What do you expect from Barcelona in reality?

[Game Boy]

I've been reading the news about Barecelona (K10, K8L, Phenom, whatever you want), and have seen lots of news about a) It only getting up to 2Ghz initially, b) It being delayed considerably, but c) Having anywhere from a 1x to a 1.5x clock-for-clock over Intel

Best case scenario:3GHz Phenom FX readily availible on the desktop Q4, 1.5x C-for-C, demolishing Intel with a $500 effectively 4.5GHz vs Intel's 3.33GHz.

Worst case scenario: 2.2Ghz Phenom X4 in short supply even in December, 1x C-for-C with Intel.

My prediction: 2.5Ghz Phenom X4 readily availible late Q4, 1.2x C-for-C with Intel
leading to a 3.0Ghz in late Q1 2008 which matches a future 3.66Ghz Penryn.

Do you think this is reasonable? Have you seen any evidence toi support best- or worst-case scenarios?

 

[TuxDave]

Originally posted by: Game Boy
I've been reading the news about Barecelona (K10, K8L, Phenom, whatever you want), and have seen lots of news about a) It only getting up to 2Ghz initially, b) It being delayed considerably, but c) Having anywhere from a 1x to a 1.5x clock-for-clock over Intel

Best case scenario:3GHz Phenom FX readily availible on the desktop Q4, 1.5x C-for-C, demolishing Intel with a $500 effectively 4.5GHz vs Intel's 3.33GHz.

Worst case scenario: 2.2Ghz Phenom X4 in short supply even in December, 1x C-for-C with Intel.

My prediction: 2.5Ghz Phenom X4 readily availible late Q4, 1.2x C-for-C with Intel
leading to a 3.0Ghz in late Q1 2008 which matches a future 3.66Ghz Penryn.

Do you think this is reasonable? Have you seen any evidence toi support best- or worst-case scenarios?

If we're talking worst case scenario... what makes you so sure that it'll be 1x clock for clock?

 

[MarcVenice]

Well, all the time they've spend engineering the new architecture and what not, should at least yield a 20% performance increase over the current gen athlon 64's/K8 cpu's, no? Putting them right next to intels c2d's clock for clock. If not, they should sell all the stock they've got and find themselfs a new job asap

 

[Game Boy]

Originally posted by: TuxDave
If we're talking worst case scenario... what makes you so sure that it'll be 1x clock for clock?

It's called worst-case for a reason.

If you disagree, what's your prediction?

 

[TuxDave]

Originally posted by: Game Boy

Originally posted by: TuxDave
If we're talking worst case scenario... what makes you so sure that it'll be 1x clock for clock?

It's called worst-case for a reason.

If you disagree, what's your prediction?

So with almost no benchmark information on a broad level, you have high confidence that Barcelona will run no slower than 1x to 1x to what Intel has at that time? Worst case for AMD should also be assuming the best case for Intel right? I was thinking if we wanted to be pessimistic you should just dump in a 0.8x to 1x but then again that's just me pulling numbers out of the air.

 

[TuxDave]

Originally posted by: MarcVenice
Well, all the time they've spend engineering the new architecture and what not, should at least yield a 20% performance increase over the current gen athlon 64's/K8 cpu's, no? Putting them right next to intels c2d's clock for clock. If not, they should sell all the stock they've got and find themselfs a new job asap

My concern is that the 'worst case scenario' has the assumption that Intel is doing nothing which doesn't sound right to me. That would be the 'best case' for AMD.

 

[RussianSensation]

Originally posted by: TuxDave

If we're talking worst case scenario... what makes you so sure that it'll be 1x clock for clock?

Dual stage 128-bit instruction pipeline vs. single stage for C2D, onboard memory controller and historically shorter and more efficient pipeline of Athlon processors virtually provides a 99% guarantee that Barcelona will be faster per clock cycle than C2D. The only question is how much...

Considering that 2.0ghz A64 3000+ was as fast and often outperformed P4 3.0ghz that's a 50% advantage in efficiency. C2D is roughly 90-100% faster than P4 per clock cycle (more in games though and less in video encoding). So if AMD were to double A64's efficiency which is easily realistic given A64's dual state 64-bit instruction pipeline, lack of L3 cache, and less efficient memory controller, it'd be roughly 50% more efficient than C2D per clock cycle. However, C2D's advantages such as robust branch prediction will probably mean Barcelona will be roughly 30-35% as efficient. That's my guess. This would allow a 2.4ghz Barcelona to compete with a 3.2ghz C2D. Problem is with 45nm Penryn, C2D should scale to 4.0ghz. However, as long as AMD is able to deliver best performance/power consumption/price ratio for low-mid-high end, they don't necessarily have to have the absolute fastest processor at the top to regain market share.

