A flagrant act of benchmarking
Now that you have the basics on the Opteron x50 series, I have to cover a few items before we get into the benchmark results. Benchmarking a workstation chip like this is akin to navigating a minefield, and I need to explain why and how we picked our way through the process.
First and foremost, we tested the Opteron 150 and 250 as workstation processors, not as server chips. Perhaps next time out we'll consider doing a suite of server-oriented benchmarks, but PC-based workstations are a more common part of our coverage around here. Because the definition of "workstation" can be a little slippery some times, I just elected to go ahead and test the Opteron x50 series against a whole range of other PC processors. Among those twenty-some competitors, you'll find the Opteron x50 predecessors, the x46 and x48 models. As you look through the results, remember that the numbers labeled "Opteron 150" represent single-processor results, and the "Opteron 250" results are from a dual-processor config. The same goes for the x46 and x48 series.
The purists among you will note that we did not use a "workstation-class" graphics chip in our tests. That is to say, we didn't use a workstation-class graphics driver, because the chips are generally the same these days. Unfortunately, though, our card said "Radeon" rather than "FireGL" on it. The Market Segmentation Police have already been notified, no doubt, and I expect to be holed up in Damage Labs in a tense standoff most of the day today.
I committed further crimes when I chose the MSI K8T Master2 motherboard for testing the Opteron processors. The Master2 only has DIMM slots hanging off of one of the two CPU sockets on the board, cutting off the second processor's ability to talk to a pair of DDR400 DIMMs via its built-in memory controller. Tyan's Thunder K8W has the magic combination of an AGP slot and a "full" dual Opteron memory configuration.
In my defense, there are some good reasons why I committed such heinous acts. The Thunder K8W showed us its eye-popping memory bandwidth potential in our recent Opteron workstation motherboard round-up, but it was only able to realize that potential with the 64-bit beta of Windows XP. Right now, there is no workstation-class version of Windows available, outside of early beta versions, that understands the non-uniform memory access (NUMA) configuration of Opteron systems, so much of the K8W's memory bandwidth goes to waste. This reality, combined with the fact that the K8W doesn't tolerate aggressive memory timings well, caused the Thunder K8W to lose out to the K8T Master2 in many of our performance tests. So I decided to go with the K8T Master2 for our purposes. Of course, future versions of Windows with NUMA awareness and 64-bitness should give dually Opteron systems a big boost over their present incarnations. Had we more time with the Opteron 250 prior to its launch, we might have been able to test with a 64-bit version of Linux on the Thunder K8W. No doubt that could be scary fast.
One more thing before your attention span snaps back into place. The MSI K8T Master2's memory config isn't exactly uniform, either. The CPU connected to the DIMM slots can access memory faster than the second CPU, which has to talk to memory by traversing the HyperTransport link between the two Opteron processors. As a result, benchmark results for dual Opterons on the K8T Master2 sometimes look a little odd. They will vary depending on which CPU the OS happened to assign the benchmark program. Sometimes that means the Opteron 246 will come out with a higher score than the Opteron 250, just based on the vicissitudes of CPU scheduling. Also, the single-processor Opteron systems are sometimes faster when applications aren't multithreaded. We've found this to be a consistent characteristic of multiprocessor Opteron systems, but it's no great detriment, as you will see from the overall performance numbers.
Sizing up the competition
We tested the Opteron 250 against its closest competitors, a pair of Xeon 3.2GHz processors. The Xeon CPUs are based on the Prestonia core, essentially a Pentium 4 Northwood with 1MB of L3 cache bolted on, plus a different CPU package and pinout. Think of it as a Pentium 4 Not-So-Extreme Edition. (Intel has versions of the Xeon 3.2GHz with 2MB L3 cache, but they tend to be pricey.) Trouble is, Intel hasn't updated the Xeon with the latest goodies for quite some time. The Xeon 3.2GHz still has a 533MHz front-side bus, and its official chipset is the E7505, which supports two channels of DDR266 memoryand nothing more. Renegade motherboard makers have produced Xeon mobos based on the 875P chipset with DDR400 support, but the overwhelming majority of Xeon 3.2GHz workstations use a 533MHz bus and DDR266 memory.
Intel also considers its Pentium 4 processor, combined with the 875P chipset, a workstation-class product, and I'm inclined to accept that. I can't see any good reason to buy a single-processor Xeon workstation when Pentium 4s are available at higher clock speeds with a faster bus, dual-channel DDR400 memory, and Serial ATA RAID built into the chipset. Still, I've included single-CPU Xeon results for completeness.
Connecting the dots
The line graphs you will see in the following pages may be familiar from our review of the Opteron 150's evil twin, the Athlon 64 FX-53. If you don't know what they are all about, I suggest reading that review for an overview. It explains why we organized the graphs as we did. For clarity's sake, let me refresh your memory about a couple of things, though.
The Pentium 4 "E" processors in the line graphsand in the bar graphs, for that matterare based on the 90nm Prescott core. The other Pentium 4 chips, including the "C" variants, are based on the 130nm Northwood core.
Also, you will not see any Athlon 64 FX processor scores in this review. That's because, for most intents including benchmarking, the Athlon 64 FX is identical to the Opteron 100-series. When you see a score for an Opteron 148, think Athlon 64 FX-51; they're both 2.2GHz AMD "Hammer" processors with dual channels of DDR400 memory. When you see an Opteron 150, think Athlon 64 FX-53.
|Some 840 EVOs still vulnerable to read speed slowdowns||36|
|Details leak out on AMD's first Zen-based desktop CPUs||48|
|Nvidia: the GeForce GTX 970 works exactly as intended||77|
|Report: 4GB of RAM coming to GTX 960 in March||97|
|Early deal of the week: A 27" G-Sync monitor for $480||34|
|Gearbox's Homeworld remake due February 25||43|
|Nvidia admits, explains GeForce GTX 970 memory allocation issue||240|
|Here's my guest appearance on tonight's Alt+Tab Show||12|
|HA. AMD in the red and nVidia in the green. Thats funny cause you know... *cough* oh forget it.||+82|