If you give birth to sure with the aim of
you require to build a grunting number-cruncher with the aim of crams a sturdy
amount of Tesla or else Xeon Phi coprocessors into a single bodywork, after
that Hewlett-Packard wants to chatter to you roughly the new to the job
ProLiant SL270s Gen8 member of staff serving at table.
The new to the job SL270s Gen8 slides into
the SL6500 "scalable system" modular bodywork, which makes it
somewhere in the sphere of concerning a rack and a blade member of staff
serving at table. The SL6500 bodywork comes in the sphere of 2U and 4U
variants, and is apparent from the "Argos" SL4500 large data scalable
systems with the aim of HP announced only remaining week, which are under
attack by large data jobs with the aim of need to position lots of storage
space on a node. With the SL6500s, the design is roughly packing having the
status of many coprocessors in the sphere of a bodywork having the status of on
the cards to raise suspended line of reasoning computing apiece watt.
HP has been peddling super-dense ProLiant
SL tackle since June 2009, with the launch of the opening SL6000s. These tackle
were a response in the sphere of part to the triumph with the aim of Dell's
Data core Solutions thing and Silicon Graphics/Rackable Systems were having
peddling dense tackle with no all the expensive management and redundancy
facial appearance with the aim of ad blade servers give birth to, and which are
legs on a wind in support of cheapskate hyperscale data core operators and
mainframe centers. Once you give birth to a massively congruence system, you
expect nodes to fail relatively than irritating to prevent it, and your
software is invented to handle with folks failures with no roaring.
HP debuted the much supplementary
impressive SL6500 systems in the sphere of October 2010, which permissible in
support of up to three GPUs to be present configured in the sphere of a
half-wide 2U member of staff serving at table node called the SL390s G7. (There
is a 1U variant of this node, too.) The SL6500 bodywork is 4U important, so you
may well position four of these 2U SL370s G7 nodes into it. Liability so gave
you a greatest of eight Xeon 5600 processors and a dozen Tesla M20X0 cards
crammed into a 4U hole.
To boost the GPU coprocessor density even
supplementary, HP rolled unfashionable a new to the job 4U important variant of
the SL390s G7 node in the sphere of April 2011 with the aim of was similarly
half-wide and permissible a wonking eight GPU coprocessors to be present
configured to a two-socket Xeon 5600 member of staff serving at table. So with
two of these darlings, you may well position four Xeons and sixteen Tesla M20X0
GPU coprocessors in the sphere of with the aim of 4U of hole. Well, you may well
if you had a sufficient amount money in support of electricity.
With the hottest system to approach
unfashionable of HP in the sphere of the scalable systems lineup, the SL6500
bodywork stays the same, but near is a new to the job SL270s Gen8 node with the
aim of is based on Intel's Xeon E5-2600 processors. This node has space in
support of the eight x16 peripheral cards, and this clock around they can be
present the new to the job Xeon Phi x86-based coprocessors and the new to the
job Tesla K10 GPU coprocessors from Nvidia.
The spec sheet in support of the new to the
job SL member of staff serving at table node says with the aim of it ropes the
Tesla M2070Q, M2075, M2090, and K10 GPU coprocessor cards and remarks with the
aim of you need to give birth to a bare minimum of 4GB of highest reminiscence
apiece socket if you install GPU coprocessors in the sphere of the node. It
besides says with the aim of on folks GPU coprocessors with the aim of give
birth to a 250 watt power represent, you give birth to to watch them down to
225 watts if you give birth to seven or else eight cards in the sphere of the
node. If you solitary position six in the sphere of, you can run these grown-up
cards by their gorged 250 watts.
The spec sheet does not reference the Xeon
Phi x86 coprocessor, but HP's revelation in support of the member of staff
serving at table node says with the aim of the Xeon Phi command be present an
option primitive subsequently time.
To supplementary shed light on, HP tells El
Reg with the aim of it command not support at all actively cooled coprocessors,
which income folks models with their own fans. So with the aim of income the
Xeon Phi 3120, which has a 300 watt power envelope and a fan on the wall like a
graphics tag is not free to be present supported. This is the cheaper $2,000
sort. But the passively cooled Xeon Phi 5110P, which sits in the sphere of a
225 watt thermal envelope and which expenditure $2,649, command be present
supported in the sphere of the SL270s Gen8 node.
HP long-established with the aim of both
the Tesla K20 and K20X GPU coprocessors, which are both passively cooled,
command be present supported in the sphere of the SL270s Gen8 node.
It would be present supplementary
attention-grabbing, of direction, if HP may well support actively cooled GPUs
inside the box, such having the status of the Quadro K5000 graphics tag from
Nvidia, or else the hottest FirePro and Radeon cards from progressive Micro
policy. By the very smallest amount, it would be present attention-grabbing if
AMD made passively cooled variants of the hottest FirePro tag, the FirePro
S10000 double-GPU cards announced only remaining week.
With the aim of FirePro S10000 is entirely
competitive with the Tesla K10 and K5000 cards in the sphere of provisions of
freezing single-precision computing, delivering 5.91 teraflops, and it is
besides able to import 1.48 teraflops by stand-in precision. The Tesla K10 has
4.58 teraflops SP, but solitary 0.19 teraflops DP and it cannot be present used
having the status of a graphics tag.
The K20 and K20X cards from Nvidia offer
1.17 and 1.31 teraflops by DP and 3.52 and 3.95 teraflops by SP, but again,
cannot organize graphics. At the bottom of actively cooled graphics cards would
allow the member of staff serving at table node to be present used having the
status of the back-end in support of visualization walls.
The SL270s Gen8 node has eight PCI-Express
x16 peripheral slots, plus lone other low-profile x8 slot and lone x8
FlexibleLOM mezzanine tag in support of linking the node unfashionable to
Ethernet or else InfiniBand networks. (LOM is brusque in support of LAN on
motherboard.) The member of staff serving at table node has eight DDR3
reminiscence slots apiece socket and maxes unfashionable by 256GB of
reminiscence using 16GB reminiscence sticks.
The Xeon E5-2600 processors with 60, 80,
95, and 115 watt thermal envelopes are supported in the sphere of the node; if
you require to benefit 130 watt parts, HP says to chatter to someone
originator. The system has a center dual-port Gigabit Ethernet welded against
the mobo, plus with the aim of FlexibleLOM slot. The thing has space in support
of eight 2.5-inch disks in support of confined node storage space.
The SL270s Gen8 member of staff serving at
table command transport in the sphere of December and command give birth to a
center slant value of $6,166 in support of a system with lone eight-core Xeon
E5-2660 rotating by 2.2GHz, 8GB of reminiscence, and a Smart Array B320i CD
controller. The system is not yet in the sphere of the online configurators by
HP, so we can't gin up a proper configuration and expose you what did you say?
It may well cost.