 

[Keysplayr]

Paper specs mean next to nothing RS. AMD should try to do what Intel did with C2D and allow pre-release benchmarks to at least give us an "idea" of the ballpark barcelona will be in. My thoughts are, if barcelona was so incredible, AMD would be shouting and gloating from the rooftops to staunch as much of Intel sales as possible until barcelona actually releases. The lack of any such "gloating" tells me that barcelona won't be much of a threat to even a C2D. Let alone Penryn. That is what I'm getting from all this. WE NEED TO SEE BENCHMARKS NOW!! hehe. Sorry for yelling.

 

[TuxDave]

Originally posted by: RussianSensation

Originally posted by: TuxDave

If we're talking worst case scenario... what makes you so sure that it'll be 1x clock for clock?

Dual stage 128-bit instruction pipeline vs. single stage for C2D, onboard memory controller and historically shorter and more efficient pipeline of Athlon processors virtually provides a 99% guarantee that Barcelona will be faster per clock cycle than C2D. The only question is how much...

Considering that 2.0ghz A64 3000+ was as fast and often outperformed P4 3.0ghz that's a 50% advantage in efficiency. C2D is roughly 90-100% faster than P4 per clock cycle (more in games though and less in video encoding). So if AMD were to double A64's efficiency which is easily realistic given A64's dual state 64-bit instruction pipeline, lack of L3 cache, and less efficient memory controller, it'd be roughly 50% more efficient than C2D per clock cycle. However, C2D's advantages such as robust branch prediction will probably mean Barcelona will be roughly 30-35% as efficient. That's my guess. This would allow a 2.4ghz Barcelona to compete with a 3.2ghz C2D. Problem is with 45nm Penryn, C2D should scale to 4.0ghz. However, as long as AMD is able to deliver best performance/power consumption/price ratio for low-mid-high end, they don't necessarily have to have the absolute fastest processor at the top to regain market share.

I'll have to quote you on that.... but due to lack of information I can't make a guess on my own.

 

[Toadster]

I'm waiting to see if "true quad core" really improves the applications that everyone whines about in Intel's non-true quad core

 

[cmdrdredd]

Originally posted by: keysplayr2003
Paper specs mean next to nothing RS. AMD should try to do what Intel did with C2D and allow pre-release benchmarks to at least give us an "idea" of the ballpark barcelona will be in. My thoughts are, if barcelona was so incredible, AMD would be shouting and gloating from the rooftops to staunch as much of Intel sales as possible until barcelona actually releases. The lack of any such "gloating" tells me that barcelona won't be much of a threat to even a C2D. Let alone Penryn. That is what I'm getting from all this. WE NEED TO SEE BENCHMARKS NOW!! hehe. Sorry for yelling.

I agree with this. See AMD is holding back for a reason. They don't want to look foolish showing a 3Ghz CPU that is slower than a 3Ghz CPU from another manufacturer. Maybe they are working on increasing performance, maybe they are not. Either way if it was indeed better they would want people to know.

I predict that their next line will compete evently with Intel's current C2D line as far as performance is concerned. The only issue is marketing it correctly.

 

[cmdrdredd]

Originally posted by: Toadster
I'm waiting to see if "true quad core" really improves the applications that everyone whines about in Intel's non-true quad core

it won't...people are using the term "true" falsely similar to Sony claiming that 1080p TV was "true HD". Really it's either HD or it's not...it's 4 cores or it's not. The arcitecture or method of doing it can be different.

 

[SlowSpyder]

I have a feeling that AMD's 'true' quad core design will allow it to scale better in software that can take advantage of the extra cores. But, who knows how it'll compete in single threaded stuff. I have no data to back this up, just a hunch at best.

Does anyone remember how AMD acted before the original K8 release? I think they were pretty quiet on the benchmarks, weren't they? I remember Intel releasing the EE P4's and AMD still didn't do/say anything about the K8's until the NDA was up. Hopefully for their sake the K10 situation is not too different then the K8 situation.

 

[cmdrdredd]

Originally posted by: SlowSpyder
I have a feeling that AMD's 'true' quad core design will allow it to scale better in software that can take advantage of the extra cores. But, who knows how it'll compete in single threaded stuff. I have no data to back this up, just a hunch at best.

Does anyone remember how AMD acted before the original K8 release? I think they were pretty quiet on the benchmarks, weren't they? I remember Intel releasing the EE P4's and AMD still didn't do/say anything about the K8's until the NDA was up. Hopefully for their sake the K10 situation is not too different then the K8 situation.

