The "perfect" Athlon 64 rating

TheRod

Distinguished
I worked on this for some time now, and I found a mostly perfect way to get the REAL Athlon 64 rating versus Pentium 4. We all know that the AMD rating is quite optimistic and it is sometimes misleading. Athlon XP are a very good example of it.

So, I decided to take into account the performance of Athlon 64 and P4 in various applications and I have normalized the results. Of course, Athlon will still get beat in Video/Audio encoding and P4 will still trail in most games. But my overall "Enhanced Athlon 64 Rating" gives a better overall view. A more realistic view based ona wide range of applications.

<b>Here is the maths…</b>
You need these Athlon 64 informations : Speed, Cache Size and Socket Type.

X = <font color=red>Speed / 100</font color=red>
Y = Cache Size (in KB) / 512
If Socket = 754 then Z = -1
If Socket = 939 then Z = 1
If Socket = 940 then Z = 0
<font color=red>If Speed >= 2.4GHz then S = 1 else S = 0</font color=red>
Enhanced Athlon 64 Rating = (X + Y + Z <font color=red>+ S</font color=red> + 9) * 100

9 is the offset to compensate Athlon 64 lower speed.
<font color=red>"S" is needed for the non-linear scaling of performance vs speed.</font color=red>

Here is an example of this equation :

Athlon 64 3200+ (S754/1Megs)
X = <b>2000</b>/100 = 20
Y = <b>1024</b>/512 = 2
Z = -1 for <b>S754</b>
S = 0 for <b>2.0GHz</b>
X + Y + Z + S + 9 = 30
30 * 100 = 3000+

So the Athlon 64 3200+ lost 200 point in this new rating, but overall, this Athlon 64 would be a better P4C/E 3.0GHz competitor. As I stated before, the A64 would crush the P4 in games, but it would not get that badly beatin by the P4 in other disciplines where the P4C/E shines.

With this "Enhanced Rating", the Athlon 64 rating move from 2800+/3800+ to 2700+/3600+. Overall there is no big change, just a little correction to some overated Athlon 64 (see 3500+). By the way, you can calculate the Athlon FX-53 rating with my equation. It can be compared to a 3700+.

NOTE : <A HREF="http://www.hardware.fr/articles/496/page12.html" target="_new">The graphic on this page</A> inspired me in this search for a better rating.

- I forget the "S" in my first post.
- I changed a bit the way to calculate the X. I know based it on speed instead of the multiplier, this helps evaluate the rating of overclocked Athlon 64. The only thing I need to evaluate now, is how the FSB overclock affect the overall rating, but I would need testing and overclocking numbers to figure this out. I don't think it impact the overall rating by much, but I know it impact it a bit.</font color=red>

--
What's the <b><font color=green>AMD Mobile Athlon 64</font color=green></b> overclocking potential? <b>It's huge!</b><P ID="edit"><FONT SIZE=-1><EM>Edited by TheRod on 06/04/04 08:12 AM.</EM></FONT></P>

trooper11

Distinguished
the 3500+ isnt as bad as oyu think. im sure you read anandtech's article comparing them and it showed that in most tests indeed its performance, along the lines of 2%-6% overall, was in line with the increased pr rating of 100 points. everyone was expecting to kill it by more, but it didnt, well everyone should have expected that.

i would agree, amking rating conservative is a much better choice and i wish it were that way just to make chips look closer on some areas where amd chips fall behind. But this is all based on point of view. The ratings as they are now do indeed equal better performance then similar rated p4's on SOME tests, but not all. In the cases taht its down its not be a huge margine, its still close enough to compete.

i just dont see the use in changing the numbers now. how much would they have to cut back to bring its video/audio/3d numbers up to par with a similar p4?

TheRod

Distinguished
I read the AnandTech article too. I know that the A64 3500+ is not that bad. But take percentage in account, the A64 3400/3500 are mostly equal in 14 of the 17 (under 3% diff.) AnandTech tests. I know AMD really wants to distingish both products, but a 3400/3450+ rating would have been better.

