Barcelona B3s are just fine

[informal]

Rumours to the contrary are wrong

By Charlie Demerjian: Friday, 11 January 2008, 2:31 PM

THERE ARE A bunch of rumours floating about AMD's B3 stepping parts, saying they are have the same bug as the B2, IE the new step did not fix things. This is categorically not the case, it was, is, and never will be true.

When the rumours started to fly, I called AMD and talked to Pat Moorehead who answered all of my questions. The B3s are not bugged, period. The fix was tested on B2 silicon and verified (Don't ask, the elves who work at the Fraternal Order of Silicon Repair Mythical Creatures Local 17 get antsy if you reveal their secrets).

Basically there is no B4, there won't be a B4, and the next stepping is Cx. The Cx parts are the 45nm Shanghai cores, so the B3 will be the end of the line for 65nm. B3s are going to be sampling in the not so distant future, production in the end of Q1, and retail availability in Q2. That part is a slip.

In summary, B3 is fine. It is a little later than people expect, but what can you do? The fixed chips are on the way, sit tight, it is only a few weeks before this soap opera ends. µ


link (http://www.theinquirer.net/gb/inquirer/news/2008/01/11/barcelona-b3s-fine)

It's inquirer but STILL 10x more credulous than that no name French site with "supposed hush hush AMD "engineer" connections ".

[knightwolf654]

if true.. SWEET!
so 45 by Q3 mabey?

[happychappy]

I trust that French site more than INQ

[Indes]

oooo looks like that french site was stealing the inqs fire?

[Jacky]

if true.. SWEET!
so 45 by Q3 mabey?

no it's not "true". 45nm is 2009. the bugfix is still due in Q2 wether or not there will be another stepping before the bugfix! so what he posts is even worse than anything else. (b3 last stepping? it couldn't get worse thant that if true)

[LowRun]

link (http://www.theinquirer.net/gb/inquirer/news/2008/01/11/barcelona-b3s-fine)

It's inquirer but STILL 10x more credulous than that no name French site with "supposed hush hush AMD "engineer" connections ".

They published that no name French site story right away, probably after reading it here, remember?

http://www.theinquirer.net/gb/inquirer/news/2008/01/10/foot-shot-day

So how does it make them a more reliable site???

And dude, just because you don't know something, it doesn't mean it have no name :rolleyes:

[knightwolf654]

no it's not "true". 45nm is 2009. the bugfix is still due in Q2 wether or not there will be another stepping before the bugfix! so what he posts is even worse than anything else. (b3 last stepping? it couldn't get worse thant that if true)

b3 last stepping is a "good" thing. AMD realizes that in order to get k10 competitive the must shrink it faster. i doubt AMD is going to wait till 2009 to produce 45nm.

[Shintai]

b3 last stepping is a "good" thing. AMD realizes that in order to get k10 competitive the must shrink it faster. i doubt AMD is going to wait till 2009 to produce 45nm.

Seriously...there will be more steppings. Mainly to reduce cost and heat. AMD wont go 45nm in 2009. And even then, do you really think they can do 100% 45nm CPUs? If you didnt know. They are still producing quite a bunch of CPUs on 90nm. Over a year after they said they went to 65nm.

[[XC] gomeler]

Agreed.. I doubt AMD will have 45nm out this year in any sort of quantity to make a fuss about. Look how much trouble they are having K10. Now if they did a K8 45nm shrink I'm sure it'd be out the door already given their experience shrinking it to 65nm.. maybe K8 + MCM would have been a better idea?

[Teqqles]

I'd love to see a 45mn K8, would make a lovely processor for my HTPC ;)

[DeathReborn]

AMD wont go 45nm in 2009.

Are you sure you want to say that?


If B3 still has the bug then being the last stepping would be another in a line of disastrousdecisions by AMD (alongside keeping Hector on the payroll). If it doesn't have the bug then it might give them just enough time to get some 45nm out the door in 09.

[Shintai]

Are you sure you want to say that?


