19 September Launch Day for FX Series Processors

[bryanW1995]

I'm interested to see how this shakes out. It could very well get to a point where AMD just can't stay in the high or mid end cpu market at all. Are we doomed to years and years of stagnation until ARM catches up to haswell-level performance?

Thanks, IDC, now I need a drink.

@virtual larry: as intel was quick to point out during AMD's "native quadcore" barcelona era, people don't pay for things like that, they pay for performance. Bobcat performs well at its price point, and AMD appears to be making some good money off of it, so it's a success thus far.

 

[Idontcare]

Are you sure about that? I don't think Bobcat is all that. It works well, but at 800Mhz, it requires 1.032v, and at 1600Mhz, requires above 1.3v. Considering competing CPUs, that seems like a lot of voltage to apply, just for 1.6Ghz. So the process it was produced on is not "all that".

Of course the process tech itself is not "all that"...this is actually my point. Architecture trumps process tech if done right.

Compare Atom on Intel's much venerated 45nm HKMG process tech to Bulldozer on TSMC's not-so-surprisingly craptastic 45nm 40nm foundry process tech.

In my opinion, FWIW, that bobcat overcame the self-inflicted technology deficit that TSMC's 40nm represents is itself a testament to the capabilities of the bobcat design team.

Applying that in a forward looking manner, cost adjusting the development of future advanced nodes, it is difficult for me to convince myself that AMD management is going to continue to ask GloFo to develop specialized process nodes like the 32nm SOI HKMG just for AMD's benefit.

I think AMD management will be actively seeking to get their 22nm design engineers to align their products with bulk-Si to ensure that AMD management can go with either TSMC or GloFo at 22nm for their high-performance MPU products. (just my opinion, not supported by any insider infos here)

 

[podspi]

I think his point about Bobcat was more technical in nature. Most consumers don't know nor care. (Edit:This is in response to bryanW)

I think the demand for high performance SOI could increase significantly. If Intel has finally decided to really go after the mobile and super-low-power markets, what makes everybody think ARM is going to have any better luck than AMD against Intel?

Yea, the ARM CPU in my phone uses next to no power, but it also pushes like ~ 10MFLOPS in Linpack. It only seems reasonable fast because there are lots of things that are hardware accelerated in integrated environments like that

Edit 2:
IDC, what makes you think future ARM-based CPUs like Denver won't utilize SOI? One would imagine that they will be helpful to high-performance ARM cores just as much as x86 cores. Atom isn't that great of a design for sure, but that shouldn't be surprising to anybody. It was a laptop-cpu designed to be a phone CPU, that blew its power budget by a LOT

If we've learned anything about Intel from C2D, Intel sometimes makes mistakes which makes them look silly. But they can get their act together. It's not like they don't have enough money to hire engineers. Their competitors aren't going to be able to depend upon braindead design forever. Or even that much longer, imho.

 

[soccerballtux]

yeah good points all. Getting tired of all these delays.
4.5ghz top line is pretty nice tech though.

 

[Red Squirrel]

I heard Bulldozer will come with a free copy of Duke Nukem Forever.

Damn, that joke does not even work anymore because DNF actually got released LOL.

Really I hope this processor will be a success, if not, I could see this being the demise of AMD. They've been working for what, 5 years on this, with no releases in between? I don't even know how they're profiting right now. They've dragged this far too long and Intel is running circles around with their 8 core processors that have been out since the Bulldozer was even announced. Ok so it's 4 with HT, but the OS sees it as 8, that's what most people care about.

I was always an AMD guy but my last few builds have been Intel because AMD just fell so much behind it's sad. I really hope they manage to pull together.

 

[Remobz]

I don't have much faith in AMD at the moment. I hope I am wrong

 

[996GT2]

I don't have much faith in AMD at the moment. I hope I am wrong

Let's just hope they don't pull a repeat of Phenom

If they do, they're pretty screwed in the desktop market

 

[wuliheron]

Its not the end of the world. Bulldozer is just another step toward trinity which will rewrite the book for desktop computing. Ideally you want 8 cpu cores, the north bridge, at least 300 streaming processors, and 1gb of edram all on the same chip for running physics, AI, and/or integrated graphics. You can change it up a little and mix a quad core cpu with more streaming processors for integrated graphics, but the idea is cram as much as you can onto the one chip excluding the system ram and long term storage. That's the future of the desktop market and having a slightly faster cpu is only a small part of the equation.