I know, it's impossible to find a single answer to AMD/Intel comparison. The only thing that will prove my "equation" right or wrong will be the release of the new P4. Hardware.fr have made a good normalised graph (see the link in my last post) and I really think it's the way to go. We can all conclude that AMD rock in gaming/compiling and we can't deny that the P4 is strong too. I just wish that AMD will not do the Barton Rating mistake with the A64. It's why I think they should have took a more conservative rating. This would ensure that this rating would not become obsolete when newer P4 will come out.

I don't want to change the way AMD rate their CPU, but I only wanted to find a way to better compare AMD/Intel offering. The best way to do so would be to substract 600 in media encoding and add 400 in gaming for all the AMD rating.

--
What's the <b><font color=green>AMD Mobile Athlon 64</font color=green></b> overclocking potential? <b>It's huge!</b>

scottchen

Splendid
LOL Good work ROD! This is GREAT WORK!!

<A HREF="http://forums.extremeoverclocking.com/myrig.php?do=view&id=17301" target="_new">My PC</A>

scottchen

Splendid
Question how would i calculate something that has FSB of 230?

<A HREF="http://forums.extremeoverclocking.com/myrig.php?do=view&id=17301" target="_new">My PC</A>

scottchen

Splendid
Oh right Rod also, sorry to dissapoint you but the mobile Athlon64s with 1mb cache also has problem going beyond 2.5Ghz without dangerously high Vcore. Friend of mine bought one, he got that Zalman passive watercooling kit, it's not great but better than air. He has the Chaintech ZNF3-250 mobo, with the mobile 3200+(10x200), he had the temp at about 25C load at stock, 1.4Vcore. He pushed it to 240mhz FSB no problem at all, using only 1.55Vcore, however going anything beyond that required Vcore jump, 1.7 was required to hit 244, 1.8 required for 247 and 1.9 was required for 252. He was pissed at that so he just returned it, and bought the desktop 3400+.

<A HREF="http://forums.extremeoverclocking.com/myrig.php?do=view&id=17301" target="_new">My PC</A>

blackphoenix77

Distinguished
If there were more than 24hours in a day, then I would feed my dog once more before I went to bed. *2 mc(2) yeah I took Geometry...but I flunked it

<b><font color=red> ATI 9600Pro </font color=red></b>
<b><font color=green> AthlonXP-M 2500+ OC'd 3200+ </font color=green></b>
<b><font color=blue> Abit NF7-S </font color=blue></b>
<b><font color=black> 2x256MB Corsair PC3200 </font color=black></b>

P4Man

Distinguished
> the A64 3400/3500 are mostly equal in 14 of the 17 (under
>3% diff.)

The difference between the 3400 and 3500 ratings is also less than 3%, in case you hadnt noticed; and further more cpu's never scale lineary in performance with their clockspeed/rating on most apps, ~75% on average would be good scaling already, which means a 3500+ ought to outperform a 3400 on average by less than 2% to earn its rating, so I'm not sure what your point is. This is statistical noise, two different motherboards or even two test runs on the same MB may well give you a far bigger variation. The 3500+ is clearly not slower than the 3400+, and its a different cpu. That alone warrants this *marginally* higher rating IMHO.

As for your exercise above.. I fail to see the point again. If there was a some consensus on how to measure performance, it might be worthwhile, but reality is that determining performance very much depends on what benchmarks/applications you use. If I developped a cpu that performs like a 4+ GHz P4 on games and office/database apps but like a Duron 600 on media encoding and scientific/mathematical/cad/3D stuff, please tell me how you'd rate it.. ? And do you think argueing over 100 PR points would be worthwhile in this context ?

= The views stated herein are my personal views, and not necessarily the views of my wife. =

TheRod

Distinguished
Question how would i calculate something that has FSB of 230?
Oups! I didn't consider overclocking! You can try it and modifiy a parameter in my equation!

--
What's the <b><font color=green>AMD Mobile Athlon 64</font color=green></b> overclocking potential? <b>It's huge!</b>

TheRod

Distinguished
sorry to dissapoint you but the mobile Athlon64s with 1mb cache also has problem going beyond 2.5Ghz
Good to know, but the newer Mobile Athlon 64 @ 1.2V might be better. And going from 2.0GHz to 2.4GHz at 1.55V is nto bad at all.

