More of the same....

[Manny Calavera]

http://www.overclockers.com/tips01166/

None of the partners we talked to are really impressed with Barcelona yet, but then again none of the true potential of the chip and chipsets have been realized due to CPU stepping issues, BIOS maturity, and generally low clock speeds. AMD simply hasn't gotten the process under control yet and after hearing our friends at the motherboard companies talk, AMD is close to near/total panic mode right now as the Q3 Barcelona product launch schedule rapidly approaches.

http://www.overclockers.com/tips01165


http://www.anandtech.com/tradeshows/showdoc.aspx?i=3006

This Anandtech article is damning,to say the least !



Omg,When is hectors head going to be found on the end of a pike ?



http://www.anandtech.com/tradeshows/showdoc.aspx?i=3006

In the end, performance was absolutely terrible. We're beginning to understand why AMD didn't let us test Barcelona last month. It's not that AMD is waiting to surprise Intel; it's that the platform just isn't ready for production yet.

 

[Scali2]

Yea, it's all coming out now, just like I've always said.
People had so much faith in AMD that they just couldn't believe I was right.
I suppose it's time for an "I told you so".

 

[quadnad]

Just like anandtech said, the hardware they had wasn't production ready, and it was apparent that it wasn't ready to be benchmarked on. That doesn't mean Barcelona is suddenly a failure, it just means that the platform isn't ready yet.

We'll see what happens in the next few months.

 

[duby229]

Except you've got anandtech, and overclockers...

Ed is the biggest Intel fanboy in this industry. He has been predicting that AMD is going to die for years now. He hasnt been right yet. There is absolutely no reason to believe that he will be right this time.

Anand is pissed that ever since he screwed AMD last year, they stopped giving him preferential treatment. They didnt give him a chip becouse of his clear bias. So he went to Taiwan and pressured some poor shmuck into giving him a chip. Except what Anand is too stupid to know is he got screwed. He got himself an early sample that wont work without a heavily modified BIOS. Of course Anand will release a about 10 benches and 9 of them will be games at 640x480 and 1 will be superpi... He will then claim that they are then end all be all. And he'll do it while claiming that his chips are representative.

 

[Scali2]

Just like anandtech said, the hardware they had wasn't production ready, and it was apparent that it wasn't ready to be benchmarked on. That doesn't mean Barcelona is suddenly a failure, it just means that the platform isn't ready yet.

We'll see what happens in the next few months.

Sadly for AMD, you can't just turn a chip around completely in a few months.
It takes years to design a complex chip like this.
In a few months you can only do minor tweaks/bugfixes and try to get your yields up and increase the clockspeed.
The IPC is roughly 40% lower than what AMD claimed. This doesn't sound like something they can fix in a few months.
The clockspeed is also about 30% lower than what AMD claimed. Now they may be able to fix that in a few months, but I wouldn't hold my breath. Still leaves them with 40% IPC to compensate for.

 

[quadnad]

Sadly for AMD, you can't just turn a chip around completely in a few months.
It takes years to design a complex chip like this.

I'm not saying the chip isn't ready, I'm saying the fact that in order to get it working you must use massively modified BIOSes which results in crippled performance.

the current boards required some heavy BIOS work before the new chips would even work, much less perform up to the expectations set by AMD.

The motherboard we tested on had minimal HT functionality and wouldn't run at memory speeds faster than DDR2-667; most 3D video cards wouldn't even work in the motherboard. Memory performance was just atrocious on the system, but the motherboard manufacturers we worked with attributed this to BIOS tuning issues that should be fixed in the very near future.

We'll see where performance is when everything is sorted out (including the clock speeds being where they should be).

 

[Scali2]

I'm not saying the chip isn't ready, I'm saying the fact that in order to get it working you must use massively modified BIOSes which results in crippled performance.

I doubt that. BIOS doesn't have that much control over the internal execution core of a processor. Especially in this case, where even the memory controller is on the CPU.

We'll see where performance is when everything is sorted out (including the clock speeds being where they should be).

I've already seen it.

 

[quadnad]

I've already seen it.

Well it's a good thing you've got your future glasses on, since I'm stuck here in the present. The Cinebench scores don't nearly count as knowing how things will pan out, either.

Again, I'm going to reserve judgment until everything is being reviewed with properly functioning hardware.

 

[Scali2]

Well it's a good thing you've got your future glasses on, since I'm stuck here in the present.

