Intel's future Sandy Bridge Xeons exposed The ongoing x64 server chip war is going to heat up considerably in the third quarter, with both Intel and AMD firing off new processors for midrange server buyers. And we now have a few more details about Intel's future "Sandy Bridge" Xeon platforms and the chipsets that drive them. Processor socket and chipset transitions …
Quote:
"Like what AMD had in late 2009 with the delivery of its first home-grown chipsets – the "Fiorano" SR56X0/SP5100 combo – and the Opteron 4100s in their C32 sockets and the Opteron 6100s in their G34 sockets, which make use of these chipsets, coming out the following spring."
The first AMD home grown chipset is AMD 760 which shipped with Athlon MP and allowed AMD to enter the SMP market. It had one of the most entertaining erratas I have ever seen (at least on MSI motherboards). System was going bonkers if you unplug the PS2 mouse :) That was in 2002 or thereabouts. Nearly 10 years ago.
I understand they hand out press releases but there's no need to smoke them up and inhale so deeply, while washing it down with Intel Kool-Aid.
I don't see Intel's slicing up of the chipset family as a benefit to anyone but Intel - OEMs can build systems with varying storage options without needing multiple silicon parts. It's Intel's attempt to grab a larger chunk of the price of higher cost servers. Nice to see as many as 8 x 6Gbps SAS ports, I believe AMD offers 6 x 3Gbps SATA at present, requiring additional chippery many times.
C32 only seems to make sense in low-end single socket, but with Bulldozer it offers 8 cores against the 4 Intel offers in the E3 Xeons. AMD looks to be equal/ahead in memory bandwidth per socket with G34 vs Intel (4 channels vs 3 or 4) and a bit behind in ultimate capacity (8 DIMMS vs 9 or 12) The AMD chipsets have as many or more PCIe lanes.
Intel will give you 8 cores in an E5 or 10 in a pricey E7, AMD offers 16. Who can say about performance until things are released, but AMD's CPU prices are relative bargins.
But the 10-core intel chips you refer to are westmere based, not SB and hence not (yet) existing in the SB range.
Furthermore, (if I got the slides correctly) there's no plan for a SB 10-core variant given that the chipset supports up to 2x8-core chips. And that's on socket 2011 which is not (yet) released in any product segment.
Care to enlighten me, or have I been drinking too early today?
The just-launched E7 xeons (westmere-ex) are Sandy Bridge, I believe, or the newest microarchitecture, in any case. (I can't always keep track of codenames for architectures vs. implementations). Or am I wrong?
I was strictly keeping things on a per-socket basis, to avoid complete madness. But AMD will give you 32 cores in a mainstream 2-socket system and it looks like Intel will give you 16 or 20 cores which are faster by some as-yet unknown fraction.
In terms of FP, which is what I care about most about it looks like Bulldozer cores are equal, clock for clock, with the newest Xeon cores on SSE2 code and somewhere closer to half as fast with AVX, when that becomes important.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
