Here's a little teaser....

[dave_graham]

dude learn to read i always said no linear not incremental.

Others kept saying wait and see more MHz will do wonders, is the stepping, etc..., that is a lie!

and what exactly have you done with Barcelona

are you just looking for another "scoop" for your website? more information that you can "leak" on upcoming products?

or, do you really want to come through here and sandbag people who have actual experience with the platform, have spent the time trying to understand the issues, etc? only you can answer that...I won't do it for you.

for the record:

B1 chips have the errata worked around in the BIOS.
BA chips have the errata fixed in silicon.
B2 chips are the "SE" parts and have not been committed to production yet (they will be shortly) and should be available by (or during) January.

i have not, anywhere, promoted the idea of performance scaling except where it concerns moving errata fixes from the BIOS to silicon. I have promoted the idea of hardware scaling (core speed increases) but yet, you somehow seem to think that I'm misleading you....interesting.

I have questioned the validity of posted results online based on what I have seen to be poor benchmarking, architecture understanding, etc. but i have not made promises on performance.

for the record, can people stop comparing a 2350 to a 5365? that's 1ghz in clock speed differences that really is a worthless comparison. of COURSE the 5365 is going to be on top. if it didn't, that'd be a big issue, now, wouldn't it?

cheers,

dave

[DoubleZero]

Dave somhow you are loosing credibiity in my book.

Check this:

So phenom at 2.5Ghz will only be 15% faster than Phenom at 2.0GHz so much for the incremental speed gains or linear speed gains.

Source

I guess you missed this:



Not perfect scaling, but better then "only 15%".

[metro.cl]

and what exactly have you done with Barcelona

are you just looking for another "scoop" for your website? more information that you can "leak" on upcoming products?

or, do you really want to come through here and sandbag people who have actual experience with the platform, have spent the time trying to understand the issues, etc? only you can answer that...I won't do it for you.

for the record:

B1 chips have the errata worked around in the BIOS.
BA chips have the errata fixed in silicon.
B2 chips are the "SE" parts and have not been committed to production yet (they will be shortly) and should be available by (or during) January.

i have not, anywhere, promoted the idea of performance scaling except where it concerns moving errata fixes from the BIOS to silicon. I have promoted the idea of hardware scaling (core speed increases) but yet, you somehow seem to think that I'm misleading you....interesting.

I have questioned the validity of posted results online based on what I have seen to be poor benchmarking, architecture understanding, etc. but i have not made promises on performance.

for the record, can people stop comparing a 2350 to a 5365? that's 1ghz in clock speed differences that really is a worthless comparison. of COURSE the 5365 is going to be on top. if it didn't, that'd be a big issue, now, wouldn't it?

cheers,

dave

hahaha you are totally sold to AMD.

Funny thing is that if i wanted i could have leaked tons of info related to K10.

Time will tell about performance but i firmly believe you arent right, K10 is not meant for desktop and not meant to beat C2D or Penryn in desktop market.

Read this i believe you know Kris and also i leak because as i've said before is the only thing i can do since we dont get the hardware we want even thought we are like Top 5 hardware review website in Spanish.

[kyosen]

One shop in Akihabara starts Opteron demo!
http://www.watch.impress.co.jp/akiba...elonademo.html
They use Tyan n3600M + Opteron 2347 x2.
...I must go there tomorrow

BTW, I've just read an official K10 Revision Guide:
http://www.amd.com/us-en/assets/cont...docs/41322.pdf
This guide describes only "DR-BA" and "DR-B2" revisions.
So, CPU-Z should shows "A" as stepping value in case of "BA" revision,
and "2" for "DR-B2".

[dave_graham]

hahaha you are totally sold to AMD.

i'm willing to devote my time and efforts to a company that actually supports the work i do. In addition, my customers have requested AMD workstations in greater proportion than Intel, so....of course, i'm going to spend the time there.

Funny thing is that if i wanted i could have leaked tons of info related to K10.

proud of you.

Time will tell about performance but i firmly believe you arent right, K10 is not meant for desktop and not meant to beat C2D or Penryn in desktop market.

and Barcelona is for Workstations and servers....sounds like you're in the wrong thread.

Read this i believe you know Kris and also i leak because as i've said before is the only thing i can do since we dont get the hardware we want even thought we are like Top 5 hardware review website in Spanish.

let me know what you need and chances are, i can provide it. I ship to Chile and Argentina all the time...no small task for me.

dave

[BeardyMan]

dude learn to read i always said no linear not incremental.