Well, how can 4 cores on an Intel platform which performs better at single threaded apps(just for arguments sake here) be slower than AMD's quad core on MP apps? You mean that it uses more of each core? would that not require rewriting the application?

I'd just like you to clarify what you ment.

 

[Deliximus]

I think depending on program, we're looking at 15-25% C-for-C over C2D with a Phenom x4 launch at 2.4-2.6ghz.
Dual cores Phenoms @ 2.8-3.0ghz

perhaps 3ghz phenoms x4 by Q2.

 

[Toadster]

you'd hope that 5 years of development would equate to an increase...

 

[SlowSpyder]

Originally posted by: cmdrdredd

Originally posted by: SlowSpyder
I have a feeling that AMD's 'true' quad core design will allow it to scale better in software that can take advantage of the extra cores. But, who knows how it'll compete in single threaded stuff. I have no data to back this up, just a hunch at best.

Does anyone remember how AMD acted before the original K8 release? I think they were pretty quiet on the benchmarks, weren't they? I remember Intel releasing the EE P4's and AMD still didn't do/say anything about the K8's until the NDA was up. Hopefully for their sake the K10 situation is not too different then the K8 situation.

Well, how can 4 cores on an Intel platform which performs better at single threaded apps(just for arguments sake here) be slower than AMD's quad core on MP apps? You mean that it uses more of each core? would that not require rewriting the application?

I'd just like you to clarify what you ment.

Even if Intel does happen to be faster in single threaded apps, it is possible for things to scale differently on the two different architectures. From what I understand Intel quads do have four cores (obviously) but the two sets have to communicate over the front side bus (can anyone claify if that's right?). The AMD quad will have all four cores communicating via the Hypertansport link and/or the L3 cache. So, it is possible for things to scale differently regardless of single threaded performance.

 

[zephyrprime]

Originally posted by: RussianSensation
Dual stage 128-bit instruction pipeline vs. single stage for C2D,

What is that?

 

[gtsing]

any know the latest news on a rough release date for the first batch of these procs?

 

[nyker96]

I believe that the Barcelona core about same as C2D but will be more energy efficient than C2Quad. However, there maybe a few spots where it shines like workstation workload but not in desktop. This is because Barcelona is designed to replace Opty and is aimed at enterprise first and desktop second. this is a prediction only nothing concrete of course.

 

[cmdrdredd]

Originally posted by: gtsing
any know the latest news on a rough release date for the first batch of these procs?

before the end of the century. :laugh:

just a joke

 

[theprodigalrebel]

I expect a significant increase over the K8 architecture - I mean, it has been five years. Five years is too long to release a CPU that doesn't demolish the previous one. Like MarcVenice said, they'd have to find a new business if they don't deliver on Barcelona.

Honestly, I expect the 2GHz Barcelona to compete with a 2.4GHz Core 2. I am guessing that the reason they haven't released any benches is because it might stall sales of their current CPUs. Intel could have afforded to take a hit on their P4 sales after they released Conroe benches - AMD needs all the sales they can get.

Or maybe that's a very simple way of looking at it.

 

[bryanW1995]

I have a feeling that a lot of amd employees are about to start looking for alternative employment. I they had ANYTHING they would have announced it a long time ago and allowed benchmarks b/c intel is killing them. You know that they would have rained on intel's july 22 parade if they could have with something like a 25-30 % clock for clock advantage. Instead, the best they could muster was a little price cut of their already cheaper than dirt processors.

 

[apoppin]

Originally posted by: bryanW1995
I have a feeling that a lot of amd employees are about to start looking for alternative employment. I they had ANYTHING they would have announced it a long time ago and allowed benchmarks b/c intel is killing them. You know that they would have rained on intel's july 22 parade if they could have with something like a 25-30 % clock for clock advantage. Instead, the best they could muster was a little price cut of their already cheaper than dirt processors.

i don't know. AMD was tight-lipped about HD2900xt but we were able to get a performance review well before [finally] launching and we knew it would be no GTX killer. Still, it turned out pretty well to compete with the GTS series.

Unfortunately, AMD is not allowed this "luxury" of 2nd Best with Barcelona. They desperately *need* a good showing against intel's best. i am not sure what to expect but i am hoping for a good showing or i think AMD won't have a 2nd chance.

 

[bryanW1995]

yeah, don't get me wrong, I HOPE that barcelona/phenom is stronger than garlic because I feel that competition is great for all of us and I think that it's 50/50 or better that there will be no amd in 12 mos without some seriously strong product releases. I just think that it makes absolutely no sense for them to withold benchmarks right now if they could possibly get people like us to ignore intel's price cuts and hold out for amd's new offerings. Every day that they don't release info is one more day that droves of people ditch their am2 and s939 mobo's and march over to the intel camp.