AMD claims no premium for four-way chips
The Opteron 6100 sales pitch
With Intel finally on par with Advanced Micro Devices in terms of CPU performance, memory bandwidth and basic server architecture, AMD is left to keep even or slightly ahead on performance and compete aggressively on price.
And with the launch of the "Magny-Cours" Opteron 6100 processors today, which make use of AMD's own chipsets to make server platforms, AMD is intent on hitting Intel in the bottom line, and maybe market share, as it prices the Opteron 6100s lower than many might have expected. Particularly AMD shareholders.
But AMD has little choice but to lower prices because with the rounding out of the Xeon product line - the "Westmere-EP" Xeon 5600s announced two weeks ago and the "Nehalem-EX" Xeon 7500s due tomorrow - Intel will have chip lines with lots of cores and memory bandwidth with which to compete on feeds and speeds against whatever AMD has.
The Nehalem-EX chips will sport eight cores plus HyperThreading, Intel's implementation of simultaneous multithreading. It will have integrated DDR3 memory controllers on the chip and use Intel's homage to AMD's HyperTransport interconnect, called QuickPath Interconnect, to make servers with four, eight, or perhaps more sockets.
The Xeon 5500s that debuted almost precisely a year ago were Intel's first processors aimed at two-socket machines that deployed QPI and offered between three and four times the memory bandwidth of that stale old frontside bus architecture.
The Xeon 5500 and the new Xeon 5600s, the latter of which come in variants with four or six cores, make use the "Tylersburg" 5520 chipset, while the Nehalem-EX and quad-core "Tukwila" Itanium 9300s (announced in early February) use the "Boxboro" chipset that Intel is only now putting into the field. Intel has three distinct server processors using two distinct chipsets.
AMD, by contrast, is trying to chop up the market in a slightly different way, and perhaps cover the needs of server makers better in the process. The exact specs of the SR5600 series chipsets used with the Opteron 6100s and the future "Lisbon" Opteron 4100s (for entry servers) have not been divulged yet, but they are probably very similar to the chipsets AMD announced last September to work with the "Istanbul" six-core Opterons.
The chipset includes the SR5690, SR5670, SR5650 I/O hubs (which does 'IOMMU' I/O virtualization technology as well as reaching out to PCI-Express 2.0 peripheral slots) paired with the SP5100 southbridge, which links to USB and SATA ports and provides links to legacy PCI slots of mobo makers want to add them in.
The three different I/O hubs, as El Reg explained last summer, different in how many PCI-Express engines are on the chip, how many PCI-Express lanes they deliver, and how much juice the I/O hub burns. The SR5600 is a single chipset with three different dials, unlike Intel's Tylersburg and Boxboro, which are very different animals.
Ditto for the Opteron 4100 chips that will come out in the second quarter and the Opteron 6100s coming out today. The Opteron 4100 is a tweaked Istanbul core that has its on-chip memory controller switched to DDR3 memory and that has an extra HT3 port so each processor in a four-socket complex can talk directly to the others. It fits into a modified Rev F 1,207-pin socket. The Opteron 6100 is basically two Opteron 4100s in a single package, plugging into a new 1,944-pin G34 socket.
The older Opterons had two memory channels per socket, as the Opteron 4100s will have as well, while the Opteron 6100s will have four channels per socket, offering lots of memory expansion. Servers with one and two sockets using the AMD SR5600 chipsets and the Opteron 4100 processors will compete with the low-end of the Xeon 5600 2P market and any aspirations Intel might have for single-socket boxes based on crimped Xeon 3600s or Core i5/i7 processors from desktops.
The Opteron 6100s will compete with the high-end of the Xeon 5600s in the 2P space and also take the fight on up to the 4P space. But, AMD's chipsets and the chips themselves are really all the same. It is really a game of packaging some components in the stack up in different ways to target different markets.
AMD does it again, AMD rocks!
As a NASA system integrator for 10 years, they have never bought a Intel server from me. They realize that AMD gives them more performance for the dollar! Keep up the good work AMD!
8 Cores? 12 Cores!
# Twelve-core and Eight–core, L2 – 512K/core, L3 - 12MB of shared L3 Cache
# AMD CoolCore™ technology, AMD PowerNow!™ technology, Enhanced C1 state, AMD CoolSpeed technology, APML (in APML enabled platforms)
# Quad-Channel LV & U/RDDR3, ECC, support for on-line spare memory
# Supports up to 3 DIMMs/channel, up to 12 DIMMS per CPU
Above From AMD's website. A dual processor machine with 24 cores is a very interesting piece of hardware and miles ahead of Intel. Moreover, the AMD architecture runs more instructions per clock cycle than an Intel and Intel's Hyperthreading is nothing more than a gimmick.
"A machine with two of the top-end 3.33 GHz Xeon X5680 six-core processors is rated at 381 on the SPECint_rate2006 integer benchmark and 257 on the SPECfp_rate2006 floating point test; with 48 GB of memory, that machine costs around $3,350.
A reference AMD platform with four sockets using the eight-core 6136 processors running at 2.4 GHz is rated at an estimated 574 on the integer test and an estimated 514 on the floating point test. With 128 GB of memory, that four-socket reference server is expected to cost just under $3,000. "
I think that is totally ok performance. Dont you agree?