Sign in with
Sign up | Sign in
Your question

Barcelona Doesn't Deliver - Benchmarks Deemed Non-Compliant

Last response: in CPUs
Share
November 9, 2007 11:41:44 PM

http://roborat64.blogspot.com/2007/11/barcelona-benchma...

Vendors are now seeing their benchmark results deemed non-compliant because Barcelona is apparently no where in sight; and this isn't even the 2.0ghz part, even the 1.9ghz is apparently MIA.

Quote:
...a Tier1 vendor like IBM can't seem to get hold of AMD's latest quad-core. If an important customer can't even ship a lower bin 1.9Ghz Barcelona months after product launch one has to wonder how many has AMD really managed get out of the door...


Expect more 300+M losses if AMD doesn't ship processors.
November 10, 2007 12:03:02 AM

itotallybelieveyou said:
nice find


I really like Roborat's Blog. He really tackles the whole "AMD is failing miserably" topic, but he does so with fact. He presents his arguments very well, and when he accidentally posts inaccurate information he corrects it and admits the mistake.

I highly recommend his blog!
November 10, 2007 12:07:13 AM

I visit his site frequently. I don't value his posts that much, but you really should take a look at the bloggers at his sites. They consist of some of the brightest minds I've ever known.
November 10, 2007 12:10:43 AM

I found one!!!
a b à CPUs
November 10, 2007 12:46:05 AM

This blog reads like an Intel version of Sharikou.
a b à CPUs
November 10, 2007 12:49:50 AM

Ok, AMD really shot themselves in the foot. I wonder if Fusion will "hurt" as bad.


^^ Hector
November 10, 2007 1:18:54 AM

TechnologyCoordinator said:
http://roborat64.blogspot.com/2007/11/barcelona-benchma...

Vendors are now seeing their benchmark results deemed non-compliant because of Barcelona is apparently no where in sight; and this isn't even the 2.0ghz part, even the 1.9ghz is apparently MIA.

Quote:
...a Tier1 vendor like IBM can't seem to get hold of AMD's latest quad-core. If an important customer can't even ship a lower bin 1.9Ghz Barcelona months after product launch one has to wonder how many has AMD really managed get out of the door...


Expect more 300+M losses if AMD doesn't ship processors.



Perhaps they were taken for the Texas SC that APPRO won the contract for. Dirk was quoted saying they shipped 10s of 1000s and just that one will take 12000. There were also a few others including Sun. They even use Budapest so those are already sold.

I think the Phenom availability will be better because they don't need the higher-quality and they have the extra two months. Barcelona has been available in retail and at some channel companies. I'm sure the plan wasn't to have to remove a SPEC entry but perhaps the B2F stepping will get to Barcelona quickly and invalidate the previous scores. We still haven't seen benches with it yet as expreview uses the ES to OC. I'm not a real OCer and I'd liek to see the new rev at actual speed. And imagine that someone could bring action because people are paying for OCers' dreams.
Anyway, I guess that's a good score and expreview did show that higher HT link and HT3 speeds do improve perf by around 12-15%. With just a 9500 I can up my active processes by 70%. That'll be worth $300.

I think demand must be high no matter how many they have shipped. If they can get back just 1% share, it will bode well for 30% by Q308. ANd with the economies of wafer sizes vs. transistor size they can raise gross margins and maybe actually stay in business for another day or two.
a b à CPUs
November 10, 2007 1:24:43 AM

BaronMatrix said:
I think the Phenom availability will be better because they don't need the higher-quality and they have the extra two months.

Well it's logical, and it's necessary so they better deliver.

BaronMatrix said:
I'm not a real OCer and I'd liek to see the new rev at actual speed.

Same, but I doubt a new rev is going to make that much difference in performance, unless AMD has something up their sleeve that they aren't letting on.

BaronMatrix said:
ANd with the economies of wafer sizes vs. transistor size they can raise gross margins and maybe actually stay in business for another day or two.

Well if Apple buys Intel inside that two day margin, I think many PC enthusiasts will boycott intel and give AMD a chance to take back the crown :lol: 
November 10, 2007 1:43:33 AM

I heard availablility was bad because of that company who was building some supercomputer. Who was that again? Anyway, they apparently bought like over 10,000 barcy's for it. That is what I have heard and read. I will try to do some research and post the links on here if I can.
a b à CPUs
November 10, 2007 2:33:55 AM

so thats where they all went
November 10, 2007 4:38:32 AM

monsterrocks said:
I heard availablility was bad because of that company who was building some supercomputer. Who was that again? Anyway, they apparently bought like over 10,000 barcy's for it. That is what I have heard and read. I will try to do some research and post the links on here if I can.