You said he returned the mobile and got a 3400+, how well it overclocked? It must be important, because the Mobile 3200+ is a lot cheaper than the desktop 3400+.

--
What's the <b><font color=green>AMD Mobile Athlon 64</font color=green></b> overclocking potential? <b>It's huge!</b>

TheRod

Distinguished
The difference between the 3400 and 3500 ratings is also less than 3%, in case you hadnt noticed; and further more cpu's never scale lineary in performance with their clockspeed/rating on most apps,
I can't argue!

As for your exercise above.. I fail to see the point again. If there was a some consensus on how to measure performance
I have no real point... I wanted to have an "overall" comparison. I know it's impossible to have really perfect rating. But, even if I like AMD and think they made good CPU, their ratign is a bit high compared to P4 when you check it overall. There is many apps/situation that really benefits from the P4 architecture. Of course, AMD marketing must be optimistic and I can't blame them for that. But, I'm not sure that the actual Athlon 64 3800+ will really beat that much a P4E @ 3.8GHz overall.

And do you think argueing over 100 PR points would be worthwhile in this context?
+/- 100, is like arguing for +/1 200MHz on Intel processor. Who would notice the difference in real world applications?

You have good points P4Man!

NOTE : Have you noticed that the Athlon 3800+ performs really well against the P4EE 3.4GHz. It's funny, I would like to know how many people got P4EE?

--
What's the <b><font color=green>AMD Mobile Athlon 64</font color=green></b> overclocking potential? <b>It's huge!</b>

P4Man

Distinguished
> I wanted to have an "overall" comparison.

Your idea of "overall" is different from mine, which is different again from AMD's,and very different from Intel's or Bapco's or Spec's, or VIA, or the pope's. okay, granted, I'm just guessing that last one . Very much more different than your "formula" is from AMD's, so its really not usefull. If you want a "better" rating than AMD's, a starting point might be to make it more differentiated for different kinds of apps. Eg, one for gaming, one for rendering, one for video/audio, office, etc. Pick a handfull of relevant benchmarks for each type of workload and see how the A64 fares against the P4.

> But, I'm not sure that the actual Athlon 64 3800+ will
>really beat that much a P4E @ 3.8GHz overall.

Again, depends what how you define "overall". Without being able to define that, such a statement is meaningless. "overall" a 3200+ might easily beat a P4 @ 3.8 GHz if UT2k4, chess and Xvid encoding is all you care about.

>Have you noticed that the Athlon 3800+ performs really well
>against the P4EE 3.4GHz

I have noticed the 3400+ does that on the apps I happen to care about.

= The views stated herein are my personal views, and not necessarily the views of my wife. =

TheRod

Distinguished
Eg, one for gaming, one for rendering, one for video/audio, office, etc. Pick a handfull of relevant benchmarks for each type of workload and see how the A64 fares against the P4.
I could add a type of apps. offset to my equation :

T = +4 for games
T = -5 for AV
T = -2 for rendering
T = +4 science
T = ...

This list could extend for a long time!!!

Ok, P4Man, you won! I will stop working on this, and I will mostly use my equation in my databases and internal stuff... I will not bother you anymore with it!

I have noticed the 3400+ does that (beat the P4EE) on the apps I happen to care about.
So, you are like me, you don't really care about Video/Audio and 3D randering performances! I like this kind of guy! HE HE HE!!!

by the way, have you read the memory compatibility article on THG. Wow! A great article for peopel who are interesting in buying an Athlon 64 based system paired with generic and HQ memory. And I must admit that ASUS still beat the competition in terms of compatibility. I don't like ASUS pricing, but I really like their quality. And even if I don't like their pricing, ma last 2 MB were ASUS (A7N8X and P3V4X).

--
What's the <b><font color=green>AMD Mobile Athlon 64</font color=green></b> overclocking potential? <b>It's huge!</b> Humm... Maybe not that huge...