Again, I'm going to reserve judgment until everything is being reviewed with properly functioning hardware.

Sounds more like you're stuck in fantasy-land to me.

Again, I'm going to reserve judgment until everything is being reviewed with properly functioning hardware.

I'm not. What I've seen so far is so close to release/final product that there's no way AMD can make any big changes now.
Just recall how well the Conroe ran, months before it was released. It ran at the expected speeds, and performance of those early samples was as good as the final product.
Again, Intel has shown a 'finished' Penryn months ago, while Penryn was supposed to be released after Barcelona.
What AMD has now is either as mature as it's going to get, or they will have to delay the release and try again.
But that will probably mean that by the time they get it up to the performance they promised, that performance is no longer competitive.

 

[quadnad]

Sounds more like you're stuck in fantasy-land to me.

I just prefer to judge a product when it's functioning properly.

 

[Scali2]

I just prefer to judge a product when it's functioning properly.

I'm saying it is functioning 'properly'.
Wanna make a little wager on it?
As I said above, they're not going to fix the 40% IPC problem... They may be able to clock it up to ~2.3-2.5 as they originally intended, but only with roughly the same IPC as they have today (so comparable to Kentsfield).
If I'm right, and the CPUs are ~2.5 GHz or lower at introduction, and perform roughly like Kentsfield clock-for-clock, I'll buy you one.
If I'm wrong, and the CPUs are actually competitive with the fastest Intel CPUs at the time, as AMD promised, you buy me one.

 

[quadnad]

I'm saying it is functioning 'properly'.
Wanna make a little wager on it?
As I said above, they're not going to fix the 40% IPC problem... They may be able to clock it up to ~2.3-2.5 as they originally intended, but only with roughly the same IPC as they have today (so comparable to Kentsfield).
If I'm right, and the CPUs are ~2.5 GHz or lower at introduction, and perform roughly like Kentsfield clock-for-clock, I'll buy you one.
If I'm wrong, and the CPUs are actually competitive with the fastest Intel CPUs at the time, as AMD promised, you buy me one.

hah, I don't think I like the idea of potentially losing money, much less if I'm right (?). Anyway, all I'm saying is that according to Anand and AMD's partners, the platform isn't ready. It may not perform up to AMD's expectations, but because everyone is saying that the platform isn't ready (and this is significantly holding the processor back) I don't think it makes sense to "call it" now.

I'm going to wait until someone has full benchmarks across the system using fully functioning hardware that doesn't require "heavy BIOS modifications" and suffers from "atrocious memory performance" etc. etc.

 

[Lou Ceifer]

I'm saying it is functioning 'properly'.
Wanna make a little wager on it?

There is a BIG difference Scali between a piece of silicon actually functioning and not being fully capable of it's intended design parameters,(i.e. a ES sample) please learn the difference if you haven't by now. Bottom line folks, ES SAMPLES DO NOT PROVIDE A CLEAR INDICATION OF A FINAL PRODUCT'S PERFORMANCE. So my friendly advice is spare us the "armchair quarterbacking" and wait to see what a production piece of silicon will ACTUALLY do before you go "proclaiming almighty judgement", much less proclaiming AMD's demise based on the performance of crippled ES samples. (this goes for everyone!)

 

[Scali2]

hah, I don't think I like the idea of potentially losing money, much less if I'm right (?).

Oh yea, I got that backwards
Loser pays ofcourse.

Anyway, all I'm saying is that according to Anand and AMD's partners, the platform isn't ready. It may not perform up to AMD's expectations, but because everyone is saying that the platform isn't ready (and this is significantly holding the processor back) I don't think it makes sense to "call it" now.

"Isn't ready" is talking about the clockspeed. Their 65 nm process isn't up to 'native' quadcores yet.
They can't do much about the architecture, so IPC won't change much. And even if they do reach the promised ~2.5 GHz, they won't be competitive. Fixing IPC to get 40% gain will mean they have to redesign the whole thing, and they can't possibly delay the processor that long. Better to have the CPU on the market and selling than to tweak it to be competitive now, but release it when that ship has sailed.
So AMD won't delay for more than a few months, and they won't be able to do much more than ramp up clockspeed and/or increase stability/reliability.

In short, we know all we need to know... It's going to be a failure no matter how you look at it.

 

[Scali2]

Bottom line folks, ES SAMPLES DO NOT PROVIDE A CLEAR INDICATION OF A FINAL PRODUCT'S PERFORMANCE.

