A64 performance @2.0Ghz

Mephistopheles

Distinguished
Feb 10, 2003
2,444
0
19,780
I was just looking at THG's Opteron/Workstation tasks review and thinking...

Well, A64 will use the same core as Opteron, right? So let's assume that, per clock, A64 performs identically (which is probably <i>not</i> the case, because of the single-channel DDR). And let's also assume that scaling the clock rates scale the performance numbers linearly, i.e. perfect scaling. So let's try to preview A64@2.0Ghz's performance...

Now think about the benchmarks we saw that compared the 3.0Ghz P4 on Canterwood and a single 1.8Ghz Opteron. I'll give you a little reminder:

<b>TEST NAME</b>/<font color=blue>3.0GhzP4result</font color=blue>/<font color=green>1.8GhzOpteron</font color=green>/<font color=red>2GhzOpteron*</font color=red>/<font color=orange>2.5GhzOpteron*</font color=orange>
<b>Lightwave 7.5</b>/<font color=blue>182</font color=blue>/<font color=green>253</font color=green>/<font color=red>227</font color=red>/<font color=orange>182</font color=orange> (lower-better)
<b>Cinema 4D</b>/<font color=blue>95</font color=blue>/<font color=green>124</font color=green>/<font color=red>97</font color=red>/<font color=orange>89</font color=orange> (lower-better)
<b>3DSMAX5</b>/<font color=blue>92</font color=blue>/<font color=green>124</font color=green>/<font color=red>97</font color=red>/<font color=orange>89</font color=orange> (lower-better)
<b>mp3maker</b>/<font color=blue>73</font color=blue>/<font color=green>112</font color=green>/<font color=red>101</font color=red>/<font color=orange>81</font color=orange> (lower-better)
<b>Main Concept</b>/<font color=blue>212</font color=blue>/<font color=green>273</font color=green>/<font color=red>246</font color=red>/<font color=orange>197</font color=orange> (lower-better)
<b>Pinnacle Studio</b>/<font color=blue>221</font color=blue>/<font color=green>314</font color=green>/<font color=red>283</font color=red>/<font color=orange>226</font color=orange> (lower-better)
<b>Winrar</b>/<font color=blue>49</font color=blue>/<font color=green>58</font color=green>/<font color=red>52</font color=red>/<font color=orange>42</font color=orange> (lower-better)
<b>PCMark2002</b>/<font color=blue>7557</font color=blue>/<font color=green>5856</font color=green>/<font color=red>6507</font color=red>/<font color=orange>8133</font color=orange> (higher-better)
<b>SiS.Sandra-Dry</b>/<font color=blue>5879</font color=blue>/<font color=green>2513</font color=green>/<font color=red>2792</font color=red>/<font color=orange>3490</font color=orange> (higher-better)
<b>SiS.Sandra-Whet</b>/<font color=blue>9170</font color=blue>/<font color=green>7223</font color=green>/<font color=red>8026</font color=red>/<font color=orange>10092</font color=orange> (higher-better)
<b>SiS.Sandra-MultiM-Int</b>/<font color=blue>14004</font color=blue>/<font color=green>7191</font color=green>/<font color=red>7990</font color=red>/<font color=orange>9988</font color=orange> (higher-better)
<b>SiS.Sandra-MultiM-FP</b>/<font color=blue>21487</font color=blue>/<font color=green>10336</font color=green>/<font color=red>11484</font color=red>/<font color=orange>14356</font color=orange> (higher-better)
*-test results obtained through linear scaling.

Now what about those numbers?... Isn't Opteron's architecture supposed to be equal or better than A64's?... Is my scaling absurd?... Is A64's performance at same clock superior to Opteron's?... Hm...

If I'm not crazy, A64 will very likely be trounced by Prescott at 32-bit. Anyone disagree? With a reason?...
 

eden

Champion
I can only comment, with even the slightest bit of information available, that perhaps when real platforms, i.e. mainboards ready for Athlon 64 and graphics in mind, will likely have somewhat better performance. Opteron uses ECC and is geared at stability and reliability first.

But even then I have no hope or hype towards Athlon 64's performance really.

--
I am my own competition. -VJK
 

Mephistopheles

Distinguished
Feb 10, 2003
2,444
0
19,780
that perhaps when real platforms, i.e. mainboards ready for Athlon 64 and graphics in mind, will likely have somewhat better performance. Opteron uses ECC and is geared at stability and reliability first.
You're being optimistic there, of course, but I hope you're right. And I've double-checked, A64 is certainly the same core as opteron - so there are no features on A64 that aren't already in Opteron. :frown:
 

eden

Champion
Possibly Opteron for its scaling in MP environments (more HT links) and the Dual Channel nature, but otherwise I cannot spot any other architectural difference.