Others kept saying wait and see more MHz will do wonders, is the stepping, etc..., that is a lie!

Calm down boy, no need for being personal.

[Eldonko]

One shop in Akihabara starts Opteron demo!
http://www.watch.impress.co.jp/akiba...elonademo.html
They use Tyan n3600M + Opteron 2347 x2.
...I must go there tomorrow

Thanks kyosen-san, please let us know what you see!

[DoubleZero]

Just a little follow up to this:
http://www.xtremesystems.org/forums/...&postcount=277

dude learn to read i always said no linear not incremental.

Others kept saying wait and see more MHz will do wonders, is the stepping, etc..., that is a lie!

Just comparing E6420 and E6660, that have the same 25% clock diference.

Quake 4: - 7%
Unreal Tournament - 7,8%
Fear - 5,5%
Mpeg Encoding - 7,7 %
H.264 Enconding - 9,5 %
Cinebench 9.5 - 11,3 %
POV Ray 3.7 - 10.9 %

Tired of proving the point, more here:

http://www.xbitlabs.com/articles/cpu...0_7.html#sect0

Pretty obvious that at higher speeds K10 will be more competitive against c2d then at lowers speeds...

So where's the lie, and who's sold on what?

[red]

Just a little follow up to this:
http://www.xtremesystems.org/forums/...&postcount=277



Just comparing E6420 and E6660, that have the same 25% clock diference.

Quake 4: - 7%
Unreal Tournament - 7,8%
Fear - 5,5%
Mpeg Encoding - 7,7 %
H.264 Enconding - 9,5 %
Cinebench 9.5 - 11,3 %
POV Ray 3.7 - 10.9 %

Tired of proving the point, more here:

http://www.xbitlabs.com/articles/cpu...0_7.html#sect0

Pretty obvious that at higher speeds K10 will be more competitive against c2d then at lowers speeds...

So where's the lie, and who's sold on what?

E6420 is 2.13GHz. E6600 is 2.4GHz 12.7% difference. lol.

Out of a possible 25/25, AMD achieves 15/25 or 60% clock scaling. Intel achieves average of 8.5 scaling from above. 8,5/12.7 is 67% clock scaling.

[metro.cl]

Just a little follow up to this:
http://www.xtremesystems.org/forums/...&postcount=277



Just comparing E6420 and E6660, that have the same 25% clock diference.

Quake 4: - 7%
Unreal Tournament - 7,8%
Fear - 5,5%
Mpeg Encoding - 7,7 %
H.264 Enconding - 9,5 %
Cinebench 9.5 - 11,3 %
POV Ray 3.7 - 10.9 %

Tired of proving the point, more here:

http://www.xbitlabs.com/articles/cpu...0_7.html#sect0

Pretty obvious that at higher speeds K10 will be more competitive against c2d then at lowers speeds...

So where's the lie, and who's sold on what?

Once again that is diferent from K10 scaling linear or better.

This same thing happened to K8 vs C2D push MHz on both on stock and performance will get closer

[DoubleZero]

E6420 is 2.13GHz. E6600 is 2.4GHz 12.7% difference. lol.

There are days where i just need more coffee

Out of a possible 25/25, AMD achieves 15/25 or 60% clock scaling. Intel achieves average of 8.5 scaling from above. 8,5/12.7 is 67% clock scaling.

K10 were scaling 500mhz not 300mhz

[red]

http://www.xbitlabs.com/articles/cpu...4-x2-6000.html
2.93 vs 2.4 22% clock difference.
Quake 4 7.1% scaling
Unreal 17.8%
Fear 11%
MPEG 16.7%
h.264 19.2%
cinebench 23.5%
povray 21.6%

Average, 16.7% scaling out of possible 22% or 76%

[PhilDoc]

http://www.xbitlabs.com/articles/cpu...4-x2-6000.html
2.93 vs 2.4 22% clock difference.
Quake 4 7.1% scaling
Unreal 17.8%
Fear 11%
MPEG 16.7%
h.264 19.2%
cinebench 23.5%
povray 21.6%

Average, 16.7% scaling out of possible 22% or 76%

Those aren't quad cores thery're dual core Intels and the Bacelona is 19/25 or 76%.