Then why do they for Intel, and why are they never crippled at Intel?
Even though Intels are released further ahead of release?
In a few months AMD may be able to do one or two more steppings, no more. The fact that these ES are crippled speaks volumes.

 

[Digital Viper-X-]

I'm saying it is functioning 'properly'.
Wanna make a little wager on it?
As I said above, they're not going to fix the 40% IPC problem... They may be able to clock it up to ~2.3-2.5 as they originally intended, but only with roughly the same IPC as they have today (so comparable to Kentsfield).
If I'm right, and the CPUs are ~2.5 GHz or lower at introduction, and perform roughly like Kentsfield clock-for-clock, I'll buy you one.
If I'm wrong, and the CPUs are actually competitive with the fastest Intel CPUs at the time, as AMD promised, you buy me one.

even your source at anandtech said it was the mobos bios that is mostly crippling barcelona, and the second contributer was the lower clock speed, but that is only for the Server based chips, even he said Agena will be clocked higher

so why do you go on this rant, are you the ANTI AMD COALITION chairperson?
relax and wait till some concrete benchmarks are out for the chip on a FUNCTIONING platform

the bios still controls alot of VITAL things, like thermal management, power management etc

 

[Powerhouse2k]

It seems that people on these forums really don't know how 'chips' used in computers (as well as other devices) are actually made.

I worked for a 'chip' mfg (Broadcom), and there were plenty of their 'chips' that while functioning properly, had something not working correctly on the actual chip die, and they needed a new build (sometimes called stepping) from the mfg to fix the issue. Builds are different from steppings, in that usually builds are fixing a problem in the mfg process, while a stepping can actually add some additional functionality to the CPU (more cache, tweaked registers, lower thermal envelope, etc).

I see the same issues with both Intel and AMD all the time. Simply put, AMD might have a "functioning properly" Barcelona, that has an issue with it's onboard cache, or a register that doesn't work properly. The CPU still is able to run and process data, but isn't able to do so efficently because of this problem. You don't have to look to far back, to remember the first AMD 'Hammer' CPU that ran at 800Mhz (similar to the Barcelona we see now), while Intel had CPU's running in the 2+Ghz range. When AMD finally released the 'Hammer' based CPU's they were running just a little slower in clockspeed than Intel's, but had a huge performance increase over them.

Will the same thing happen again with Barcelona? We just don't know at the moment. But until I see benchmarks from the [H]ard, or other trustworthy websites, I wouldn't say that AMD sucks and Intel is King, based off the information I'm reading today.

Do I care if AMD's CPU's suck or are great? Well yes, but not because I'm a fanboy. It's because competition in the CPU industry has been very good for the computer industry. Because of AMD, Intel has had to rethink their CPU strategy and release more powerful processors quicker than ever before to compete against AMD. Remember when Intel would just release a 'New CPU' with a slight bump in clockspeed and that was it? Thankfully those days are long gone, due in part to the competition from AMD.

That being said, AMD isn't doing to well financially currently, and if Barcelona isn't competitive with Intel's stuff, AMD will have a huge problem on it's hand, possibly having to leave the CPU market. If this happens, I could see Intel going back to its old ways since it wouldn't have the competiton to worry about anymore. This would be bad for the computer industry as a whole, and is why a lot of people (myself included) are rooting for AMD to release something competitve against Intel.

 

[Powerhouse2k]

What I've seen so far is so close to release/final product that there's no way AMD can make any big changes now.
Just recall how well the Conroe ran, months before it was released. It ran at the expected speeds, and performance of those early samples was as good as the final product.
Again, Intel has shown a 'finished' Penryn months ago, while Penryn was supposed to be released after Barcelona.
What AMD has now is either as mature as it's going to get, or they will have to delay the release and try again.
But that will probably mean that by the time they get it up to the performance they promised, that performance is no longer competitive.

Actually, AMD can make a big change, that being with their partner IBM. Perhaps you might recall IBM a few weeks ago mentioning that their new CPU's are running at 4.7+Ghz (up from the 2.6Ghz in the initial generation of the their CPU).http://www.dailytech.com/article.aspx?newsid=7384

If you read the details about this, it's because they were able to work out some problems with their design and fixed some problems with the mfg process. Since AMD and IBM are partners, it's not unrealistic to bet that AMD might licence these new process' from IBM to allow their CPU's to run just as fast.