If B3 still has the bug then being the last stepping would be another in a line of disastrousdecisions by AMD (alongside keeping Hector on the payroll). If it doesn't have the bug then it might give them just enough time to get some 45nm out the door in 09.

Was a typo, they will first be 45nm in 2009 and not in 2008 ;-)

[3NZ0]

Great news for AMD, shame it takes so long to do so.

Intel needs to be put back under pressure again, the sooner AMD do so, the better.

[DeathReborn]

Was a typo, they will first be 45nm in 2009 and not in 2008 ;-)

Was just checking ;) When are Intel going to be 100% 45nm? I do believe Intel still manufactures some CPU's on 90nm aswell (Itanium). It will take at least until late 08 for Itanium to be shrunk with Tukwila.

AMD will probably still be making 90nm processors at the beginning of 09, it's where their performance has been lately.

[naokaji]

Was just checking ;) When are Intel going to be 100% 45nm? I do believe Intel still manufactures some CPU's on 90nm aswell (Itanium). It will take at least until late 08 for Itanium to be shrunk with Tukwila.

AMD will probably still be making 90nm processors at the beginning of 09, it's where their performance has been lately.

one can still buy old netburst cpu's....

anyway, while amd sure does need to fix the TLB bug and even if they manage to do so, people have to remember that getting rid of that will not do anything to fix the performance of phenom. the initial reviews of phenom where done without the bios workaround, so performance of a phenom without the TLB bug will be no better than what it was at launch.

[ownage]

It was also said that Retail barcelona's would not have the TLB bug.

[Shintai]

Was just checking ;) When are Intel going to be 100% 45nm? I do believe Intel still manufactures some CPU's on 90nm aswell (Itanium). It will take at least until late 08 for Itanium to be shrunk with Tukwila.

AMD will probably still be making 90nm processors at the beginning of 09, it's where their performance has been lately.

Intel still makes some 1000nm CPUs. So maybe they are abd to compare with. But all regular CPUs are shipped as 65 and 45nm.

Itanium will skip 45nm btw and go straight to 32.

[Nedjo]

its funny to see how Intel boys only want to see what suits them best, and suddenly Charlie who is really the only reliable jurno when it comes to CPU's (regardless of the fact that he is writing for the Inq) dosn't know s*it, and some unkown French site is more reliable!

Oh one more thing 45nm K10 will be out for sales in THIS year!!

[Brother Esau]

Shintai, whats you're position at Intel:rolleyes:

[AbelJemka]

http://www.elitebastards.com/cms/
B3 = no bug :up:

[lukija]

http://www.elitebastards.com/cms/
B3 = no bug :up:

Can't load page?

http://www.elitebastards.com/cms/index.php?option=com_content&task=view&id=517&Itemid=29

This one works for me.

[JAWS]

After reading this potentially worrying story, we contacted AMD to get their take on the matter straight from the horse's mouth. Their response couldn't have been much clearer, as they stated that the story surrounding 'B3' silicon not fixing the TLB issue is "simply not true", and that "B3 silicon does not have the TLB erratum. It is fixed.".

Is this the same AMD that strung us along for months with K10 hype? :rolleyes: the only word that comes to mind thinking about k10 is "broken"

I hope they fix something in K10, not only TLB.

[metro.cl]

Is this the same AMD that strung us along for months with K10 hype? :rolleyes: the only word that comes to mind thinking about k10 is "broken"

I hope they fix something in K10, not only TLB.

I thought the same imagine how low would AMD stock if the B3 wasnt fixed we would never get the real answer from AMD.

Also this is the same Inq that said AMD eng. were dancing around K10 because in one of the first steppings it runned according to plan or even more...

[ryboto]

Is this the same AMD that strung us along for months with K10 hype? :rolleyes: the only word that comes to mind thinking about k10 is "broken"

I hope they fix something in K10, not only TLB.

It was marketing PR. There were instances, albeit few, where the K10 does outperform Core2 by wide margins. In this case, I can't really see how lying about the B3 would be beneficial.