[red]

Where do you get Barcelona 19/25?

Here I go again
http://www.xbitlabs.com/articles/cpu...me-qx6850.html
qx6800 vs q6600 both 1066fsb, 2.93 vs 2.4
Quake 11.4%
Unreal (none)
Fear 10.5%
MPEG (18.1% for divx and 16.8% for xvid)
h.264 12.1%
cinebench 20.7%
povray 25.3%

16/22=73

[PhilDoc]

Where do you get Barcelona 19/25?

Here I go again
http://www.xbitlabs.com/articles/cpu...me-qx6850.html
qx6800 vs q6600 both 1066fsb, 2.93 vs 2.4
Quake 11.4%
Unreal (none)
Fear 10.5%
MPEG (18.1% for divx and 16.8% for xvid)
h.264 12.1%
cinebench 20.7%
povray 25.3%

16/22=73

From the Anand article and I only get 14.3/22 or 65%

[red]

my point of posting these numbers is that Intel scales a tangible amount with clock speed despite what others insist

[PhilDoc]

my point of posting these numbers is that Intel scales a tangible amount with clock speed despite what others insist

Sorry, just bored atm. I think the Intels scale very well and you've shown that.

[KTE]

I'm trying to roughly compare things here, on average.

K8 90nm-> K8 is 65nm - K10 is 65nm (shouldn't be much clock speed bin differences, short maturing time)
P4 65nm-> Core 2 is 65nm - Penryn is 45nm (should be much clock speed bin differences but long maturing time)

We know that staying on the same fab level does make clock speed raising easier as the process matures, but while that'll hold roughly true for AMD dual cores like it did for Intel dual cores, I doubt it too much when passing over to a fully different and new quad core.

K10 server: I've looked at Anandtech review long and hard and extrapolated backwards the dual Opteron 2224SE 3.2GHz scores for perfect clock scaling (which is obviously rare), down to 2x 2GHz and compared to 1x Opteron 2350 2GHz. A few times I used the 3GHz instead (where scores were available - twice I think). Flaws will arise since clock scaling won't be linear, but not by hugely.

Overall Opteron 2350 @ 2GHz (4 core) comes to be 52.2% (or 152.2%-100% for some) better than 2x Opteron 2224SE @ 2GHz (4 core).

Now that certainly is inline with what AMD has been saying.

Take one high Intel optimized benchmark score out of the equation, LINPACK and you get an average mean performance increase, clock per clock of:

39.28% for K10 server chip in server benches. That's not far from a Core 2 over P4...

August '07 HKEPC desktop Wolfdale 2330MHz was overall ~9.8% better than a E6550 2330MHz. Anandtech's was around 5%. We'll take the first one.

That's including the one high Divx score with Intel SSE4 specific optimization.

Take away that score and the mean average performance increase clock per clock basis is:

7.45% for a desktop chip with desktop benches.

Boards used were? Two separate ones IIRC, a GA-P35-DQ6 with DDR3 1066 for the Wolfdale and a GA-G33-DS3R DDR2 1066(?) for the E6550. Which ones runs 1066MHz on RAM default or did they overclock for RAM to the same speed? I'm not sure.

14.156sec SPI 1M at stock for that Wolfdale IIRC.

Tigerton/Clovertown: is beat clock per clock basis by K10 Opteron overall. With 2.13GHz and many times even with 2.33GHz parts compared to 2GHz.

Desktop: no idea.

Problem: Intel is competing with higher clocks and good performance per clock. AMD with better power efficiency and lower prices but lower clocks. Will outperform at the equal price market with its low clocks yet, but it has nothing for mid-high end.
AMD loses out when comparing top offerings due to low clock speeds against an already very matured chip and fab process. Price is very good. They just need higher clock speeds to compete more on a level basis. It'll be tight but I reckon AMD takes this clock per clock basis, and maybe by even price. They just need a 3GHz Agena X4 part by sometime Q1 when desktop Yorkfield will be releasing 3.16GHz.

Do I give a damn which chip kills what? No, because it kills inside weak minds only. Competition is always healthy, even Intel knows that. If the highest Barcelona quad beats or levels the highest Penryn with highest clock possible by Q2 08, then the same excitement and effort turns out for Nehalem, and so on.

[PhilDoc]

I'm trying to roughly compare things here, on average.