 

[pxc]

IBM had an initial Power 6 target of 5GHz, so 4.7GHz is a little off. AMD had a target of under current K8 speeds and it having trouble meeting goals in the 2S space. Uniprocessor versions were being shown at higher frequencies.

The Power 6 and Barcelona are completely different designs so i'm not sure how IBM can "help" in this case, especially in a design that's already finished.

 

[Powerhouse2k]

Part of IBM's help would be in a their new versions of copper interconnects on the CPU die, as well as High-K, and Strained Silicon. All of these contributed to lowering the heat generated by the CPU, allowing IBM to increase the Ghz.

So while the CPU design stays the same, the material used would be 'new and improved'.

 

[Scali2]

even your source at anandtech said it was the mobos bios that is mostly crippling barcelona, and the second contributer was the lower clock speed, but that is only for the Server based chips, even he said Agena will be clocked higher

Yes, but I don't consider Anandtech very knowledgeable on the topic of lowlevel BIOS/CPU architecture. I'm quite sure I have a better insight on the situation than they do.

so why do you go on this rant, are you the ANTI AMD COALITION chairperson?
relax and wait till some concrete benchmarks are out for the chip on a FUNCTIONING platform

I could say the same about you. You are the one calling me names here, when you have absolutely no proof that I could be wrong.

the bios still controls alot of VITAL things, like thermal management, power management etc

Yes, and it's highly unlikely that any of those features would be crippling performance, because they would simply be turned off if that was the case. They're probably turned off anyway, when they're benchmarking the maximum performance.

 

[pxc]

So while the CPU design stays the same, the material used would be 'new and improved'.

Adding those things would be a new revision of the CPU, something that usually takes AMD 12-18 months.

 

[Scali2]

You don't have to look to far back, to remember the first AMD 'Hammer' CPU that ran at 800Mhz (similar to the Barcelona we see now), while Intel had CPU's running in the 2+Ghz range. When AMD finally released the 'Hammer' based CPU's they were running just a little slower in clockspeed than Intel's, but had a huge performance increase over them.

Yes, and the IPC was virtually unchanged from the early 800 MHz Hammers to the latest.
As I've been saying in this thread, it's the same this time. They are trying to ramp up clockspeed, but the IPC is a lost cause at this point. The gap (40%) is too large for them to fix in this short period of time.
Their production may not be consistent yet, but they're hand-picking these engineering samples. So, if some of them have faulty cache or something, they can still pick out a few with working cache. If they all have faulty cache, it's a structural problem, and they need to overhaul the chip's design, which takes time... Time that AMD doesn't have. At this point it's quite certain that the chips are functioning properly except perhaps for some minor issues, else they would never have made it out of AMD's labs in the first place, and AMD would have put the release dates further ahead. Instead they're telling their customers like Cray that the chips will arrive shortly.

Will the same thing happen again with Barcelona? We just don't know at the moment. But until I see benchmarks from the [H]ard, or other trustworthy websites, I wouldn't say that AMD sucks and Intel is King, based off the information I'm reading today.

Probably, but I have better sources.

Remember when Intel would just release a 'New CPU' with a slight bump in clockspeed and that was it? Thankfully those days are long gone, due in part to the competition from AMD.

That's not true at all.
By the time AMD came to the game, Intel was already making 486 processors.
So in short:
- Intel invented the microprocessor with the 4-bit 4004
- Intel then created 8-bit and 16-bit variations, the 16-bit one being the most famous: the 8086, which is the reason why we still call the CPUs x86 today.
- Intel then developed the 80286 with a protected mode.
- Intel extended the instructionset to 32-bit with the 80386.
- Intel integrated the x87 coprocessor in the 80486, which had previously been sold separately.

All the while they both improved IPC and clockspeed with every generation. All this before AMD even debuted its own x86-clone.
After AMD joined, even though they weren't very competitive, except for the low-budget region, which was basically Intel's last-generation, Intel continued to release new architectures and push performance further:
- Intel releases the Pentium, the first superscalar x86 implementation, which was a huge culture-shock in terms of performance. 64-bit databus, large and fast caches, massively improved FPU performance, and two ALU pipelines that basically still hold the record in terms of IPC for most common instructions.
- Intel releases the Pentium Pro, a radical new design with full OOO, improving IPC especially on compiler-generated code, and allowing for far higher clockspeeds.
- Intel introduces SIMD instructionsets in the form of MMX and later SSE.