I hope you're right, I just bet $XXX with my friend that AMD's Phenom would be "decent".
November 10, 2007 6:29:02 AM

monsterrocks said:
I heard availablility was bad because of that company who was building some supercomputer. Who was that again? Anyway, they apparently bought like over 10,000 barcy's for it. That is what I have heard and read. I will try to do some research and post the links on here if I can.


If they're running 2500 wafers a week with 100 good die on each wafer (that's being generous) then they shouldn't be having this much trouble keeping up. That's a million parts a month if they're keeping up a good clip in the fab.

The green team is clearly having issues.
November 10, 2007 6:56:31 AM

BaronMatrix said:
There were also a few others including Sun. They even use Budapest so those are already sold.

Care to provide a link to back this up?

Also,
Quote:
I guess that's a good score and expreview did show that higher HT link and HT3 speeds do improve perf by around 12-15%. With just a 9500 I can up my active processes by 70%.


Want to provide a link before stuffing words into Expreview's mouth?
a b à CPUs
November 10, 2007 7:11:13 AM

Heh Intel stayed away from a Monolithic design for a reason, perhaps AMD would have been better off making a dual dual core A64 to get Into the Quad Core Industry and waited for 45nm and a Monolithic Quad Core, cause Intel is already on there second generation of Quad Cores with 8 cores well within reach @ 45nm, and even if AMD does catch Intel they have Nehalem samples already surfacing.

AMD's troubles started with 65nm, makes you wonder if they solved it with the Phenom - doesnt seem like it.

Perhaps we all see AMD in the wrong light here - The K8 was actually the first successful product and now they have returned to there normal spot, and AMD is a tiny company compared to Intel - other then the K8 they never really had success, perhaps AMD's real spot is the budget and bang for the buck segment like the K6 and K7 days.
November 10, 2007 7:12:55 AM

Coolest paper launch evar
November 10, 2007 9:19:51 AM

Impressive, you know whats more pathetic than failed launch? Its those who dedicate their time exlusively to bash some company :sarcastic:  They kick dead horse and then pick the microscope and describe in detail how every molecule is rotting :p 
November 10, 2007 11:17:39 AM

Barcy most likely has lower yields but with a year and a half of old systems plus new system I would see this as more of a demand issue. I would suspect barcy due to old and new systems is the most demanded CPU of all time.

Imagine the demand on conroe if all the mobos sold for a year and half prier would have wanted upgrades. Many of those who could have upgraded probably didn't because of the technical knowledge need to flash the bios. The higher cost of the conroe than barcy at the time would have delayed many from upgrading. Just imagine the demand if the X6800 would have been $380 and the E6600 for $239 on launch. The price alone would have caused an extra month of supply shortages.

While this does not look good for AMD its a lot better than 8 months ago when low demand was the issue.

Note: I didnt use E6300 and E6400 because barcy has no differences other than clock speed.
November 10, 2007 2:25:51 PM

yomamafor1 said:
Care to provide a link to back this up?

Also,
Quote:
I guess that's a good score and expreview did show that higher HT link and HT3 speeds do improve perf by around 12-15%. With just a 9500 I can up my active processes by 70%.


Want to provide a link before stuffing words into Expreview's mouth?



What is it with you? Right on this very site there were several posts about how Sun would lose money because Budapest was delayed. Someone even posted - maybe you - a set of scores with HT link increased and CrySis went from 64 to 71 or so on expreview.

As far as adding processes that's common sense. Two extra cores means many extra processes.



November 10, 2007 3:04:32 PM

Baron, it's not the HT bus that increases the performance, it's the RAM speed. If you bothered to check the Expreview links more closely, you will see they compared a 1.3GHz HT bus to a 2.1GHz HT bus - there is no difference in performance.
November 10, 2007 3:43:55 PM

epsilon84 said:
Baron, it's not the HT bus that increases the performance, it's the RAM speed. If you bothered to check the Expreview links more closely, you will see they compared a 1.3GHz HT bus to a 2.1GHz HT bus - there is no difference in performance.



Yes, hence the deduction that Phenom will be more sensitive to RAM speed than Intel offerings.
November 10, 2007 4:28:14 PM

BaronMatrix said:
What is it with you?

I guess I'm just too sensitive to FUDs...

Quote:
Right on this very site there were several posts about how Sun would lose money because Budapest was delayed.

But has it even officially launched? I scoured the internet, and found 0 news that suggested Budapest had launched. Yes, it was delayed, but it has yet to be launched.


Someone even posted - maybe you - a set of scores with HT link increased and CrySis went from 64 to 71 or so on expreview.
said:

Someone even posted - maybe you - a set of scores with HT link increased and CrySis went from 64 to 71 or so on expreview.

