|
|||||||
|
|
![]() |
|
|
AMD Steps into the 64-Bit Ring with Opteron by Timothy Prickett Morgan It's taken a few years, but Advanced Micro Devices has finally launched a 64-bit alternative to Intel's Itanium processors, with its "SledgeHammer" Opteron chip. While there will be demand for 32-bit Intel and clone processors for many years to come, big jobs will inevitably require 64-bit computing. AMD's Opteron, which can run both 32- and 64-bit applications, may give Intel lots of grief if it takes off in the server space because Itanium can't run 32-bit X86 code and has its own instruction set. AMD has been banging at the corporate door for years, and after trying to get onto the corporate desktops with its Athlon chips, it has figured out it is easier to get into shops through the data center. Athlon processors are increasingly popular as platforms for running technical workloads on Linux clusters, and the Opterons, because of the substantial architectural improvements AMD has made with the Hammer chips and its associated chipset, are going to be quite attractive. Linux support will be available immediately, but because few commercial applications are running on Linux, the impact on commercial users will be less impressive than if Windows and Unix were supported on the AMD chip. Windows Server 2003 will be available for the chips in the Standard and Enterprise Editions in beta by mid-2003, and will probably be in production by the end of the year or early next year. The word on the street is that AMD and Microsoft are working on a version of Windows Server 2003 Datacenter Edition for Opteron-based servers, but we hear this will not be ready to roll out in production environments until mid-2004. (The 32-bit versions of Windows 2000 will apparently work on the boxes, however.) That a Datacenter Edition exists on the Opteron machines is surprising, because the Opteron and its HyperTransport interconnect only scale to eight-way symmetric multiprocessing right now. The AMD Opteron chip/8000 chipset combo is based on two-way cell boards that are clustered into an eight-way machine through a four-port HyperTransport hub. This is very similar to the design IBM has for its "Summit" chipset for the xSeries 440. However, IBM is using four-way cell boards in the xSeries 440. The existence of a future Datacenter Edition for Opteron machines seems to indicate that AMD is thinking of building bigger boxes, perhaps with 16 or 32 processors in a single-system image. AMD could get to 16 processors by moving from four to eight ports in the HyperTransport switch or by moving to four-way cell boards. If AMD did both at the same time, it would have a 32-way machine. AMD has never officially given out clock speeds or prices, but in presentations concerning the performance of the Opterons, the company provides charts that show it available in 1.2 GHz, 1.4 GHz, 1.8 GHz, and 2 GHz. Some press reports indicate AMD will debut Opteron at 1.6 GHz and 1.8 GHz; others say a 2 GHz port will also debut. AMD refused to comment on the Opteron chip before the announcement, which took place yesterday in New York. The chip includes 128 KB of L2 cache and 1 MB of on-chip L2 cache. The Opteron design also includes an on-chip DDR-SDRAM memory controller--an industry first and the source of many performance benefits. This allows the memory controller to run at the same core frequency as the Opteron processor (off chip memory controllers typically run at half speed or slower). And with each chip having its own memory controller, memory bandwidth scales as Opterons are added to a processor complex. This is a big deal. Each processor can have eight DIMMS, and that means 8 GB of main memory per processor using today's DRR-SDRAM technology. That gives an eight-way Opteron box 64 GB of main memory and 5.3 GB/sec of memory bandwidth to play with. Here's why that matters: The performance specs I have seen indicate that, clock-for-clock on the SPECint2000 and SPECfp2000 processor benchmarks, the 64-bit Opterons have about a 30 percent performance advantage on number-crunching jobs and about twice the performance on integer work, compared with the "Prestonia" Pentium 4 Xeon DP processors. That's a big performance boost, one you would expect when comparing 64-bit applications to 32-bit applications. Servers based on the Xeon DP and Xeon MP processors from Intel are what AMD seems to be targeting explicitly. But, implicitly, the Opteron is also going to go after Itanium. If the 2 GHz Opterons sell for under $800, as expected, and can deliver about the same performance as the future "Madison" Itanium 2 chips (as I estimate they will be able to do on commercial workloads), how will Intel be able to sell Madisons in all but the largest machines that scale beyond AMD designs? Intel will have to slash Itanium 2 prices from their current high of several thousand dollars a pop just to keep AMD out. Anyone looking for an eight-way, 64-bit server running Windows or Linux in late 2003 or throughout 2004 will have to take a look at the AMD alternative. This prospect can't be making Intel happy. While none of the big server makers has committed to supporting the Opteron and its 8000 chipset, what is obvious is that vendors that hate Intel's Itanium but want to extend their reach into the X86 market would be wise to endorse Opteron, even if they don't push it hard. Hewlett-Packard has thus far bet all of its chips (so to speak) on Itanium, but it is remotely possible--although unlikely--that HP will support Opterons in ProLiant or rx Series servers running Linux or Windows 2000 and, later, Windows 2003. Sun Microsystems, which hates Intel and Microsoft, quietly endorsed the Opteron platform before the launch, and it intends to run its 32-bit Solaris and Linux platforms on the AMD processor and is evaluating the possibility of using the X86-64 extensions. Anything that hurts both Intel and Microsoft at the same time is going to attract Sun. IBM appeared on stage with AMD in New York and vaguely committed to supporting the platform because of the manner in which 32- and 64-bit applications can run on the same processor. IBM's exactly plans were unclear at the Opteron announcement, but it is clear that IBM does not want to endorse Itanium, because of the credibility it gives to HP's strategy of using that chip. Anything that hurts HP is going to attract IBM, even if it does hurt Intel. Dell has been mum on its plans for AMD processors, but Dell, like IBM and Sun, is as loyal to its own bottom line as it is to its partners Intel and Microsoft. If it makes sense for these companies to launch AMD-based servers, and if this can give them leverage with AMD and against Intel, you can bet IBM, Sun, Dell, and maybe even HP, will not let the white box server makers rule the AMD server market for very long. Next week, we'll take a look at the new Opteron servers and compare them with Intel-based machines.
|
Editor
Contact the Editors |
| Copyright © 1996-2008 Guild Companies, Inc. All Rights Reserved. |