Opteron fairs well, according to Anandtech...

L

Luminescent

Veteran
I urge you to Explore Anandtech's performance evaluation of the Opteron at 1.8GHz here. It exceeds the performance of a 3.0MHz P4 in many of the benchmarks, with only a 1.8GHz core.

Remember, the P4's SSE2 unit can execute 2 double precision ops and 4 single precision ops; it executes just as many ops per cycle as the Hammer, yet it manages to exceed opteron's performance on SSE2 apps by a small margin (with a ~30% clockspeed advantage and performance tuned chipset). This is an impressive feat, as are the results of the the results of the opteron on the other tests.
 
Yeah I was pleasantly surprised. Now they just need to get the frequency up.

Athlon64 is actually worrying me. It'll have HALF the memory bandwidth of the opteron (a single 64bit memory interface). I find that in conjucntion with the reputed clock scaling problems worrisome. :(
 
Many sane minded people have been worried about clock scaling and rightly so. The 2 stage increase is offset by the increased logic and I doubt AMD has better process technologies than Intel that will allow it to scale clock rate via that avenue. I've said it before, expect a new Hammer iteration soon, likely with an extended pipeline to improve frequency scaling.

Half the bandwidth is another concern, lower latency will only improve bandwidth efficiency. The question is will this be enough to compete against the newer P4 iterations and the newer chipset iterations, which support higher frequency FSB and RAM? Ultimately, I don't think they'll be a clean winner, they'll probably trade blows, as each will win with various applications. I bet the application X has been optimised for processor Y talk will be at an all time high.

In anycase, the P4 SSE2 throughput is rougly half speed, so the P4 -- IIRC this is for most operations, see the optimisation guide or CPU spec sheet. I wonder if this will change in Prescott?
 
I had the same reaction to the Anand benchmarks over in the Ace's Hardware forums...until Johan put me in my place. He's now convinced me that the gaming benchmarks are benchmarking not the CPUs but rather the various platforms' PCI bus implementations.

This shouldn't be a huge surprise given the configurations: some of the fastest CPUs in the world, a video card very capable of performing quite well on DX7 games at 640x480, all over the PCI bus. The Opteron's excellent performance is then easily explained by its direct HT link to the Southbridge, while the other platforms have to go through the Northbridge first. The kicker is the almost complete lack of any performance difference between the AXP 3000+ and the AXP 2200+...and these are supposed to be benchmarking the CPU? Fat chance.

Now, this isn't to say that Opteron/Athlon64 won't be great gaming CPUs--XbitLabs' tests of a beta A64 suggest that already, and Opteron's 128-bit memory bus will help even more. It's just to say that Anand's benches don't necessarily have any bearing on reality.
 
They fare fairly well, but Anand needs to test with a PCI card other than a GF4MX (as Mloot shows in his excellent threads, which I linked in 3DGB&N), IMIO*.

* In My Ignorant Opinion.
 
You must log in or register to reply here.
Back
Top