Only then does AMD start to become competitive with its K7/K8... but really, what has AMD innovated here? In the history of x86, and CPUs in general, Intel has done most of the work. Okay, so the Pentium 4 wasn't such a big hit in terms of performance... Even so, it's only a very small part of the total Intel history, and the Pentium 4 did have a lot of innovative technologies... It just didn't work out. Doesn't mean Intel didn't put any effort in it.

 

[Scali2]

Adding those things would be a new revision of the CPU, something that usually takes AMD 12-18 months.

My point exactly. If AMD does that, they won't be up against Penryn, but against Nehalem, and it won't look pretty.
Not to mention that they won't be selling many CPUs in the meantime, because after next month's pricecuts, quadcores are entering the mainstream, and AMD has nothing to compete.
They're better off with a slow quadcore than with no quadcore, so AMD isn't going to delay the CPU for 12-18 months. They're going to patch it up a tad in a delay of 2-3 months, then it's off.

 

[Scali2]

Actually, AMD can make a big change, that being with their partner IBM. Perhaps you might recall IBM a few weeks ago mentioning that their new CPU's are running at 4.7+Ghz (up from the 2.6Ghz in the initial generation of the their CPU).http://www.dailytech.com/article.aspx?newsid=7384

You know why CPUs like POWER6 and Pentium 4 run at such high clockspeeds?
Because they were *designed* that way.
They have long pipelines with very simplified logic (IBM even gave up OOO altogether).
AMD uses a completely different design, which is very similar to their current design, and is therefore physically incapable of running anywhere near 5 GHz in the near future.
They're aiming for low clockspeeds and high IPC (more or less like Intel's competitor to the POWER6, the Itanium2).
Even AMD themselves expected it to run at lower speeds than K8. They quoted figures of 2.3-2.5 GHz. Apparently they're having trouble to even reach that, at the moment... so 5 GHz is WAY off.

If anything, the Pentium 4 proved that clockspeed is not the way to go for x86 (at least, not now. Breakthroughs in manufacturing technology could change that overnight. By the looks of it, Core2 is already growing towards Pentium 4). POWER6 probably has different rules, although we don't know for sure, since IBM is the only one making them. If Athlon64 didn't exist, nobody would know that a design with lower clockspeed and higher IPC would give better absolute performance.

 

[Powerhouse2k]

Adding those things would be a new revision of the CPU, something that usually takes AMD 12-18 months.

Are you sure? CPU revisions usually involve a design change, or a process shrink. Since the CPU design wouldn't change, and they aren't switching to a 45nm process, this could be a relatively simple change.


So are you an engineer working for IBM or AMD that could say that it would take that long? Not like I am either, just making a point.

Their production may not be consistent yet, but they're hand-picking these engineering samples. So, if some of them have faulty cache or something, they can still pick out a few with working cache. If they all have faulty cache, it's a structural problem, and they need to overhaul the chip's design, which takes time... Time that AMD doesn't have. At this point it's quite certain that the chips are functioning properly except perhaps for some minor issues, else they would never have made it out of AMD's labs in the first place, and AMD would have put the release dates further ahead. Instead they're telling their customers like Cray that the chips will arrive shortly.

Actually, having engineering samples, could mean they all come from the same batch, which they could all have the same problems. Again, might not be an actual design problem, so much as a Mfg problem. The CPU's that AMD gave to their partners could have been given so they (the partners) could fix their motherboards to accept and work with them, and not meant to benchmark with (hence the lower clock speeds), which seems to be the case if you read the Anandtech article again:

"But the demos that AMD ran were on its own motherboards, not on motherboards from its partners. AMD's partners just recently received their first "production quality" Barcelona samples, and as expected, the current boards required some heavy BIOS work before the new chips would even work, much less perform up to the expectations set by AMD.

The motherboard we tested on had minimal HT functionality and wouldn't run at memory speeds faster than DDR2-667; most 3D video cards wouldn't even work in the motherboard. Memory performance was just atrocious on the system, but the motherboard manufacturers we worked with attributed this to BIOS tuning issues that should be fixed in the very near future."

With this information, any CPU from any company wouldnt be shown in a good light. Again, I'll wait till the parts get in the hands of websites like the [H]ard before I make any kind of call on how good or bad the CPU is.

Probably, but I have better sources.

Well, my sources say that Barcelona (Server) will be released in July at 2.6Ghz and the Agena (Desktop) will be out in December at 2.9Ghz. Oh, and my sources can beat up your sources too...