[freeloader]

Fix or no fix. If AMD doesn't turn around by Q3/08 they'll be in Chapter 11 before '09. They simply can't continue to burn through money and miss operating schedules the way they have in the last year. You can't keep going to the well for water, when the well is dry.

[terrace215]

So, AMD has decided to rename B3 to "B2.1", and will call another try "B3". And while the world waits for their (admittedly pushed-back-to-Q2 "B3" parts) they can feast on some new 1.8GHz parts that AMD claims their customers would rather have.

AMD engineers must be DANCING IN THE AISLES over these latest developments.

[BrowncoatGR]

No they just tested the fix on B2 internally and are going to implement it in B3 commercially. At least that's how i understand it. I'm really surprised though that ppl here still haven't realised that Barcelona outperforms both Core2 and Penryn in many server apps by wide margin and that they also scale better when multiple cores are used unless there is no data interdependency and there is limited memory bandwidth needed for the app(i.e. 3d rendering).

[Shintai]

No they just tested the fix on B2 internally and are going to implement it in B3 commercially. At least that's how i understand it. I'm really surprised though that ppl here still haven't realised that Barcelona outperforms both Core2 and Penryn in many server apps by wide margin and that they also scale better when multiple cores are used unless there is no data interdependency and there is limited memory bandwidth needed for the app(i.e. 3d rendering).

I dont know what world you live in. But for dual sockets, AMD is chanceless. With 4 sockets..its a hard battle and they usually lose due to lack of frequency. 8 Sockets..sure.

So for the standard servers deployed. AMD dont win many server applications. Maybe some synthetics. But not real world. And for rendering I hope you are joking. Even dual clovers beat dual barcelonas pretty easily.

[KTE]

ATM no one apart from AMD or some close friends really know any facts about B3 stepping. AMD said they'd fix it. That's the last we know and the rest is just the usual gibberish as is starting to become common with anything that has _AMD_ in the description. Why not just wait and see.
Their circumstance is also obviously clear. They have to fix it, it's not even an option to not have any major desktop/server/workstation shippings after 2 years of borrowed cash spent, 8 months after official new architecture release.

[Face]

http://hardocp.com/news.html?news=MzAwODMsLCxoZW50aHVzaWFzdCwsLDE=

Contrary to other reports, the new B3 version of the Phenom processor does fix the TLB errata issue. You can take that to the bank.


http://www.betanews.com/article/AMD_delays_more_CPU_rollouts_this_time_quadcore_Ph enoms/1200080044

AMD delays more CPU rollouts, this time quad-core Phenoms

"We expect to launch our AMD Phenom triple-core product in Q1 and our AMD Phenom 9700 and 9900 quad core processors in Q2 2008," an AMD spokesperson told BetaNews today.

[hollo]

well, it'd be amazing if b3 was also bugged, i could barely believe it when i read the other thread.
it'd be equally amazing if b3 was the last 65nm k10 stepping

[BrowncoatGR]

I dont know what world you live in. But for dual sockets, AMD is chanceless. With 4 sockets..its a hard battle and they usually lose due to lack of frequency. 8 Sockets..sure.

So for the standard servers deployed. AMD dont win many server applications. Maybe some synthetics. But not real world. And for rendering I hope you are joking. Even dual clovers beat dual barcelonas pretty easily.

http://www.anandtech.com/showdoc.aspx?i=3162&p=9
http://www.anandtech.com/showdoc.aspx?i=3162&p=10
http://www.anandtech.com/showdoc.aspx?i=3162&p=11 (the second bench. fritzchess is really irrelevant)
http://connexitor.com/blog/pivot/entry.php?id=191
:welcome: the Mysql results alone are amazing
Also you misunderstood me. I meant that in tasks that are light on data interdependency and dont require much bandwidth C2Q scales better than usual and is faster than K10.

