Slashdot Mirror
Linux Shootout: Opteron 150 vs. Xeon 3.6GHz Nocona
danalien writes "Anandtech with their previous review have stirred up a bit of controversy, and they've released their follow-up review where they pit AMD's Opteron 150 vs Intel's Xeon 3.6 Nocona (on linux)."
Athlon 64 is the name used for the desktop line, and Opteron is the name used for the server/workstation processors.
Athlon64 3000+ (2GHz): $167
Pentium 4 3.4GHz Extreme Edition: $1025
Provided you have a NUMA-aware operating system, that is. The OS needs to know which memory is attached to which processor, since access to memory attached to the same processor on which a thread is running will obviously be faster and lower latency than going across hypertransport to a different processor and waiting for an answer.
After all is said and done it became difficult (nearly impossible?) to justify the Xeon processor in a UP configuration over the Opteron 150,
Huh? Here are some numbers:
- POV-Ray 3.50c: Opteron is 40% faster
- Crafty v19.15: Opteron is 70% faster
- TSCP: 10% faster
- PostgresSQL test-insert and test-select: Opteron takes 60% of the time it takes Xeon
- MySQL test-insert: Opteron takes 80% of the time it takes Xeon.
In almost every benchmark, where proper optimizations are used (and why shouldn't they be? Who in his/her right mind would not use proper optimizations??), the Opteron destroys the Xeon.There are older dual and quad Opteron vs Xeon reviews around.
Humorously, the also say this:
Now we know that the Nocona is here, and it's getting slaughtered at the Altar of The Opteron.
Belief is the currency of delusion.
How do they know? By cache coherence signals transferred between the CPUs. This isn't free and consumes bus bandwidth.
The first CPU can't "instantly" write the value either, because it must first obtain exclusive ownership of that cache line by checking with the other CPUs.
On the Opteron architecture (we call this NUMA, or "point-to-point"), as soon as one CPU writes a value to the 'GO!!' area, well, that's _just the beginning_. It has to tell another CPU in the system that it just did that. etc etc
It has to use some communication resource to update the other CPUs on the state of that cache line. Just like the bus-based situation.
This still leaves me wondering why an Opteron 250 (2.4GHz, 1MB L2 cache) seems to so seriously outperform an Athlon 64 3500+ (2.2GHz, 512KB L2 cache).
When people says that the first article was bad, it's because it was really bad: 64-bit binaries for Intel vs. 32-bit binaries for AMD, copy&pasted benchmark results from previous 32-bit benchmarks, tests (PI digit computation) that measured the libc optimization instead of the actual benchmark (when removing the printf() it got about a 10x boost). People on aceshardware forums were posting TSCP scores about 2x what Anandtech got, on the same processor. So the A64 3500+ scores you saw in that article are trash. Forget them.
I wouldn't touch VIA and SIS with a ten foot pole for my own systems, even less for servers. Plenty of bad experiences.
While in the past VIA and SIS have been really bad chipsets, modern VIA chipsets (KT266A and up) are rock stable and perform well. I have had both KT333 and KT600 boards which have never failed. SIS, while I have no first hand experience, I am told is similar.
Perhaps this is what you are looking for
The design intended to become the Pentium 5 (Tejas) was cancelled in favour of Pentium M derivatives. Intel basically had to give up on the Netburst micro-architecture and is now concentrating on increased parallelism (multiple cores) rather than extreme clock rates.
I bought a board with a SiS 745 chipset and it has been perfect, at least under windows. They provide a nice dual-channel PCI bus, even. SiS used to be a horrible joke but they've come a long, long way.
"You're right," Fisheye says. "I should have set it on 'whip' or 'chop.'"