K8 90nm-> K8 is 65nm - K10 is 65nm (shouldn't be much clock speed bin differences, short maturing time)
P4 65nm-> Core 2 is 65nm - Penryn is 45nm (should be much clock speed bin differences but long maturing time)

We know that staying on the same fab level does make clock speed raising easier as the process matures, but while that'll hold roughly true for AMD dual cores like it did for Intel dual cores, I doubt it too much when passing over to a fully different and new quad core.

K10 server: I've looked at Anandtech review long and hard and extrapolated backwards the dual Opteron 2224SE 3.2GHz scores for perfect clock scaling (which is obviously rare), down to 2x 2GHz and compared to 1x Opteron 2350 2GHz. A few times I used the 3GHz instead (where scores were available - twice I think). Flaws will arise since clock scaling won't be linear, but not by hugely.

Overall Opteron 2350 @ 2GHz (4 core) comes to be 52.2% (or 152.2%-100% for some) better than 2x Opteron 2224SE @ 2GHz (4 core).

Now that certainly is inline with what AMD has been saying.

Take one high Intel optimized benchmark score out of the equation, LINPACK and you get an average mean performance increase, clock per clock of:

39.28% for K10 server chip in server benches. That's not far from a Core 2 over P4...

August '07 HKEPC desktop Wolfdale 2330MHz was overall ~9.8% better than a E6550 2330MHz. Anandtech's was around 5%. We'll take the first one.

That's including the one high Divx score with Intel SSE4 specific optimization.

Take away that score and the mean average performance increase clock per clock basis is:

7.45% for a desktop chip with desktop benches.

Boards used were? Two separate ones IIRC, a GA-P35-DQ6 with DDR3 1066 for the Wolfdale and a GA-G33-DS3R DDR2 1066(?) for the E6550. Which ones runs 1066MHz on RAM default or did they overclock for RAM to the same speed? I'm not sure.

14.156sec SPI 1M at stock for that Wolfdale IIRC.

Tigerton/Clovertown: is beat clock per clock basis by K10 Opteron overall. With 2.13GHz and many times even with 2.33GHz parts compared to 2GHz.

Desktop: no idea.

Problem: Intel is competing with higher clocks and good performance per clock. AMD with better power efficiency and lower prices but lower clocks. Will outperform at the equal price market with its low clocks yet, but it has nothing for mid-high end.
AMD loses out when comparing top offerings due to low clock speeds against an already very matured chip and fab process. Price is very good. They just need higher clock speeds to compete more on a level basis. It'll be tight but I reckon AMD takes this clock per clock basis, and maybe by even price. They just need a 3GHz Agena X4 part by sometime Q1 when desktop Yorkfield will be releasing 3.16GHz.

Do I give a damn which chip kills what? No, because it kills inside weak minds only. Competition is always healthy, even Intel knows that. If the highest Barcelona quad beats or levels the highest Penryn with highest clock possible by Q2 08, then the same excitement and effort turns out for Nehalem, and so on.

Did pretty much the same thing, but came with some different numbers using the articles last page. If you thow in the B1 stepping that was used, then maybe a little higher, don't really know.

Came to the same conclusion that AMD needs to get to 3G. If they do we may have a nice little price war, which is great for us. Plus Intel may have some tricks of their own, which would mean even faster cpus. All in all, I think we as consumers are going to win bigtime.

[gone_fishin]

Came across some interesting cpu bench scores.
In an article,
http://www.crn.com/hardware/201805044
"The Test Center was provided with an engineering sample server equipped with dual Barcelona CPUs and 16Gbytes of RAM"

and used PasMarks benchtest to obtain a cpumark score of 4693.

Going to Passmarks site, you can see an updated table,
http://www.cpubenchmark.net/cpu_look...45+%40+2.33GHz

Including Xeon 5355 dual cpu (8 core it means) that get 6207 cpumarks.

So AMD = 4693 vs Intel's 6207.
And the highest end Intel 5365 is not on the charts yet.

[zakelwe]

Well , waiting for results still. Considering that this processor has been released and is in final production form it is a very poor showing up to now when it comes to actual results.

As a reference

http://www.xtremesystems.org/forums/...ghlight=conroe

That's a Conroe 4 months before release, not too many bugs or excuses there. Just goes to show how shoddy this release has been. How many more months can it be strung out that AMD will rule ... if you wait just a little bit longer. I am sure I have been hearing this from just before 65nm K8 and that ended in nothing but a massive let down with its poor overclocking and slow cache. Slow cache, hmm, that rings a bell with the K10 ...