[Shintai]

http://www.anandtech.com/showdoc.aspx?i=3162&p=9
http://www.anandtech.com/showdoc.aspx?i=3162&p=10
http://www.anandtech.com/showdoc.aspx?i=3162&p=11 (the second bench. fritzchess is really irrelevant)
http://connexitor.com/blog/pivot/entry.php?id=191
:welcome: the Mysql results alone are amazing
Also you misunderstood me. I meant that in tasks that are light on data interdependency and dont require much bandwidth C2Q scales better than usual and is faster than K10.

Unfortunately synthetic benching is usually wrong. Now lets see what Anand have to say, when he bench for the other company so to say.

http://www.anandtech.com/IT/showdoc.aspx?i=3099&p=6

Now this is MSSQL and not MySQL tho.

However you can see its not always as black and white. Real world knowledge and test is abit more helpful.

And another note, I dont even think they used the same disksystems and controllers. Apples to oranges in the serverspace where the storage system is half the server for SQL.

[BrowncoatGR]

Unfortunately synthetic benching is usually wrong. Now lets see what Anand have to say, when he bench for the other company so to say.

http://www.anandtech.com/IT/showdoc.aspx?i=3099&p=6

Now this is MSSQL and not MySQL tho.

However you can see its not always as black and white. Real world knowledge and test is abit more helpful.
Well MySQL is not a synthetic, neither is OpenLDAP. Even in the bench you posted the Xeon has an advantage of up to 27% while it sports a 33% faster clock. Since K10 seems to scale rather well with clockspeed one would assume that at the same clockspeed K10 is actually at least slightly faster. I would actually like to see some Oracle and Postgresql numbers for both architectures.

[Zytek_Fan]

Good to hear. More options for the people.
And hopefully something competitive.

[Shintai]

Well MySQL is not a synthetic, neither is OpenLDAP. Even in the bench you posted the Xeon has an advantage of up to 27% while it sports a 33% faster clock. Since K10 seems to scale rather well with clockspeed one would assume that at the same clockspeed K10 is actually at least slightly faster. I would actually like to see some Oracle and Postgresql numbers for both architectures.

You telling me it wasnt tested and done with synthetic queries? Just looping the same thing over and over? Also you should know things dont scale well. Not even for AMD either. Especially not when it might be disk, memory etc limited. You simply dont scale well nomatter who you are when just making 1 of many components faster.

And tell me please, what was the disksystem used in the bench you linked to? What controllers was used?

[DeepThought86]

You telling me it wasnt tested and done with synthetic queries? Just looping the same thing over and over? Also you should know things dont scale well. Not even for AMD either. Especially not when it might be disk, memory etc limited. You simply dont scale well nomatter who you are when just making 1 of many components faster.

And tell me please, what was the disksystem used in the bench you linked to? What controllers was used?

http://www.anandtech.com/IT/showdoc.aspx?i=3099&p=5

Same RAID controller and drives used. And do you really think benchmark writers are so stupid that they run the same queries again and again? 65nm or 45, Intel's unbalanced mobile-derived chips do great on single-threaded tasks and start to falter in server tasks

[accord99]

http://www.anandtech.com/IT/showdoc.aspx?i=3099&p=5

Same RAID controller and drives used. And do you really think benchmark writers are so stupid that they run the same queries again and again? 65nm or 45, Intel's unbalanced mobile-derived chips do great on single-threaded tasks and start to falter in server tasks
Unbalanced or not, these Intel chips' combination of raw power, reasonable power consumption and ability to clock at high speeds give them an overwhelming performance advantage in the vast majority of applications, single-threaded or otherwise. So much so that AMD has yet to release a single benchmark on a major enterprise benchmark for a Barcelona system.

[informal]

http://www.anandtech.com/IT/showdoc.aspx?i=3099&p=5

Same RAID controller and drives used. And do you really think benchmark writers are so stupid that they run the same queries again and again? 65nm or 45, Intel's unbalanced mobile-derived chips do great on single-threaded tasks and start to falter in server tasks

Don't even try to argue with him,you'll end up the same as freeloader did(look at my sig for colorful description :p: )

[freeloader]

I dont know what world you live in. But for dual sockets, AMD is chanceless. With 4 sockets..its a hard battle and they usually lose due to lack of frequency. 8 Sockets..sure.