 

[Idontcare]

I think his point about Bobcat was more technical in nature. Most consumers don't know nor care. (Edit:This is in response to bryanW)

I think the demand for high performance SOI could increase significantly. If Intel has finally decided to really go after the mobile and super-low-power markets, what makes everybody think ARM is going to have any better luck than AMD against Intel?

Yea, the ARM CPU in my phone uses next to no power, but it also pushes like ~ 10MFLOPS in Linpack. It only seems reasonable fast because there are lots of things that are hardware accelerated in integrated environments like that

Edit 2:
IDC, what makes you think future ARM-based CPUs like Denver won't utilize SOI? One would imagine that they will be helpful to high-performance ARM cores just as much as x86 cores. Atom isn't that great of a design for sure, but that shouldn't be surprising to anybody. It was a laptop-cpu designed to be a phone CPU, that blew its power budget by a LOT

If we've learned anything about Intel from C2D, Intel sometimes makes mistakes which makes them look silly. But they can get their act together. It's not like they don't have enough money to hire engineers. Their competitors aren't going to be able to depend upon braindead design forever. Or even that much longer, imho.

I've been arguing for a couple years now that SOI would be killer for anything that is already implemented in bulk-Si, GPU's moving to SOI would be awesome, ARM moving to SOI would be even more awesome given the mobility apps that are involved.

Remember that EETimes article in which ARM reduced to practice and showed a 40% reduction in power-consumption combined with a 7% reduction in die-size for the same circuits at the same node by going SOI vs. Bulk?

The results show that 45-nm high-performance SOI technology can provide up to 40 percent power savings and a 7 percent circuit area reduction compared to bulk CMOS low-power technology, operating at the same speed. This same implementation also demonstrated 20 percent higher operating frequency capability over bulk while saving 30 percent in total power in specific test applications.

My tune recently has changed only because of the fact that despite these demonstrations of massive benefits to be had from SOI, no one to date has felt compelled to pursue it outside of AMD and even then only for their high-performance x86 MPU's.

Heck AMD even went bulk-Si for brazos instead of opting for SOI. If that doesn't create a foreboding prognosis of SOI's future then I don't know what else would.

 

[OBLAMA2009]

most tech ballers gave up on amd a long time ago and got nehalem or sb. unless they have something that really outperforms sb on 80% of benchmarks, bulldozer is going to be relegated to ghetto boxes gist like phenom

 

[podspi]

IDC, you could be right. Intel doesn't seem to think it needs SOI! Still, SOI could give GF's 22nm a little extra oomph, which will be nice since it will be competing against Intel's 14-nm tri-gate.

Doesn't the effectiveness of SOI decrease on lower nodes anyway?

 

[RussianSensation]

AMD has a very small window to make an impression. Even if Bulldozer equals SB, IB and SB-E will launch in the next 6 months, and are almost certain to retake the lead in the mid-range and high-end markets. AMD may be able to compete with current SB processors (or perhaps even outperform them), but Intel isn't even trying right now. It's been 8 months since 2500k/2600k launched and Intel hasn't even bothered to release any faster steppings of those chips. The minute Intel feels real pressure, they will bring out 3.7-4.2+ ghz SB/IB chips. Much faster clocked 2500k/2600k derivatives will again turn this race into which firm has the superior IPC performance - which has been Intel for the last 5 years.

What about clock speeds? Well, if AMD's top end CPU will have a TurboBoost of 4.5ghz, then it doesn't look like it'll have much more room for improvement on 32nm. Their only chance to regain leadership is if in the next 2-3 years most software starts to take advantage of 8+ threads.

 

[AnandThenMan]

...Intel isn't even trying right now. It's been 8 months since 2500k/2600k launched and Intel hasn't even bothered to release any faster steppings of those chips. The minute Intel feels real pressure, they will bring out 3.7-4.2+ ghz SB/IB chips. Much faster clocked 2500k/2600k derivatives will again turn this race into which firm has the superior IPC performance - which has been Intel for the last 5 years.