 

[pxc]

Are you sure? CPU revisions usually involve a design change, or a process shrink. Since the CPU design wouldn't change, and they aren't switching to a 45nm process, this could be a relatively simple change.

Changing the transistors the way you suggested would require tweaks thoughout the whole chip.

 

[Scali2]

Actually, having engineering samples, could mean they all come from the same batch, which they could all have the same problems.

Faulty batches SURELY are not going to leave AMD's labs and end up in the hands of big review sites.

Again, might not be an actual design problem, so much as a Mfg problem. The CPU's that AMD gave to their partners could have been given so they (the partners) could fix their motherboards to accept and work with them, and not meant to benchmark with (hence the lower clock speeds), which seems to be the case if you read the Anandtech article again

First of all I think it is VERY unlikely that AMD sends 'broken' CPUs out to its partners. Secondly, even if they do, I'm sure they would do it with a huge NDA attached to it ("I can let you benchmark this CPU, but then I'm going to have to kill you", or something to that effect).
Thirdly, that leaves us the question where the GOOD CPUs from AMD are? Especially since they're so near their planned release date. Look at Intel, and how they can demo perfectly working ES chips months before actual release.
This close to release, AMD should already be in full production and trying to build up stock for when the release is up. If they can't even get a few proper samples to their partners, something stinks.

"But the demos that AMD ran were on its own motherboards, not on motherboards from its partners. AMD's partners just recently received their first "production quality" Barcelona samples, and as expected, the current boards required some heavy BIOS work before the new chips would even work, much less perform up to the expectations set by AMD.

And what do you think "production quality" means? Exactly. Not flawed samples from bad batches.

With this information, any CPU from any company wouldnt be shown in a good light. Again, I'll wait till the parts get in the hands of websites like the [H]ard before I make any kind of call on how good or bad the CPU is.

The real question is why there are all these lousy excuses, but no proper chips, when you actually believe they're going to be released in about a month.
1 + 1 = 2.

Well, my sources say that Barcelona (Server) will be released in July at 2.6Ghz and the Agena (Desktop) will be out in December at 2.9Ghz. Oh, and my sources can beat up your sources too...

My sources have pre-ordered these chips, and got a negative answer from AMD about whether they'd ship on time.
So my sources have had to tell their customers that their orders are going to be late.

 

[dexvx]

Actually, AMD can make a big change, that being with their partner IBM. Perhaps you might recall IBM a few weeks ago mentioning that their new CPU's are running at 4.7+Ghz (up from the 2.6Ghz in the initial generation of the their CPU).http://www.dailytech.com/article.aspx?newsid=7384

Unfortunately, the target speed for the Power6 was 5+Ghz. Also, the Power6 is like a Pentium-4; deeply pipelined. While the official number hasn't been out, its been stated that its longer than that of the XBox 360's Xenon (which has 21):

Xenon pipelines: http://arstechnica.com/articles/paedia/cpu/xbox360-2.ars/3
Power6 pipeline: http://arstechnica.com/news.ars/post/20070212-8823.html

Part of IBM's help would be in a their new versions of copper interconnects on the CPU die, as well as High-K, and Strained Silicon. All of these contributed to lowering the heat generated by the CPU, allowing IBM to increase the Ghz.

I find it highly unlikely you worked on the low-level hardware for Broadcom. Making material changes takes at least 3-4 months and then another 9-12 months of debugging and fixing whatever problems there are. Even with a "simple" material revision, we're looking at about a year of time.

You don't have to look to far back, to remember the first AMD 'Hammer' CPU that ran at 800Mhz (similar to the Barcelona we see now), while Intel had CPU's running in the 2+Ghz range. When AMD finally released the 'Hammer' based CPU's they were running just a little slower in clockspeed than Intel's, but had a huge performance increase over them

This is most likely as to what you're referring to: http://www.anandtech.com/printarticle.aspx?i=1591

Those were, unfortunately, A0 steppings fresh from the FAB. Barcelona, according to Anandtech is already several steppings beyond that. Moreover even at that low frequency (800Mhz), the IPC has *not* changed significantly from launch.

Is there room for variation? Certainly. The claimed +40% general IPC improvement over Kentsfield? Highly doubtful. If we use Cinebench as a baseline, than its even more doubtful because Cinebench performed very favorably on the A64's, even when compared with the Core2.

 

[Powerhouse2k]

Love your posts Scali2, you almost sound like a lawyer trying to twist the truth.