So for the standard servers deployed. AMD dont win many server applications. Maybe some synthetics. But not real world. And for rendering I hope you are joking. Even dual clovers beat dual barcelonas pretty easily.

Servers are more dependant on disk I/O than anything else. All things being equal, you'd be hard pressed to tell the difference between an AMD or Intel server.

[Donnie27]

Montvale was supposed to be 65nm but it shipped as 90nm after a year delay. As was said Itanium will skip 45 and go straight to the 32nm Poulson.

http://www.news.com/Big-Itanium-change-coming-with-Poulson/2100-1006_3-6047449.html

[Zytek_Fan]

How large is the market for Itanium these days? I don't see it being tremendously popular...

[Shintai]

How large is the market for Itanium these days? I don't see it being tremendously popular...

You dont see Sparc and Power series being popular either. Yet its a multibillion revenue business with very high margins.

It said that in the first quarter of this year, Itanium revenue was 54% of UltraSparc's and 41% of Power's. In 2002, Itanium revenue was just 1% of both UltraSparc and Power revenue, and by 2006, it had grown to 40% and 26%, respectively.

They estimate the revenue for Itanium this year will be above 5 billion$ and 6.6 billion$ in 2009. Or about the same as AMDs entire revenue this year for Itanium...

[Monkeywoman]

anyone know if any site has done any benchies with the B3 revision? i think they should be up to par with the intel chips; clock for clock.

[BrowncoatGR]

You telling me it wasnt tested and done with synthetic queries? Just looping the same thing over and over? Also you should know things dont scale well. Not even for AMD either. Especially not when it might be disk, memory etc limited. You simply dont scale well nomatter who you are when just making 1 of many components faster.

And tell me please, what was the disksystem used in the bench you linked to? What controllers was used?

Im telling you i dont know and neither do you, because its not mentioned. As for scaling, if we dont get any real results with the same systems we are both just speculating, in particular since there are only a few graphs available, no raw data and no slower or faster parts for either architecture.

[STaRGaZeR]

anyone know if any site has done any benchies with the B3 revision? i think they should be up to par with the intel chips; clock for clock.

Why? The architecture is the same.

[donitsi]

Well atleast it would be a closer match if they could get the nb speed to run at cpu speed.

[informal]

Why? The architecture is the same.

If one TLB bug patched with BIOS can decrease perf. by 5-30%,what would be the effect of 1 or 2 serious BIOS patched bugs in nortbridge part of K10?
We now know that the TLB one is hard fixed and will not impact performance like the patch does,do you think AMD would miss the chance with B3 to address one or two serious bugs elsewhere?

[STaRGaZeR]

If one TLB bug patched with BIOS can decrease perf. by 5-30%,what would be the effect of 1 or 2 serious BIOS patched bugs in nortbridge part of K10?
We now know that the TLB one is hard fixed and will not impact performance like the patch does,do you think AMD would miss the chance with B3 to address one or two serious bugs elsewhere?

Of course not, but again, the perfomance is expected to be the same, but without bugs. The same of Phenom without the TLB fix. There is no perfomance gain in adressing something that causes crashes, not perfomance penalties.

[Monkeywoman]

the B3s should also overclock higher and use less energy...at least thats the case with lintel stepping's

[informal]

Of course not, but again, the perfomance is expected to be the same, but without bugs. The same of Phenom without the TLB fix. There is no perfomance gain in adressing something that causes crashes, not perfomance penalties.

I think you misunderstood me.
There are other patched bugs by current BIOSes.You(as general public) not knowing this is nothing uncommon.The only difference with TLB is that it was discovered too late and had to be publicized since systems weren't equipped with microcode updates to address the errata on time(and systems were locking up in certain situations;the fix is penalizing everything,not only the VM intensive stuff)
I'm not saying that addressing these will bring anything to the table,i'm saying AMD wouldn't do another respin of the silicon(B3) and miss the opportunity to fix anything else(check K10 errata list,it's not short).

