-
Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!
Phenom/Barcelona Reviews/Previews and Pre-release Discussion
- Thread starter ]-[itman
- Start date
Try your equation with a time of 65s and a clockspeed of 4.8GHz. This is scaling of 100% efficiency with a performance increase of 100% and a clockspeed increase of 100%. But your equation considers it only a performance increase of 50% and a clockspeed increase of 100%. Your equation is obviously incorrect.
Still using the wrong efficiency equation. The Opteron even with its slightly higher efficiency would not catch the 3GHz Xeon at 3GHz and its doubtful that its efficiency at 2.5GHz to 3GHz would be the same as from 2GHz to 2.5GHz given the nature of WME.
Randy Allen promised 2.5GHz by December.
Even if the 2GHz Opteron gets the win for POV-ray, it still only wins in SPECjbb2005 (and will probably lose when its official score is published considering the official 2xX5365 score is over 230,000), POV-ray and Myri-Match while losing in Valve VRAD, Cinebench, Eulers, Panorama Factory, picColor and WME compared to the 2GHz Xeon.
- Joined
- Aug 30, 2004
- Location
- Santa Barbara, CA
Keep it clean and on topic folks. 
- Joined
- Apr 9, 2006
- Location
- Falcon Heights, MN
Keep it clean and on topic folks.
amen
I will personally probably buy amd in the next couple years just to help fund competition. I will do my part
I will really start watching the amd procs as soon as someone on this forum gets one.
Last edited:
![]-[itman](/forums/data/avatars/m/9/9309.jpg?1650325391){kind=avatar}
- Thread Starter
- #146
Actually 100% improvement would be as close to 0 sec as possible, but now you're getting more into calculas as this is obviously not possible to perform a task in 0 sec. You're looking at it backwards. It's not 100% from 65s to 130s, but rather 50% from 130s to 65s. Look here and here, at anandtech, they calculate the % improvement in Windows Movie Maker from the 3ghz Core2 from the 2.93ghz Core2 at 5.8% using my same calculation.
[(74.4-70.1)/74.4] = 5.8%
Using the correct efficiency calculations you're looking at a Xeon 2.5ghz pulling in a performance of 444.3s compared to 447s of the 2.5ghz K10 which means the difference is only 0.6%. . . you really don't think a 3ghz K10 opteron wouldn't match a 3ghz Xeon? This is also discounting the poor scaling between 2ghz and 2.33ghz of the Xeon which at this point I don't think is trustworthy.
Perhaps it will be December, as I said, there are rumors that faster Opterons will be out by late Oct. or Nov. I guess we'll have to wait and see.
This is exactly what I said, 3 for Opteron, 5 for Xeon as the WME win is quite slim and will disappear once faster Opterons appear.
- Thread Starter
- #147
No offense, but that video is almost a month old and really doesn't show anything useful. . .
- Joined
- Feb 11, 2002
- Location
- U.P. Michigan
- Thread Starter
- #149
Something else I noticed in the cinebench at techreport.
Looking at the eight-core results, going from 2ghz to 2.5ghz on K10 yields a 31.1% performance increase meaning a 124% efficiency! Could this be an anamoly? Sure. Could it also be an example of the bandwidth hungry core we hear about finally getting some bandwidth to unleash it's potential? Also a possibility. Personally, I think it's a combination of both. I think in certain examples, the increased memory bandwidth given will allow the scaling of a Phenom cpu to be more efficient than even Barcelona has shown, but I also think something else may be going on in this benchmark we don't know about. Just my observation/opinion.
Looking at the eight-core results, going from 2ghz to 2.5ghz on K10 yields a 31.1% performance increase meaning a 124% efficiency! Could this be an anamoly? Sure. Could it also be an example of the bandwidth hungry core we hear about finally getting some bandwidth to unleash it's potential? Also a possibility. Personally, I think it's a combination of both. I think in certain examples, the increased memory bandwidth given will allow the scaling of a Phenom cpu to be more efficient than even Barcelona has shown, but I also think something else may be going on in this benchmark we don't know about. Just my observation/opinion.