So where to begin?

1. The Problems I was refering to didn't mean 'Faulty Batches', but could be any number of issues that required the Partners to 'Heavily Modify' their BIOS' to work with the CPU.

2. The CPU's didn't end up in the hands of a big review site (in this case Anandtech). Again, if you read the article over there, basically they got there information through 'Word of Mouth' from some AMD Partner(s). Anandtech didn't actually have the CPU and put it through any kind of testing.

"None of the partners we talked to are really impressed with Barcelona yet, but then again none of the true potential of the chip and chipsets have been realized due to CPU stepping issues, BIOS maturity, and generally low clock speeds."

3. There are NDA's, which is why Anandtech couldn't state who the AMD 'Partners' were, and why the CPU pictured had the serial number covered up.

4. As for Production, it's no secret that AMD is having problems there, with only 2 fabs making it's chips. Its not like Intel, where they can use 1 whole fab for their next generation chips to refine the process, while all the rest (what are they at, 8 or 10 fabs now?) keep cranking out existing CPU's. So for AMD, they have to stop producing some chips, to produce some Barcelona 'test' CPU's, then resume production on their existing CPU's. Again, this is a very big problem for AMD, no doubt about it.

5. Production Quality means, in this case, that the CPU functions, but is still in need of some tweaks. What those tweaks are, I couldn't tell you, but obviously they have made some tweaks, as the article stats that...

"The chips we saw at Computex ran at either 1.4GHz or 1.6GHz, the latter being the latest B01 stepping. AMD's Tier-1 server partners are starting to see 1.8GHz chips, but that's the fastest we've heard of at this point. "

So obviously there was some change made from the B01 stepping to the one the Tier-1 partners got.

6. My 'Sources' was a joke, as I see way to many people use that line in these and other Forums, to make them look like they are in the know.

7. Given AMD's supply problem that I listed above (no. 4 above), which Tier are your sources? Since AMD will give all it's initial Barcelona Supply to their Tier-1 partners first (Dell, etc, I'm assuming), obiously Tier-2 and lower will have later shipments.

 

[Powerhouse2k]

I find it highly unlikely you worked on the low-level hardware for Broadcom. Making material changes takes at least 3-4 months and then another 9-12 months of debugging and fixing whatever problems there are. Even with a "simple" material revision, we're looking at about a year of time.

Those were, unfortunately, A0 steppings fresh from the FAB. Barcelona, according to Anandtech is already several steppings beyond that. Moreover even at that low frequency (800Mhz), the IPC has *not* changed significantly from launch.

Believe what you want about where I worked, I was there, were you? Broadcom is a Fabless company, they sent all their designs to UMC and TSMC for production.

So even if your timetable is correct (don't really know, again I don't work for IBM or AMD), IBM release the information about their new CPU in January (with working samples). AMD being a partner with IBM could have signed a new agreement with IBM for this new tech. So 3-4 months puts that around April-May to make the material changes, with the rest of the year to debug and fix 'problems'. Could actually explain the issues with AMD's current steppings.

As for the original Hammer A0 steppings fresh from the FAB, well, who is to say how many steppings (and the time frame they were tested) were in between A0 and the AMD partners that used B01.

 

[RadiationMan]

So even if your timetable is correct (don't really know, again I don't work for IBM or AMD), IBM release the information about their new CPU in January (with working samples).

I cant find links to working Power6 samples, my internet is messed up right now. Can you provide some? I googled some searches but my browser keeps closing when I click on a link.

 

[Scali2]

Love your posts Scali2, you almost sound like a lawyer trying to twist the truth.

Whatever, I'm not going to play your numbers game, because it would only lead to me repeating myself, which is useless.
And everything you mentioned is beside the point I made earlier: Where are the good chips?
And as dexvx also stresses... how are they ever going to fix that IPC?
BIOS or bad batches etc simply aren't plausible explanations. Not when you're about 5 weeks away from your release date. You can't fix such fundamental problems in such a short amount of time.

Now as you may or may not know, I dabble in medical data visualization, so you can bet that I have the inside track with some of those tier-1 partners. We love big servers with lots of cores.