TheRod

Distinguished
The difference between the 3400 and 3500 ratings is also less than 3%, in case you hadnt noticed; and further more cpu's never scale lineary in performance with their clockspeed/rating on most apps, ~75% on average would be good scaling already, which means a 3500+ ought to outperform a 3400 on average by less than 2% to earn its rating, so I'm not sure what your point is.
Just a little note. I just checked back my "equation" and I can confirm you that the 3500+ get a +100 over the 3400+ in it.

A3400+ (s754/2200/1024K)
<b>(</b> (-1) + (11*2) + (1024/512) + 9 <b>)</b> * 100 = MyPR 3200+
A3500+ (s939/2200/512K)
<b>(</b> (1) + (11*2) + (512/512) + 9 <b>)</b> * 100 = MyPR 3300+

--
What's the <b><font color=green>AMD Mobile Athlon 64</font color=green></b> overclocking potential? <b>It's huge!</b> Humm... Maybe not that huge...

P4Man

Distinguished
>I could add a type of apps. offset to my equation :

You got it backwards again. First you have to define those types, then select the benchmarks, see how the A64 compares to the P4 and ONLY THEN you could try and make a formula that fits the datapoints.

If you do that, no one will be able to argue against the formula, only against the selection/weighing of benchmarks.

>So, you are like me, you don't really care about
>Video/Audio and 3D randering performances! I like this kind
>of guy! HE HE HE!!!

I do care about video encoding, but I use Xvid, not Divx (xvid is totally free, and IMHO much better than DivX). A64 (and even AXP) trashes the P4 on Xvid encoding. As for audio.. I've got ~40 GB worth of MP3's that I ripped from CD, and converted to VBR MP3's. What took long was feeding all those audio Cd's into my CD drive, and the ripping, not the encoding. Encoding was almost on the fly (my drive can only rip at 8x). Besides, I did that once, took me a few hours, but since I don't buy 10 new CD's every day, I could care less about MP3 encoding speed.

= The views stated herein are my personal views, and not necessarily the views of my wife. =

rcj187

Distinguished
Y = Cache Size (in Kb) / 512
you mean KB not Kb. b denotes bits while B is bytes. one byte is 8bits.

Vapor

Distinguished
How about instead of multiplying by 100 at the end, you multiply by the FSB/2. That way a 3500+ @2.6GHz (or 237MHz FSB) would be a 3910+ while a regular 2.6GHz chip would only be a 3800+ on your system. The +1 (for 2.4GHz or higher) should only apply to stock speed (otherwise this would be a 4030+, which is excessive).

Anyway, I would like to see an article with an P4ES running at 3.8, 3.6, (4.0?), etc. just to see how aggressively AMD predicted Intel's performance at those speeds.

Maxtor disgraces the six letters that make Matrox.

scottchen

Splendid
The Athlon64 3400+ desktop hit 250FSB with 1.725Vcore stable. He got it to 263 with 1.9Vcore but not too stable, 260 was stable with 1.9Vcore. However the downside with this chip is that it required 1.7Vcore to hit 240FSB, while the mobile 3200+ only needed 1.55V

The crap thing about the new mobile chips is they have 512k cache, but the great thing is they only generate 35watts of heat at 1.2Vcore

<A HREF="http://forums.extremeoverclocking.com/myrig.php?do=view&id=17301" target="_new">My PC</A>

trooper11

Distinguished
actaully i do alot of mp3/wav ensocing and dbpower amp actaully does it faster on the thalon 64 then a p4 , so for me its better.

