This comment is pure BS:
"However, Tim has sought to keep ScienceMark independent by diversifying the development team and by publishing much of the source code for the benchmarks at the ScienceMark website."

almost zero source code is available.

You know what's really missing from this review?

Where's the FUD graph!? ("Folding Units per Day," why, what did you think I meant?) (Pipe down, gerbs... uniform WUs across platforms, natch.) I mean, seriously, how can you possibly claim to show real-world application benchmarks without showing what is using 90+% of the CPU time on most of the readers' systems here?

AMD needs the dual channel non registered version of the FX51, imo, AS WELL AS a PCI Express chipset(s) to pair with it. Its going to be a matter of months or less before this will be the minimum standard.

If, as it seems to be, Win 64 is seriously delayed, the concept to be wary of is not Win 64 (even though that's where the eyes are looking), but rather in what kind of a delay that could also put into a 64 version of Longhorn. It would be frustration beyond belief, imo, for people to have nice 64 chips, only to be left out in the cold as the rest of the world maybe spends a year or more in the Longhorn environment. Longhorn is "more complicated" and is taking long to produce. Win XP 64 seems to be taking years. I think you begin to get the picture... UGLY!

However Intel is about to lower the IPC of their processors by lengthening the pipeline. Remember the P3 to the P4 was a shift of 12 slot pipeline to a 20. Now we are making the move to 30, or so it sounds like. This is in the "hopes" that we get greater gigahertz out of the chips, but remember the initial versions are likely to be slower than the 3.0 P4 we currently have. The ringer will be PCI Express chipsets, which will intrigue folks the more we learn. The downside there will be DDR2, which will promise high datarates, but the penalty in latency will be, like the pipeline, reminders of a year or two ago. The more we move ahead, the more often it sometimes looks like we move backwards. Additionally, for a real kicker, I have a strong suspicion the math units in the Prescott, can no longer be run at double the CPU speed. That has been one of the saving graces to keep the P4 in the running in math intensive stuff (games, Excel, etc) above its SSE2 abilities. This won't get talked about much if it happens, but the early benchmarks comparing Prescott 2.8 to P4 2.8 hinted at it. Its a guess but I have a strong feeling from seeing some of those that the SSE3 is faster than SSE2, but that the math units in Prescott have to run at core speed now. The probably belief is that so much software is SSE2 optimized now it doesn't matter right? hehehe If you go Prescott in 2004, I think you might have some serious thinking to do before you do it.

Its looking like you'll probably want a DDR1/PCI Express/Prescott or P4EE if you are wanting to be Longhorn bound,

and the 939 non reg mem FX (and PCI Express) if you care more about raw performance. Win 64 I worry could VERY easily shift into a late 2004 or early 2005 item, and Longhorn 64 become a 2006/07 item.

When the pipeline goes up, and the 939 FX comes out I predict the performance is going to start tilting massively (even in 32 bit mode) to the FX chip.

Anyway many changes in 2004, but about half are not as good as the marketing dept's would have you believe... Isn't it always that way ?

:(

Tomshardware review starts off good, but the rest makes me vomit. He hand picked the test after the more popular benches.

good review... but i do have one thing to bitch about. I hear people talking right and left (not necesarily here) about how the processor is useless because there is no 64bit OS (other than linux) to support it. People are not buying it because of a lack of OS support. Does this drive anyone else freaking nuts? If AMD DIDNT announce this so strongly as being a 64bit processor, it seems like theyd be selling MORE chips.

Damn morons.

I am an Anonymous Gerbil----and I shall NOT go away, so frig you-----you elitist "I have-a-name" Bastids !!

You hear that, Scott?????

We AG's rule !
We AG's iz the Future!

Bit expensive for a platform with a socket that'll be obsolete soon.

Although Intel's the big bully in this story, i do hope that they wake up and answer with something as significant. 64-bit significant, not "pull out the next batch of Xeons and rename them" significant.

I don't think they are in any position to do that at the moment. They are probably still working on launching prescott. And once they have that done, and move it to a new stepping, they will jack up the clock speed like crazy.....or add 2 megs of cashe to all there processors.

I wouldn't be suprise they are working on something like Yamhill...

What company in its right mind is going to produce S754 boards for less than a year and then have the entire socket die on them?

BTX, PCI-Express, Socket 939, DDR2....
Too much stuff happening, too many standards changing (i don't really mind the RAM nor the socket for its pin count, but the PCI/BTX/Dual-Channel are significant.

Although Intel's the big bully in this story, i do hope that they wake up and answer with something as significant. 64-bit significant, not "pull out the next batch of Xeons and rename them" significant.

I hope this forces Intel to do *something*. Those lazy bitch have become... very lazy.

Well I've pretty much cemented in my mind that an A64 will be powering my new system this spring. Hopefully socket 939 will happen sooner rather than later because I'd really like to wait for that but damn is the 3400+ looking sweet to me.

Tom does not turn anything in Intel's favor.
Amd better for games, Intel better for video encoding.

End of story. All the sites have benchmarks that show the same thing.

If you can not accept that sorry for you, Amd fanboy.

I have nothing against Amd or pro Intel, it all depends what use you are going to make of your computer.

I LOVED the review Damage, I like how you went into all the different types of 3d rendering apps. It's great to see you still including different speed P4's and AthlonXPs in there. I got one gripe tho...Really, is that small picture of the 3400+ really nessesary on the front page? It's kinda tacky.

>Interesting. I had expected the 3400+ to be a fair bit slower than the
>Athlon 64 FX-51, because matrix multiplication generally requires quite >a bit of memory bandwidth

Not true. xGEMM Performance is usually only CPU limited.

What sort of temps do these processors typically run at idle and at load?

I really wanted to see the 3000+ in the comparasions. I think it would have meant a lot to people waffling on processor choices. Also, the news about the 754 dying is making me think a lot harder about OPtereon since the 940 wouldn't die as fast.

I'm a little suspicious of that ScienceMark "Blas DGEMM" benchmark. It seems curious that all the codepaths for the A-64 give exactly the same result. In fact, it argues that something else is going on. My guess would be you're seeing numbers that are cache-limited in some way. If that's the case, then this might be one test where the A64/3000 will show a significant difference in scores vs its bigger brothers.

Good article!

Aww, I think the little pic of the 3400+ is cute. I'd leave it!!

Looking at this, I have come to the conclusion that dual-channel Socket939 type that won't be needing registered memory would be mind-blowing in terms of performance. The unbuffered advantage of 3400+ with dual-channel prowness of FX51 is a winning formula. Looking at how cache isn't that much of a problem with 3200+-3000+ comparison, even a 512KB version would fare well, I would like to believe, if supplied with good dual-channel unbuffered goodness.

Oh, man. I'm SO gonna get A64 in summer.

Is there a typo in the Science Mark MolyDyne graph on page 13? The XP 2800+ seems to be running awfully high on the chart and the A64 3200 seems to be down below the XP 2500+.

Good one

AMD is stacking there arsenal of CPUs

The article has been updated to reflect the Athlon 64 3400+ launch price of $417. AMD sent me the wrong #@^&! price. I've also revised my conclusions based on this new info.

no GNU/Linux tests? I want to see what these things can do in 64 bit mode!

Hey Damage, quick thing I noticed. The 3400+ is actually priced at $417. You were off about $200 :)

http://www.amd.com/us-en/Corporate/VirtualPressRoom/0,,51_104_543,0...
http://www.amd.com/us-en/Processors/ProductInformation/0,,30_118_60...

w00t...great review...and one "hot" lil chip...hot damn :)

Good work damage! Keep it up.