 

[Powerhouse2k]

Unfortunately, the target speed for the Power6 was 5+Ghz. Also, the Power6 is like a Pentium-4; deeply pipelined. While the official number hasn't been out, its been stated that its longer than that of the XBox 360's Xenon (which has 21)

To answer your question, no the Power6 isn't all that deeply pipelined (I believe the current A64's have either 12 or 14):

http://arstechnica.com/news.ars/post/20070521-ibms-power6-flies-the-coop-at-4-7ghz.html

"POWER6 would debut north of the 5GHz mark, so Big Blue fell a bit short on that score. But such a shortcoming is pretty forgivable when you consider the fact that POWER6 has the same number of pipeline stages as its predecessor (15 integer stages), and it doesn't draw much more power."

 

[Scali2]

"POWER6 would debut north of the 5GHz mark, so Big Blue fell a bit short on that score. But such a shortcoming is pretty forgivable when you consider the fact that POWER6 has the same number of pipeline stages as its predecessor (15 integer stages), and it doesn't draw much more power."

Read the small print though.
They removed OOO, that cuts out quite a few pipeline stages.
So despite the less complex design, it still has 15 stages. Which is quite a lot.
The original Pentium was the last non-OOO processor, and had 5 pipeline stages (I believe the MMX-version actually had 6). Its OOO-enabled counterpart, the Pentium Pro, had 14(!) pipeline stages (or 12+2, depending on how you count). Both were 200 MHz parts at the time, with comparable IPC.

The original POWER processors, without OOO, only had 4 stages if I'm not mistaken.

 

[Powerhouse2k]

What small print? Dexvx posted that the Power6 had more that 21 pipelines, and I posted that in fact it only has 15. I also just read that in fact the A64's have 12 pipelines (the Pentium 4's have 20, Pentium D had up to 31, and the Core2D have 14).

http://www.extremetech.com/article2/0,1697,1988794,00.asp

As for your ramblings about OOO and older CPU's, that really has nothing to do with what Dexvx posted and I answered. Nor does it have anything to do with this Discussion on AMD's Barcelona.


As for some good Barcelona info regarding IPC:

http://www.beyond3d.com/content/news/224

http://www.realworldtech.com/page.cfm?ArticleID=RWT051607033728

 

[osalcido]

Yea, it's all coming out now, just like I've always said.
People had so much faith in AMD that they just couldn't believe I was right.
I suppose it's time for an "I told you so".

Hate to break it to you but 90% of knowledgeable members already believed this would happen. Your constant bickering with duby wasnt enlightening to anybody.....

 

[Digital Viper-X-]

Yes, but I don't consider Anandtech very knowledgeable on the topic of lowlevel BIOS/CPU architecture. I'm quite sure I have a better insight on the situation than they do.

I could say the same about you. You are the one calling me names here, when you have absolutely no proof that I could be wrong.

Yes, and it's highly unlikely that any of those features would be crippling performance, because they would simply be turned off if that was the case. They're probably turned off anyway, when they're benchmarking the maximum performance.

I'm sorry I have no proof your wrong, but you have no proof that you are right either, no, some words from anand on the subject without extensive testing don't count as proof to a good majority of people.

I consider thermal and power managment to be a huge issue for barc, considering its new features, what if they couldn't control the power states properly and cpus were only using 2 cores instead of 4? or if the cpu was being throttled, etc

 

[BigDH01]

What small print? Dexvx posted that the Power6 had more that 21 pipelines, and I posted that in fact it only has 15. I also just read that in fact the A64's have 12 pipelines (the Pentium 4's have 20, Pentium D had up to 31, and the Core2D have 14).

http://www.extremetech.com/article2/0,1697,1988794,00.asp

As for your ramblings about OOO and older CPU's, that really has nothing to do with what Dexvx posted and I answered. Nor does it have anything to do with this Discussion on AMD's Barcelona.


As for some good Barcelona info regarding IPC:

http://www.beyond3d.com/content/news/224

http://www.realworldtech.com/page.cfm?ArticleID=RWT051607033728

Depends if you're speaking integer or floating point pipeline stages. AFAIK, the Athlon 64 has more stages in its FPU than the Power 6 or the Core 2. Not sure what bearing it has on this convo but I thought I'd throw it out there.

I believe the Power 6 has 15 stages in its integer and fpu pipelines and the Core 2 has 14 each. I can't recall exactly how many stages the Athlon 64 has in its FPU but I believe it's close to 20.

Looked it up, it's 17 for Athlon 64 FPU.

 

[LstBrunnenG]

On paper Barcelona looked like a competitive design, and I don't think anyone here is hoping for it to be an 2900XT-sized flop. I'm not saying that Barcelona is a flop yet, but news like this is not good news.