TheRod

Distinguished
I do care about video encoding, but I use Xvid, not Divx (xvid is totally free, and IMHO much better than DivX). A64 (and even AXP) trashes the P4 on Xvid encoding. As for audio.. I've got ~40 GB worth of MP3's that I ripped from CD, and converted to VBR MP3's. What took long was feeding all those audio Cd's into my CD drive, and the ripping, not the encoding. Encoding was almost on the fly (my drive can only rip at 8x). Besides, I did that once, took me a few hours, but since I don't buy 10 new CD's every day, I could care less about MP3 encoding speed.
I totally agree with you, I find Audio/Video encoding benchmarks irrevelent for the majority of users. As you stated, ripping MP3 cna be done so fast today, we ususally waits for the CDROM drive to catch on. Video Encoding is so wrongly hyped, I do have a MINI-DV camera and 90% of the time I do editing and I only encode when my editing is done. And my old Athlon XP @ 2.0GHz can do the job well. I really don't care how long it takes to convert my final edit in MPEG2, since it took me hours to edit... waiting a couple more minutes for the rendering to be done is so irrevelent.

DivX encoding, it's the same thing, who sits in front of his PC waiting a complete DVD to be ripped. No one!

Only Audio/Video professionnal should care about those numbers. And many professional used "rendering station/farm". And if Audio/Video is VERY important to a user, it should consider buying MAC G5, they are awesome Audio/Video working beast!

--
What's the <b><font color=green>AMD Mobile Athlon 64</font color=green></b> overclocking potential? <b>It's huge!</b> Humm... Maybe not that huge...

TheRod

Distinguished
you mean KB not Kb
Oups! Corrected, thank you!

--
What's the <b><font color=green>AMD Mobile Athlon 64</font color=green></b> overclocking potential? <b>It's huge!</b> Humm... Maybe not that huge...

TheRod

Distinguished
How about instead of multiplying by 100 at the end, you multiply by the FSB/2. That way a 3500+ @2.6GHz (or 237MHz FSB) would be a 3910+ while a regular 2.6GHz chip would only be a 3800+ on your system.
Not a bad idea, but if I start to consider all overclocking parameter, It could become very complicated. Let's say you RUN the FSB at 250MHz and run you memory at 4/5 because your DDR400 can't run at high speed. This would mess the thing... And with if you set 2T instead of 1T in the BIOS, because you wanted to attain higher stability in your overclocking... There is too much parameters to consider. And usually overclocker don't care much about the rating, they care much about the percentage increase they can achieve.

But it's not a bad to try add some FSB speed to the equation. But we should need some testing to know if your idea of FSB/2 is as good as it sounds. But I'm not sure the FSB have a huge impact on overall performance. The final speed should be near of a regular 2.6GHz Athlon 64 with a bit higher rating.

The +1 (for 2.4GHz or higher) should only apply to stock speed (otherwise this would be a 4030+, which is excessive).
I think this +1 should stay, because to Athlon 64 architecture scale better than the P4 architecture. And I just checked my equation and the 2.4 @ 2.6

--
What's the <b><font color=green>AMD Mobile Athlon 64</font color=green></b> overclocking potential? <b>It's huge!</b> Humm... Maybe not that huge...

TheRod

Distinguished
The Athlon64 3400+ desktop hit 250FSB with 1.725Vcore stable.
But I doubt it run at 250x12 = 3000MHz, It probably run at 250x10 = 2500MHz.

--
What's the <b><font color=green>AMD Mobile Athlon 64</font color=green></b> overclocking potential? <b>It's huge!</b> Humm... Maybe not that huge...

TheRod

Distinguished
How about instead of multiplying by 100 at the end, you multiply by the FSB/2.
I modified a bit my equation, it's now based on CPU speed instead of multiplier. So an overclocked Athlon 64 (s939/512K) @ 2.6GHz would be rated 3800+.

The only thing I need to know now, it's how the FSB/Memory overclock affect the overall performance. To know this, I would need numbers this kind of numbers : 200*10, 222*9 250*8 scores comparison. And It's not easy to find, since I don't personnaly have an Athlon 64 to test this.

--
What's the <b><font color=green>AMD Mobile Athlon 64</font color=green></b> overclocking potential? <b>It's huge!</b> Humm... Maybe not that huge...

scottchen

Splendid
250x10 and 263x9, so i guess chaintech board's max is 263, damn i guess MSI's better, able to hit 300mhz.

<A HREF="http://forums.extremeoverclocking.com/myrig.php?do=view&id=17301" target="_new">My PC</A>