Engineering Sample of A64 Benchmarks

[pIII_Man]

i guarantee in realwold apps...a overclocked tualatin will kill any sdram p4!

If i put my k6 in a Ferrari it would be faster than your your pentium 4 or Athlon XP :tongue:

 

[Kelledin]

do you think applications will be easier than drivers to make 64bit? is that just a question of recompiling, or does it require re-writing from scratch in 64bit?

Applications will generally be easier to port. The 64-bit cleanliness hurdle still applies when making userland applications stable, but the PCI memory mapping quirks shouldn't matter in userland. Everything else (NUMA and extra register usage) is really optional performance tuning.

Optimizing register usage is generally a compiler issue. AMD's NUMA implementation is slightly tricky for multithreaded apps, because the more multiple threads share data, the more traffic will have to go between the inter-CPU HyperTransport links--which incurs a bit of a performance hit.

also, i am almost cerain the answer to this is yes, but, if (or when) intel eventually releases a 64bit Pentium, will it be 100% compatible with 64bit drivers/applications/OS' etc. that have been written based on the Athlon 64?

Well...that's a good question. I can think of two scenarios:

1) AMD64 really catches on before Intel can get a Yamhill(ish) implementation going. In this case, Intel will probably (wisely) make a set of Yamhill extensions 100% compatible with AMD64. W00t!

2) Intel manages to get their Yamhill(ish) implementation going before AMD64 really catches on. In this case, Intel can make theirs just slightly incompatible with the AMD64 instruction set, and due to Intel's dominant marketshare, AMD will be cut out of the 64-bit loop. Even if the AMD64 ISA manages to hang on, we'll be stuck with two divergent, incompatible 64-bit x86 ISAs. Suck! :frown:

I'd consider (1) to be the most likely outcome. (2) is still possible, and there are even rumors of 64-bit Yamhill(ish) extensions in Prescott. Deciding whether Yamhill would be compatible or incompatible with AMD64 will probably be a simple matter of microcode changes, so Intel could make that decision even after committing to a Prescott tapeout.

<i>I can love my fellow man...but I'm damned if I'll love yours.</i><P ID="edit"><FONT SIZE=-1><EM>Edited by kelledin on 08/12/03 01:06 AM.</EM></FONT></P>

 

[wolverinero79]

Just a few notes:

A.) Grabbing "benchmark data" from AMDZone is worse than believing everything that comes of The Inquirer. I think it's funny in the mentioned benchmark link that they briefly mention the lower amount of memory for the P4 and Athlon systems and add this blurb

"The P4 and Athlon system were run with Cas latency at 2 while the Opteron was only able to run at Cas 2.5 which is a distinct disadvantage for it"

While there may be a little disadvantage from this, i think doubling the memory (at least...they don't give the number of 256 memory sticks used on the P4/Athlon machines) MORE MORE MORE MORE than compensates for this. This site could have gone even more rediculous by using DDR266 and i'm actually surprised they didn't.

B.) I think the reason Yamhill is still so far out (what do the rumors say 2005-2007, somewhere in there?), because Intel doesn't recognize the extreme need for desktop user 32/64 support. And i tend to agree with them on this conjecture.

C.) I think some good points have been made that show that we don't really know if 64 bits (on average) will increase or decrease performance on most real world applications

Here's the smartest thing AMD can do. AMD is changing their arch around with the A64, right? Right. A lot of that will be new speedpaths and improved efficiency. They also will be making the switch to 64 bits. What they should try to do is blind everyone to the fact that this chip with 32 bits would have done much better than current technology ANYWAY (because of arch enhancements) and hint that it's because of the 32/64 bit support.

We'll just have to see...

I'm just your average habitual smiler =D

 

[TheRod]

B.) I think the reason Yamhill is still so far out (what do the rumors say 2005-2007, somewhere in there?), because Intel doesn't recognize the extreme need for desktop user 32/64 support. And i tend to agree with them on this conjecture.

C.) I think some good points have been made that show that we don't really know if 64 bits (on average) will increase or decrease performance on most real world applications

It might not be essantial to have 64 Bits APP now! But if AMD do is job well! they will "impose" their 64 Bits technology for the future...

So instead of MMX/SSE/SSE2 AMD processors. We may see Intel with AMD64 thecnology. This would gives AMD a big BOOST in reputation. Since many people still think that AMD makes only low/mid qualitu CPUs.

--
Would you buy a GPS enabled soap bar?

 

[Kelledin]

While there may be a little disadvantage from this, i think doubling the memory (at least...they don't give the number of 256 memory sticks used on the P4/Athlon machines) MORE MORE MORE MORE than compensates for this.

Not if the extra memory capacity isn't put to use. If the system never swaps even without the extra memory, then the added memory capacity is no benefit at all. And 512MB is actually a LOT for simple benchmarks, gameplay, desktop usage, and even professional apps. Photoshop, 3DSMax, and some other professional apps could benefit from gobs and gobs of memory, but generally not in an artificially-condensed benchmark scenario.

On Linux, it might matter, since Linux takes all unused memory and puts it to use for little piddling tasks, just so that memory will add some benefit instead of sitting idle. I don't think Windows manages memory like that, though.

<i>I can love my fellow man...but I'm damned if I'll love yours.</i>

 

[SJJM]

I thought that Win xp changed the how it uses its memory so that it keeps more info so that programs run fasters and that 512mb is basic amount of memory to have programs to run well.

<font color=blue>"You know, that my backstab attack does double the damage. I can make an off button for him." </font color=blue> 😎 <P ID="edit"><FONT SIZE=-1><EM>Edited by SJJM on 08/12/03 04:40 PM.</EM></FONT></P>

 

[SJJM]

I don't see why it would make that much of a difference to change to 64bit. As the reveiws show that even in 32bit that cpu is still running fast. I feel that it will not help with changing to 64bit software. In fact I think they are trying to make it in with microsoft, by trying to get people to move to 64bit software. Think of all the people who are going to go out and buy windows 64 just for a64.

<font color=blue>"You know, that my backstab attack does double the damage. I can make an off button for him." </font color=blue> 😎

 

[MaximumGoat]

"Think of all the people who are going to go out and buy windows 64 just for a64."

thats what i find strange, as it wont be very many people, as amd only has a small market share, and a lot of them will keep their existing OS.

"keyboard error or no keyboard present, press F1 to continue or DEL to enter setup" spot the deliberate mistake

 

[eden]

Microsoft will NEVER produce a product if they won't reap tons of cash from it.

This is just one more in their lap of schemes. Windows 64, if ever out, will go public like never before seen. Microsoft will advertise the living hell out of it, or make sure it does by OEMs. So trust me, if it will be out, OEMs will be forced to adopt Athlon 64s, and use Windows 64, on account buyers will fall for the new number.

THAT'S marketting. If AMD gets that succesfully, they WILL win this battle. So, I am still baffled as to how MS will really accept AMD's dare, being a small company will little resources, relatively.

--
<A HREF="http://www.lochel.com/THGC/album.html" target="_new"><font color=green><b>A sexual experience like never before seen</font color=green></b></A>
Site has now even more sexy members, for your pleasure.

 

[b3arPaW]

"One may argue about how importantly the amount of memory affects those benchmarks, but 2G on opteron vs. 256M on P4? Shame on AMDZONE!"

Wrong - AXP and P4 rigs used two high end corsair sticks. 512 megs of ram is more than overpumping for almost all bechmarks.

opteron rig only gets latency penalty from such a massive amount of ram, and amdzone shows us how easly opteron platform handles such amounts of ram.

Anyway, opteron impressed me with beautiful performance and even more beautiful scaling.