As an avid overclocker since the Celeron 300A all I can say is

PUT UP OR SHUT UP.

Or can you not put up AMD merchants? Show us the scores.

Don't bother quoting me and arguing the other side, just put up some earth shaking benchmarks in the desktop space. If Conroe can get good scores 4 months before release then Phenom should be able to provide similar 2 months before if it is that good, or are AMD just dicking around still in the quicksand ?

This is another HD2900XT all over again.

Regards

Andy

[dave_graham]

Well , waiting for results still. Considering that this processor has been released and is in final production form it is a very poor showing up to now when it comes to actual results.

As a reference

http://www.xtremesystems.org/forums/...ghlight=conroe

That's a Conroe 4 months before release, not too many bugs or excuses there. Just goes to show how shoddy this release has been. How many more months can it be strung out that AMD will rule ... if you wait just a little bit longer. I am sure I have been hearing this from just before 65nm K8 and that ended in nothing but a massive let down with its poor overclocking and slow cache. Slow cache, hmm, that rings a bell with the K10 ...

As an avid overclocker since the Celeron 300A all I can say is

PUT UP OR SHUT UP.

Or can you not put up AMD merchants? Show us the scores.

Don't bother quoting me and arguing the other side, just put up some earth shaking benchmarks in the desktop space. If Conroe can get good scores 4 months before release then Phenom should be able to provide similar 2 months before if it is that good, or are AMD just dicking around still in the quicksand ?

This is another HD2900XT all over again.

Regards

Andy

and your point is?

why don't you let the folks who actually know something about the platform get a chance to get test the processors out besides hacking through a list of "Well, Intel did this..." stuff? we get it, you're not impressed....don't really care.

i have been upfront in claiming that B1 cores with their BIOS workaround DO cause performance deficits. BA will eliminate that, so, why don't you shut up and let me (and S7) get back to getting the platforms in line. you don't see me crapping in your threads, so....please, stop trying to attack people.

dave

[Movieman]

Well , waiting for results still. Considering that this processor has been released and is in final production form it is a very poor showing up to now when it comes to actual results.

As a reference

http://www.xtremesystems.org/forums/...ghlight=conroe

That's a Conroe 4 months before release, not too many bugs or excuses there. Just goes to show how shoddy this release has been. How many more months can it be strung out that AMD will rule ... if you wait just a little bit longer. I am sure I have been hearing this from just before 65nm K8 and that ended in nothing but a massive let down with its poor overclocking and slow cache. Slow cache, hmm, that rings a bell with the K10 ...

As an avid overclocker since the Celeron 300A all I can say is

PUT UP OR SHUT UP.

Or can you not put up AMD merchants? Show us the scores.

Don't bother quoting me and arguing the other side, just put up some earth shaking benchmarks in the desktop space. If Conroe can get good scores 4 months before release then Phenom should be able to provide similar 2 months before if it is that good, or are AMD just dicking around still in the quicksand ?

This is another HD2900XT all over again.

Regards

Andy

Ohh Andy, your on a roll today!
I think what happenned is that AMD backed themself into a corner.
They had moved the release date I think twice and knew they couldn't do it again BUT, they still had some issues with the chips so they kept it all quiet as long as possible( great job doing that) with the hope that they would solve the issues before release. It seems they didn't but JUST seems that way.
I think that given 60-90 days they will correct most of the issues and then is when we should make our judgements not today..
Just my 2 cents pal..

[akaBruno]

We are talking about a server chip here folks. Not an overclocking cpu for gaming punks. You don't OC these things!!! We don't even have a solid mobo chipset yet to really see anything to compare. Let alone comparing them to a 1G+ cpu advantage.

C2D is great. No doubt. Give it a rest on your laurels.

[Movieman]

We are talking about a server chip here folks. Not an overclocking cpu for gaming punks. You don't OC these things!!! We don't even have a solid mobo chipset yet to really see anything to compare. Let alone comparing them to a 1G+ cpu advantage.

C2D is great. No doubt. Give it a rest on your laurels.

I beg to differ but yes you can OC server chips and effectively..
Limitations, yes, due to bios constraints mostly, but it can be done and done with excellent results.