[donitsi]

(check K10 errata list,it's not short).

Where can i find this, linky? Thanks

[informal]

Here (http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/41322.pdf)is the one and only(not containing the latest "TLB bug")

[onewingedangel]

Unless b3 also allows k10 to clock significantly higher, fixing the TLB bug alone doesn't do a whole lot to improve AMD's chances - so if b3 is the final 65nm revision lets hope its a good one (and clocks at least 10% faster than current revisions)

[Shintai]

the B3s should also overclock higher and use less energy...at least thats the case with lintel stepping's

Thats not right. B1-B2-B3 all used the same power and overclocked just the same. It was first G0 that changed.

Same is true for AMD. Also taking their K8 etc history into it.

[STaRGaZeR]

I think you misunderstood me.
There are other patched bugs by current BIOSes.You(as general public) not knowing this is nothing uncommon.The only difference with TLB is that it was discovered too late and had to be publicized since systems weren't equipped with microcode updates to address the errata on time(and systems were locking up in certain situations;the fix is penalizing everything,not only the VM intensive stuff)
I'm not saying that addressing these will bring anything to the table,i'm saying AMD wouldn't do another respin of the silicon(B3) and miss the opportunity to fix anything else(check K10 errata list,it's not short).

No, I think this is not a misunderstanding... I was talking earlier about Monkeywoman post, about perfomance. You replied to that, and I completely agree with you that AMD will not miss the chance to adress a few bugs. I then, I said those bugs don't affect perfomance. Your are talking about those bugs now, and I repeat they have nothing to do with perfomance. The only bug (AFAIK) that drops perfomance is the TLB one, and that will be hard fixed in the B3 revision of the silicon.

And please, don't act like some kind of super-specialist, because you aren't, like many others here, because someone with more knowledge can pwnt you, and thas has already happen before. The bugs adressed by microcode updates are commonly known by everyone I have contact with.

Anyway, the B3 respin won't bring anything new to the table. Same perfomance of B2 without TLB patch, TLB and another bugs fixed. AMD needs way more than that to compete with Intel.

[informal]

And please, don't act like some kind of super-specialist, because you aren't, like many others here, because someone with more knowledge can pwnt you, and thas has already happen before. The bugs adressed by microcode updates are commonly known by everyone I have contact with.


Ah just lol...
You seem to know a lot about other people you don't actually know and you like posting things with no basis in reality.Quite amusing :)

[Shintai]

So Informal..whats the latest hype?

Huge speed boost? Massive clock increase?

Or perhaps the more realistic info is still dead in the water?

[savantu]

B3 will bring Warp speed to Wordpad...

[STaRGaZeR]

Ah just lol...
You seem to know a lot about other people you don't actually know and you like posting things with no basis in reality.Quite amusing :)

Maybe I miss it in your post, but where is your answer to "Barcelona B3s are just fine", or to my post? Oh wait, a YAIOH again (Yet Another Informal Offtopic or Hype) :ROTF:

Seriously, who is the one with no basis in reality? :)

[hyc]

Servers are more dependant on disk I/O than anything else. All things being equal, you'd be hard pressed to tell the difference between an AMD or Intel server.

Not all workloads are equal though. The OpenLDAP tests are for a read-only authentication workload, which is a pretty common real-world application for LDAP. In these tests, where the entire DB fits in RAM, disk speed is irrelevant.

Once you get the disk involved, I suspect the magnitude of difference in speeds (disk vs RAM) totally overshadows the CPU characteristics. E.g. these tests

http://connexitor.com/blog/pivot/entry.php?id=193
http://connexitor.com/blog/pivot/entry.php?id=185

A lot of it comes down to the software design; you're no longer seeing a meaningful test of the CPUs... Though it also touches on another point - when you have such large working sets, 2MB vs 4MB vs 6MB of on-chip cache isn't going to make any difference.