- Joined
- Feb 11, 2002
- Location
- U.P. Michigan
If numbers don't add up, that means something is bad not good.
Not all variables are accounted for in the production of these numbers.
That's an infinite performance increase. So when somebody runs 100m in 20s, and another person runs it in 10s, is the second person only 50% faster than the first?.
So with the correct scaling equation, the 3GHz Opteron would 381s. Still slower than the 3GHz Xeon and that's only assuming it maintains as high a scaling factor.
Based on AMD's recent performance and credibility of AMD rumors, I wouldn't bet on it.
Well, then the Myrimatch lead is also slim and regardless, Xeons are still faster clock-for-clock.
- Thread Starter
- #152
Yes. If you have problems with the equation, go take it up with Anandtech, like I said, they calculated their results the same way; or go argue with a physics teacher as I'm sure they'd tell you the same.
Actually, if you work it from the baseline the equation was derived from, it would put the opteron at 350.2s, a 7% lead over the Xeon. Even if you use 381s, the Opteron would still be within 1.2%, a tie.
You consider a 5.3% performance advantage with a 20% deficit a slim lead(2.5ghz K10 beats a 3ghz Xeon)?
It's these kind of open-ended, unqualified conclusions when we have only seen previews of K10's performance that bother me.
And clearly they calculated it wrong like you did. I'm sure a physics teacher will take no more than a few seconds to tell you somebody going at 100 km/hr is 100% faster than somebody going at 50 km/hr.
I consider that Myri-Match which performs absolutely poorly with more than a few threads not to say very much at all.
There's enough evidence to make reasonable predictions.
- Thread Starter
- #154
Ummm, no. You just switched from a percent decrease to a percent increase, they're slightly seperate calculations. It may be counter-intuitive, but many things in math are. For ease of use.
http://www.marshu.com/articles/calculate-percentage-increase-decrease-percent-calculator.php
A simpler, less accurate way but gives the same results for all my calculations,
http://mathforum.org/library/drmath/view/58138.html
And finally exactly how I do my calculations,
http://retail.about.com/od/retailingmath/a/retail_percents.htm
If you still refuse to believe me, go to a math forum and ask them if my calculations are right, they'll confirm.
And? A lot of programs have trouble with threading, even retail programs, i'm sure they'll fix it over time. You can't just say I don't like that benchmark so it doesn't count. I could do the same thing for the Valve bench as it has yet to have any real application that I know of. I don't because I still think it's a valid test of certain cpu abilities. If you have a good reason for throwing out Myri-Match, I'm open to it.
Perhaps, but you haven't specified any predictions, you've made unqualified statements based on weak evidence of what you're trying to support.
- Joined
- May 26, 2004
- Location
- Japan, Tokyo, Ueno.
- Joined
- Aug 30, 2004
- Location
- Santa Barbara, CA
I see a 2Mb L3 in there. What would the cache latencies be? Would the overall latency increase or decrease?
- Thread Starter
- #157
I guess we'll have to wait and see for sure, but if Anandtech's initial results hold true, Mr. Allen's assesment may have been a little conservative,
http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=3092&p=6
Or perhaps he's using mode instead of mean as even the geometric mean would yield 18%.
- Joined
- Nov 17, 2005
How well do intels core2 scale with 25% increase in clock? I would like to know 2ghz -2.5ghz and 2.33 -3ghz.
- Thread Starter
- #159
Well, overall latency increases when there is an L3 cache miss, but greatly decreases if the cpu can find the data in L3 rather than going to main memory. I believe L3 can also be used for inter-core communication, I could be wrong about that though. Either way, obviously AMD engineers thought it would help more often than it would hurt. I don't know if we'll ever see no-L3 cache K10's, that would be the only way to know for sure where it helps and where it hurts. When the L3 cache increases to 6mb with the next core revision(I believe when they go to 45nm), I think we'll see the L3 cache have a bit more noticeable effect.
- Thread Starter
- #160
It depends on the application, just like the K10. We'll have to wait and see some more reviews to know for sure in general which scales better.
Similar threads