However, like I said, I hold no more expectations from this CPU. If it's awesome, it's awesome, if it fails, oh well, tuff luck, they had it coming.
On the other hand a Prescott still amazes me. I'm a huge fan of the P4 architecture, love everything about it, and I can't wait to see how the increase Trace Cache size will affect. Doesn't matter if it's not too much, but it helps us analyze and understand it more.

--
I am my own competition. -VJK
 

juin

Distinguished
May 19, 2001
3,323
0
20,780
P4 3.0 on workstation can beat a Opteron 2.5 GHZ on dual channel to think a A64 can do it 2.4 ghz MAX clock speed.On game and support for DDR 400 will help more that Dual Channel.All score is comparing a 3.0 GHZ northwood to say a presscott will lose with 1 GHZ advance and twice the bandwith.This clock speed disparity is made that vs lowest clock speed of presscot vs highest speed of A64.Not to account that in 2004 grantdale and faster FSB coming to leave Intel with even more bandwith but higher latency.


I have made my point clear before A64 will not be faster that a Northwood at release.

I dont like french test
 

juin

Distinguished
May 19, 2001
3,323
0
20,780
WHO by now you love even more the P4 that me personalie there are many flaw but most cannot be solve unless there a ISA change that why i fell that Itanium is the only posible way to next generation level of performance.The instant everything

I dont like french test
 

Crashman

Polypheme
Former Staff
So it appears that the inefficiency of the P4 is meaningless when compared to the low clockspeed processors it has to compete against. That is to say, it beats it's inefficiency with brute force, AMD can't finesse their way to better performance, and AMD needs more clock speed than they can come up with.

<font color=blue>Watts mean squat if you don't have quality!</font color=blue>
 

imgod2u

Distinguished
Jul 1, 2002
890
0
18,980
So it appears that the inefficiency of the P4 is meaningless when compared to the low clockspeed processors it has to compete against. That is to say, it beats it's inefficiency with brute force, AMD can't finesse their way to better performance, and AMD needs more clock speed than they can come up with.

I fail to see how shoving 9 execution units with 9 issueing ports where around 70%+ of that remains idle most of the time counts as "efficiency". Even in the best case (absolutely no data dependencies, no memory latencies, the processor is filled to the fullest extent), 67% of that die would still go idle. All that translates into wasted die space, wasted transistors and a lot of heat.
Now, the P4 isn't any better, it's just as inefficient a processor but it's not "making up for it" with anything.

"We are Microsoft, resistance is futile." - Bill Gates, 2015.
 

Crashman

Polypheme
Former Staff
I was thinking more allong the lines of how P4's compared to PIII's and traditional Athlons, clock for clock.

<font color=blue>Watts mean squat if you don't have quality!</font color=blue>
 
The only problem is that if Itanium became mainstream right now, AMD would be out of business, as Intel would immediately render the P4 AND all Athlons obsolete with the first mainstream introduction of a desktop Itanium.

Intel would then effectively be a monopoly....thus....until then, Intel with their P4 has the opportunity to leech more money out of us, while maintaining marketshare and production profits...

Intel is in the position that EVERY business would like to be in...in regards to desktop processors.

....ok...so call me a conspiracy theorist...LOL!!! :smile:

<font color=blue> Ok, so you have to put your "2 cents" in, but its value is only "A penny's worth". Who gets that extra penny? </font color=blue>
 

imgod2u

Distinguished
Jul 1, 2002
890
0
18,980
I would guess that Intel would eventually be forced to liscense out IA-64, whether due to a court ruling or due to the fact that it has to offer AMD (which does have a sizable market share right now) some incentive to drop x86-64.

"We are Microsoft, resistance is futile." - Bill Gates, 2015.
 

Mephistopheles

Distinguished
Feb 10, 2003
2,444
0
19,780
So it appears that the inefficiency of the P4 is meaningless when compared to the low clockspeed processors it has to compete against. That is to say, it beats it's inefficiency with brute force, AMD can't finesse their way to better performance, and AMD needs more clock speed than they can come up with.
One thing is giving the P4 a big advantage right now: SSE2. SSE2-optimised programs that have been compiled with Intel's compilers are fast...
 

imgod2u

Distinguished
Jul 1, 2002
890
0
18,980
It's not just SSE2. The P4 is very picky about the code it runs. There are a lot of factors when it comes to optimization. Avoid bitshifting like the plague, avoid node-based data structures like Barbara Streisand's bedroom, try to keep your loops small and tight and avoid branching if you can.