So what you're saying is Intel is holding back technology due to lack of competition and short changing us in the process? I distinctly remember many arguing that would never happen, if for example AMD went out of business, Intel would continue to push the envelope and not rip people off due to lack of competitive pressure.

 

[OCGuy]

So what you're saying is Intel is holding back technology due to lack of competition and short changing us in the process? I distinctly remember many arguing that would never happen, if for example AMD went out of business, Intel would continue to push the envelope and not rip people off due to lack of competitive pressure.

Of course they are. Time for AMD to step up and push them.

It isn't Intel's fault, they are there to make profit. So is AMD.

 

[Riek]

AMD has a very small window to make an impression. Even if Bulldozer equals SB, IB and SB-E will launch in the next 6 months, and are almost certain to retake the lead in the mid-range and high-end markets. AMD may be able to compete with current SB processors (or perhaps even outperform them), but Intel isn't even trying right now. It's been 8 months since 2500k/2600k launched and Intel hasn't even bothered to release any faster steppings of those chips. The minute Intel feels real pressure, they will bring out 3.7-4.2+ ghz SB/IB chips. Much faster clocked 2500k/2600k derivatives will again turn this race into which firm has the superior IPC performance - which has been Intel for the last 5 years.

Ow and BD doesn't get an update in 2012? They go to an BD+ with one extra module (+25% resources in multithreading)

and much faster clocks?, 300MHz is a <10% performance difference at best.. That means if BD would average the same as 2600, intel can emeditaly go past by ~5%. That alone would be a great accomplishment by AMD.. to be only 5% worse on average (which would be insignificant comapred to the years difference it is now). So BD 95W being able to compete with 2600 woudl make AMD lightyears ahead and closer for a complete year to come then they were before.

SB-E will have a hard time with the 125W Bulldozers I believe also (it won't be a walk in the park comapred to thuban). BD actually improves alot more in low threaded compared to its 95W counterparts, which is not true for SB-E. (+7.6% for BD, 2.6% for SB-E)
If you look at the difference in multithreaded the difference between 8*2.8 or 8*3.6 is huge. (close to 30% extra for BD, SB-E is +50% cores -3% clocks.).
(assuming those rumoured clocks are correct ofcourse..)

What about clock speeds? Well, if AMD's top end CPU will have a TurboBoost of 4.5ghz, then it doesn't look like it'll have much more room for improvement on 32nm. Their only chance to regain leadership is if in the next 2-3 years most software starts to take advantage of 8+ threads.

No it would assure that the whole cpu would at least be capable of running 4.5GHz. That said a turboboost of 4.5GHz doesn't say squad about room for improvement on 32nm. For all we know they can boost it to +5GHz on a mature 32nm. It is not the process that holds back the frequencies, it the power consumption and the cpu design that does.

 

[Rifter]

So what you're saying is Intel is holding back technology due to lack of competition and short changing us in the process? I distinctly remember many arguing that would never happen, if for example AMD went out of business, Intel would continue to push the envelope and not rip people off due to lack of competitive pressure.

AMD would do the exact same thing in intels shoes, as a matter of fact they have in the past. Now we have intel $1000 EE cpus in the past we had $1000 AMD Athlon64 FX-62 CPU's.

Both companies are out there to make money and will spend as little as possible to make as much as possible.

AMD needs to get there crap together and be competative for a change instead of falling farther behind or we are going to be in a bad spot as far as CPU tech goes.

 

[RussianSensation]

So what you're saying is Intel is holding back technology due to lack of competition and short changing us in the process? I distinctly remember many arguing that would never happen, if for example AMD went out of business, Intel would continue to push the envelope and not rip people off due to lack of competitive pressure.

Remember Pentium 4 and Pentium D days? Intel was increasing clock speeds almost every quarter. There is no question that Intel isn't updating clock speeds as fast as they used to in the past. Why would they? They are probably getting higher yields at 3.3-3.4ghz SB than they would at 3.8-4.0ghz. The lower clocks are already fast enough to beat anything from AMD. It's been 8 months since SB launched and 2500k/2600k haven't received any updates. That's unheard of in the period of Athlon XP+ and Athlon 64. There was no way Intel could sit there for 8 months during those periods as AMD would have left them in the dust. We need competitive AMD back. :thumbsup:

60&#37; profit margins on Intel's end....