Thank you Epsilon... this guy really needs to take a closer look at the numbers.

Baron, you've been saying HT3 would help. How would it helped?
November 10, 2007 10:43:48 PM

yomamafor1 said:
I guess I'm just too sensitive to FUDs...

Quote:
Right on this very site there were several posts about how Sun would lose money because Budapest was delayed.

But has it even officially launched? I scoured the internet, and found 0 news that suggested Budapest had launched. Yes, it was delayed, but it has yet to be launched.


Thank you Epsilon... this guy really needs to take a closer look at the numbers.

Baron, you've been saying HT3 would help. How would it helped?




Stop saying FUD. FUD stands for Fear Uncertanity, and Doubt. You can't clock AM2\HT1.1 as high. The Inq got a 790FX board and got a BE 5000+ to 3.5GHz. That's much higher than nForce or 580X. HT3 has been reported clocking to 500MHz(FudZilla). Try that with nForce 590. That means that 3GHz only needs a 6X rather than 15X and faster FSB has been proven to be better than higher multipliers.
November 10, 2007 10:49:40 PM

BaronMatrix said:
Stop saying FUD. FUD stands for Fear Uncertanity, and Doubt.

Yes, thank you very much for your first grade education.


You can't clock AM2\HT1.1 as high. The Inq got a 790FX board and got a BE 5000+ to 3.5GHz. That's much higher than nForce or 580X. HT3 has been reported clocking to 500MHz(FudZilla). Try that with nForce 590. That means that 3GHz only needs a 6X rather than 15X and faster FSB has been proven to be better than higher multipliers. said:

You can't clock AM2\HT1.1 as high. The Inq got a 790FX board and got a BE 5000+ to 3.5GHz. That's much higher than nForce or 580X. HT3 has been reported clocking to 500MHz(FudZilla). Try that with nForce 590. That means that 3GHz only needs a 6X rather than 15X and faster FSB has been proven to be better than higher multipliers.


No one doubted 790X's FSB overclockability. No one doubted HT3 can reach record speed. But how do those two things help Phenom in gaining higher performance? As Epsilon said, Phenom is more sensitive to RAM speed than HT3 speed.
November 11, 2007 12:10:09 AM

No one doubted 790X's FSB overclockability. No one doubted HT3 can reach record speed. But how do those two things help Phenom in gaining higher performance? As Epsilon said, Phenom is more sensitive to RAM speed than HT3 speed.

No AMD is more susceptible to latency not RAM speed because of the IMC. And the RAM was only at 485MHz, not 533MHz(1066). The Inq has screenies that show 790FX does fix the RAM divider issue as the OC utility they used on 5000+ showed the RAM running at rated speed.

Anyway, I guess the whole point is I only buy AMD for whatever reasons I want. They are competitive. We'll, I guess, see full-blown benches soon as AMD has been showing off a Dell XPS with Phenom.

I would say that it will be really fast.
November 11, 2007 1:30:18 AM

BaronMatrix said:


No AMD is more susceptible to latency not RAM speed because of the IMC.

Ok. If this is the case, then how does having a HT3, which boasts larger bandwidth, help with the latency issue?

Quote:

And the RAM was only at 485MHz, not 533MHz(1066). The Inq has screenies that show 790FX does fix the RAM divider issue as the OC utility they used on 5000+ showed the RAM running at rated speed.

:lol:  :lol:  :lol:  :lol:  :lol: 

I think you're royally confused. 3425Mhz with 380Mhz RAM? How is that fixed? Fixed would be RAM speed at 400Mhz, but why is it 20Mhz short? Can you please explain to me Baron? :pt1cable:  :pt1cable: 



Anyway, I guess the whole point is I only buy AMD for whatever reasons I want. They are competitive. We'll, I guess, see full-blown benches soon as AMD has been showing off a Dell XPS with Phenom.

[b said:
I would say that it will be really fast.]Anyway, I guess the whole point is I only buy AMD for whatever reasons I want. They are competitive. We'll, I guess, see full-blown benches soon as AMD has been showing off a Dell XPS with Phenom.

I would say that it will be really fast.
[/b]

Thank you for your opinion
. My take is that currently only AMD offers quad VGA capability. Therefore, you can put four 3870 X2 on it, and have 8 GPUs working for you. Since Nvidia only has three way SLI available, Dell would be dumb if they stick with Nvidia, although offer better performance per GPU, but forgo the possibility of having double amount of GPUs.
November 11, 2007 3:15:08 AM

So funny, even a caveman will laugh
November 11, 2007 11:43:12 AM

Quote:
No AMD is more susceptible to latency not RAM speed because of the IMC. And the RAM was only at 485MHz, not 533MHz(1066). The Inq has screenies that show 790FX does fix the RAM divider issue as the OC utility they used on 5000+ showed the RAM running at rated speed.