"We are Microsoft, resistance is futile." - Bill Gates, 2015.
 

juin

Distinguished
May 19, 2001
3,323
0
20,780
It's not just SSE2. The P4 is very picky about the code it runs. There are a lot of factors when it comes to optimization. Avoid bitshifting like the plague

Worse that plague it a set of 10 handbrake on a P4.

I dont like french test
 

juin

Distinguished
May 19, 2001
3,323
0
20,780
I think that Intel may want to force it ISA by licensing to transmeta and AMD may be the best.As for AMD if they fail in X86-64 they are out of buiness if they have licensing now and start develop they may be able to have a good Ia-64 cpu before it here on the desktop.

I dont like french test
 

dwellman

Splendid
Dec 14, 2002
3,792
0
22,790
I like it when people use <i>efficiency</i> (ask Eden).

There's this nice management philosophy called <i> Theory of Constraints</i>. The P4 architecture is a virtual microcosm of TOC principles. One of which is <i>efficiency </i> is a misnomer, at best, and a meaningless metric the rest of the time. What most of us are concerned about is <i>throughput</i> i.e., results.

It is ok that parts of the P4 remain "idle", because it (like any linear entity) can only operate as fast as its slowest part, which seems to be fast enough.

Regards,

Dave

__________________________________________________
<b><font color=red>Three great virtues of a programmer are: laziness, impatience, and hubris.</font color=red><b>
 

juin

Distinguished
May 19, 2001
3,323
0
20,780
That intel big issue is biggest competitor to itanium is there own X86 chip.

On compiler side intel produce the best X86 and a weak IA-64 compiler only fast on FP wich is easier that INT.Must keep X86 but must develop IA-64.

Chip by slowing down X86 chip that will led to let the market to AMD if AMD have more that 50% of X86-32 bit market why move to IA-64 when there allready 50% of useur have a 64 bit capable CPU.

Must keep X86 chip and must continue Itanium chip.

On software side even worse.Try to create a market with no software and on ISV side is to port software to a imaginary market.

OS side Windows 2003 for Itanium the 1 os made with Itanium in mind was made also for itanium good news WIN 2003 are fast and well made but IA64 version cost a premium like all IA64 software as they lack of mass production.

Confident and reputation of intel will play a bigger part that the itanium itself.Unless Itanium can be faster that any X86 chip by a very good margin 75% or 100% it will not be take as new contender for Desktop CPU.

I dont like french test
 

Mephistopheles

Distinguished
Feb 10, 2003
2,444
0
19,780
Yes, indeed... If Intel could do x86 with 64 bit extensions and turn Prescott into a fully-fledged 64-bit high performance chip, they'd be shooting their own feet (i.e. rendering Itanium uninsteresting...)

Anyway, the thing about 64-bit migration is that it requires a great deal of software support and enthusiasm. In that sense, if x86-64 lost its hype, it would be very bad for them.

Compilers' performance is very important indeed...
 

Kemche

Distinguished
Oct 5, 2001
284
0
18,780
Compilers' performance is very important indeed...
Yes, indeed. Until now AMD has been relying on Intel to devlope compilers. But for 64bit AMD has to have 3rd party develope for them. As you know Intel probably has the best compiler out there, I wonder how the new 64bit compiler will preforme vs. intel's 32bit compiler.

KG

"Artificial intelligence is no match for natural stupidity." - Sarah Chambers
 

eden

Champion
No.
You should read articles around the web including on AMD's website to know more, this has been quite an old thing really.

--
I am my own competition. -VJK
 

juin

Distinguished
May 19, 2001
3,323
0
20,780
Simple i surprise of this statement form you on P4 wich i found there some architecture flaw or route that intel i take that i think is not the best.

I dont like french test
 

eden

Champion
Personally I have always been amazed. Have you not read my posts over 6 months back? I am sure I expressed many times by love for it. It's a great promising architecture. It is weak now, but it is the most potential-filled one.

--
<A HREF="http://www.lochel.com/THGC/html/news.html" target="_new">Join the Tom's Hardware Guide Community forum album, send us your pics!</A>