM |
No silicon yet? | ? | 2009/09/22 11:18 AM |
No silicon yet? | IntelUser2000 | 2009/09/22 12:13 PM |
No silicon yet? | none | 2009/09/22 01:40 PM |
No silicon yet? | Wilco | 2009/09/22 02:53 PM |
No silicon yet? | a reader | 2009/09/23 07:45 AM |
No silicon yet? | ? | 2009/09/22 11:34 PM |
Why China Mobile choose Marvell PXA920? | Michael S | 2009/09/22 02:49 AM |
Why China Mobile choose Marvell PXA920? | Wilco | 2009/09/22 02:23 PM |
Why China Mobile choose Marvell PXA920? | Michael S | 2009/09/22 02:46 PM |
Why China Mobile choose Marvell PXA920? | Wilco | 2009/09/22 03:03 PM |
Why China Mobile choose Marvell PXA920? | Michael S | 2009/09/22 04:48 PM |
Why China Mobile choose Marvell PXA920? | Paul | 2009/09/22 05:46 PM |
Why China Mobile choose Marvell PXA920? | Anon4 | 2009/09/23 03:53 AM |