First sentence says AMD needs low latency, not high speed. Second and third sentences complain about the speed of RAM used in benching AMD. Non sequitur?

Mind you, it already sounds strange to complain about 485MHz RAM being a bottleneck when Core 2 is frequently benched with cheap 266-400MHz RAM. CL4 RAM at >500MHz is practically a top-of-the-line bin.

It is also strange to differentiate HT1.1 from HT3 boards for Phenom. RAM doesn't use HT to communicate with the IMC. I understand that the different HT versions may happen to use different base clocks and that this could lead to somewhat different RAM speeds/multipliers, but we already covered that above; RAM speed has not been the main factor in the IMC design. An integer divider and low latency are more important, and what version HT is used has no bearing.
November 11, 2007 12:53:40 PM

BaronMatrix said:
No AMD is more susceptible to latency not RAM speed because of the IMC. And the RAM was only at 485MHz, not 533MHz(1066). The Inq has screenies that show 790FX does fix the RAM divider issue as the OC utility they used on 5000+ showed the RAM running at rated speed.


http://www.anandtech.com/memory/showdoc.aspx?i=2800&p=7

AMD looks very reliant on RAM speed to me. DDR2-800 is a good 5 - 10% faster than DDR2-667, and 10 - 20% faster than DDR2-533.

I imagine Phenom would show similar scaling, but obviously support is extended DDR2-1066.

Quote:
I would say that it will be really fast.


You said that about 4x4 too. :lol:  :whistle: 
November 13, 2007 7:25:32 AM

yomamafor1 said:
Ok. If this is the case, then how does having a HT3, which boasts larger bandwidth, help with the latency issue?


You do know that higher RAM speed will need a higher HT or FBS speed to reach its full potential right?

You do know that having a DDR2 1333MHz RAM on a 800MHz Pentium 4/D bus is not going reach the RAM's max performance or potential right?

You do know that the same theory applies to AMD's HT1, HT3 as well right?

Just in case you dont: Having a fast RAM module on HT1 = useless, therefore, HT3 will help increase performance when paired with fast memory.

Same theory in Intels term (some people only look for the word INTEL in a post): Having fast RAM modules running on 800MHz FBS = useless, P35, X38 will help increase performance when paired with fast memory.

:pt1cable:  :heink: 
November 13, 2007 2:29:12 PM

monsterrocks said:
I heard availablility was bad because of that company who was building some supercomputer. Who was that again? Anyway, they apparently bought like over 10,000 barcy's for it. That is what I have heard and read. I will try to do some research and post the links on here if I can.


Some company named AMD bought all the barcy's in order to build a supercomputer. They needed a new machine to replace Henri and turn poor performance into plausible sounding lies.

"AMD IS NOT AS GOOD AS INTEL"

"Oops, needs tweaking"
November 13, 2007 4:29:25 PM

On the plus side, we don't have any "My 2.0 GHz Barcelona runs too hot!!!" threads.
November 13, 2007 5:48:30 PM

pete4r said:
You do know that higher RAM speed will need a higher HT or FBS speed to reach its full potential right?

You do know that having a DDR2 1333MHz RAM on a 800MHz Pentium 4/D bus is not going reach the RAM's max performance or potential right?

You do know that the same theory applies to AMD's HT1, HT3 as well right?

Just in case you dont: Having a fast RAM module on HT1 = useless, therefore, HT3 will help increase performance when paired with fast memory.

Same theory in Intels term (some people only look for the word INTEL in a post): Having fast RAM modules running on 800MHz FBS = useless, P35, X38 will help increase performance when paired with fast memory.


Very well, but how much of this bandwidth is being used?

This is from Techreport:
http://techreport.com/articles.x/13176/3


Notice the maximum memory bandwidth Barcelona / Phenom can take advantage of is 11534Mb/s, or 11.5Gb/s. Bear in mind that Sandra memory test is a synthetic memory benchmark, so in real life the number should be even less.

This is the specification for HT2.0 and HT3.0
HT2.0:
* 22.4 GB/s Aggregate Bandwidth
* 11.2 GB/s (89.6 Gb/s) per Link

HT3.0:
* 41.6 GB/s Aggregate Bandwidth
* 20.8 GB/s (166.4 Gb/s) per Link

If K10 can't even take full potential of HT2.0, how is having a larger bandwidth help them? DDR2 800 dual channel has bandwidth of 12.8Gb/s, but how much of it is actually being used?
a b à CPUs
November 13, 2007 8:47:08 PM

exit2dos said:
On the plus side, we don't have any "My 2.0 GHz Barcelona runs too hot!!!" threads.

Aye ;) 
!