 

[RussianSensation]

Ow and BD doesn't get an update in 2012? They go to an BD+ with one extra module (+25&#37; resources in multithreading)

How many programs do you personally use that take advantage of more than 4 fast cores?

That means if BD would average the same as 2600. So BD 95W being able to compete with 2600 woudl make AMD lightyears ahead and closer for a complete year to come then they were before.

AMD will be more competitive, no doubt. But if top-end BD will come clocked closer to 4.0ghz, it either means it's way slower per clock OR it will blow the doors off SB. The fact that AMD had to re-spin BD already to get those clock speeds way higher likely means they are way behind in IPC. You think AMD will release an 8-core 4.0-4.5ghz BD for $300 with IPC on par with SB? That chip would be 30% faster than a 2600k.............Such a chip would be sold for $500+, not $300. That likely means BD has no chance of competing with SB per clock -- that's why they are throwing more cores against SB and trying to boost clock speeds with a new stepping (hence the delay).

SB-E will have a hard time with the 125W Bulldozers I believe also (it won't be a walk in the park comapred to thuban).

Based on what information? How in the world is Intel going to justify $600-1000 SB-E chips if a $300 BD chip will be as competitive?

SB is already 50% faster per clock than Phenom II is. 2600k @ 3.4ghz has overclocking headroom up to 4.6ghz+ with a mild Vcore increase. In fact, "The firm claims a Bulldozer module can achieve 80% of the performance of two complete cores of the same capability." If you think BD will be at par with SB per clock, then a 3.8ghz-4.5ghz 8-core AMD chip would completely blow Intel chips away. Then why haven't they released a 2.8ghz 8-core BD then? They wouldn't even need a 3.8ghz chip then......

BD actually improves alot more in low threaded compared to its 95W counterparts, which is not true for SB-E.

I wasn't aware that we had information which showed how much faster BD will be per clocked compared to Phenom II. Do you have a link?

No it would assure that the whole cpu would at least be capable of running 4.5GHz. That said a turboboost of 4.5GHz doesn't say squad about room for improvement on 32nm.

Of course it does. The more complex/deep the pipeline of the chip is, the harder it is to scale its frequency on the same 32nm process. So if BD has such an "easy" time scaling to 4.5ghz, that either means its efficiency per clock is nowhere near Intel's OR AMD is squeezing as much as possible out of it. It's very hard to get past 4.8ghz on SB on air cooling. You think AMD will be able to suddenly break 5.0ghz on BD without problems and still achieve similar IPC, while also giving us 8 cores? ---> and all that for $300? Sounds like a no compromise chip. Last time we had this was with the FX line of Athlon 64 - those CPUs had the fastest performance in almost all scenarios and they also cost $1000.

For all we know they can boost it to +5GHz on a mature 32nm. It is not the process that holds back the frequencies, it the power consumption and the cpu design that does.

The process and CPU design are intertwined. So basically you think AMD can release an 8-core equivalent SB processor clocked at 5.0ghz on 32nm all in 1 generation for $300? :whiste: Something has to give -- BD will be slower than SB per clock.

 

[Riek]

How many programs do you personally use that take advantage of more than 4 fast cores?

If there aren't any nobody would be interested in SB-E... and you don't need one program, use 2 programs that can use 4cores et voila.

AMD will be more competitive, no doubt. But if top-end BD will come clocked closer to 4.0ghz, it either means it's way slower per clock OR it will blow the doors off SB. The fact that AMD had to re-spin BD already to get those clock speeds way higher likely means they are way behind in IPC. You think AMD will release an 8-core 4.0-4.5ghz BD for $300 with IPC on par with SB? That chip would be 30&#37; faster than a 2600k.............Such a chip would be sold for $500+, not $300. That likely means BD has no chance of competing with SB per clock -- that's why they are throwing more cores against SB and trying to boost clock speeds with a new stepping (hence the delay).

Who is talking about same IPC? BD is designed with a certain thing in mind and those are high clocks with a high ipc. (yes anything like K8 or higher is an high ipc design).Again BD was never designed to be on par by clock... it was designed to allow higher frequencies and to allow more throughput (cores) per square mm.

Based on what information? How in the world is Intel going to justify $600-1000 SB-E chips if a $300 BD chip will be as competitive?

Since SB-E will have advantages BD doesn't have.. like 12threads vs 8 (50% more threads). I also never stated that BD will be better or equal then 6core SB-E. I said that SB-E can have some trouble with BD. That it won't be like current Gulftown vs thuban. in contrast to what most people believe: single threaded will barely improve with SB-E and the improvement will be mostly in the threaded part.. the thing which should also be a strong point in BD. ergo SB-E will not have the advantage we are seeing today.

SB is already 50% faster per clock than Phenom II is. 2600k @ 3.4ghz has overclocking headroom up to 4.6ghz+ with a mild Vcore increase. In fact, "The firm claims a Bulldozer module can achieve 80% of the performance of two complete cores of the same capability." If you think BD will be at par with SB per clock, then a 3.8ghz-4.5ghz 8-core AMD chip would completely blow Intel chips away. Then why haven't they released a 2.8ghz 8-core BD then? They wouldn't even need a 3.8ghz chip then......

Again with the fixation on ipc. overclocking doesn't matter!!! What matters is what both can achieve at stock within a certain TDP. Again with the misunderstandings.. You seem to read what you want to read. But yes, if BD (8150) has the same performance as the 2600 in single threaded then yes that BD will be alot faster than the 2600 in multithreaded.. 3800->3400 is 10% lower. using 4core+HT (20%) is lower then
4200 -> 3600 which is 14% lower over 8core with a throughput limitation

so 90*4*1,2 = <4,2 times the performance ofg a single thread in best case.
86 * 4 * 1,8 = <6,1 times the performance of a single thread in best case. (which is about the same as the theoretical max improvement of SB-E in multithreaded)

Again BD won't be faster than SB-E 6c, but it will be extremely close in some applications and might even take some.
(i add this so i can quote it when you again twist everything)

I wasn't aware that we had information which showed how much faster BD will be per clocked compared to Phenom II. Do you have a link?

Again with the ipc fixation.

Of course it does. The more complex/deep the pipeline of the chip is, the harder it is to scale its frequency on the same 32nm process. So if BD has such an "easy" time scaling to 4.5ghz, that either means its efficiency per clock is nowhere near Intel's OR AMD is squeezing as much as possible out of it. It's very hard to get past 4.8ghz on SB on air cooling. You think AMD will be able to suddenly break 5.0ghz on BD without problems and still achieve similar IPC, while also giving us 8 cores$1000.

thesmaller the logic that needs to be passed per clock the higher the clockspeeds. So no it is easier to scale the frequency on a processor optimiwed for that (given their are no other path bottlenecks).

The efficiency per clock is never stated to be equal. You are just focussing on the one measure while overthrozing all the others. Stop fixating on SB as the ultimate design... the design that determines max ipc and max frequency possible... How come SB uses less power (in non avx code) as nehalem while having a higher ipc andhigher frequencies on the same process... it must be magic!!! because according tro your logic that wouldn't be possible.

The process and CPU design are intertwined. So basically you think AMD can release an 8-core equivalent SB processor clocked at 5.0ghz on 32nm all in 1 generation for $300? :whiste: Something has to give -- BD will be slower than SB per clock.

The process are not the same and nor are the cpu designs. You trying to define the max frequency of BD by using SB as a measurement stick is completely rediculous . Thats like trying to determine the olive growth in italy by measuring the corn growth in America.

 

[Idontcare]

Again BD was never designed to be on par by clock... it was designed to allow higher frequencies and to allow more throughput (cores) per square mm.

I cannot imagine the fanboy backlash that would ensue if Intel came out and said their products were going to do the same.

If higher clocks, more cores, and lower IPC/core were the way of the future then I guess we should prepare ourselves for Intel dusting off their Netburst portfolio and delivering some 24-core P4 chips on 22nm :hmm:

 

[Troll Trolling]

I cannot imagine the fanboy backlash that would ensue if Intel came out and said their products were going to do the same.

If higher clocks, more cores, and lower IPC/core were the way of the future then I guess we should prepare ourselves for Intel dusting off their Netburst portfolio and delivering some 24-core P4 chips on 22nm :hmm:

I think it would be fun if Intel produced a small number of Pentium 4s with the 22nm lithography.

Wonder how it would clock, power consumption and how many Netbursts cores they could fit in the same die size as Sandy Bridge.