Bulldozer delayed til Q3?

The bulldozer has potential for both an IPC increase and IPC decrease. As for charging more money, 8 core and much higher base + turbo frequencies can make up somewhat for a less than great improvement in IPC.

I thought one of those recent slides they presented indicated that were increasing the IPC?
 
I do not think AMD has ever officially stated that. They have made made several claims of performance where they claimed increases versus magny-cours (X% more performance and Y% more cores at the same power draw) but none of these really require a large IPC improvement. I am thinking X was 50% and Y was 33%. However magny-cours is clocked low. An increase in clock speed + more cores can arrive at the same performance figures.
 
I do not think AMD has ever officially stated that. They have made made several claims of performance where they claimed increases versus magny-cours (X% more performance and Y% more cores at the same power draw) but none of these really require a large IPC improvement. I am thinking X was 50% and Y was 33%. However magny-cours is clocked low. An increase in clock speed + more cores can arrive at the same performance figures.

JF did say that IPC increases. I think the problem with BD is that is obviously isn't cabable in a single core of hitting the max IPC of even a Phenom. This is going to kill it in synthetic tests, but real world the BD should be closer to its max and therefore be an IPC upgrade over PHII.

Its design seem to be that of fighting smarter not harder.
 
Yeah, AMD's going to charge premium price (over $300 for the top unit) for CPU' that are barely faster than the ones they are EOL'ing. If so, economic suicide must be their new MO.

Furthermore, the author of that article is the same dude who wrote this:

http://www.theinquirer.net/inquirer/news/1005560/amd-socket-am2-has-a-secret-weapon

Barely faster?
Why would AMD be having a MASSIVE gaming hoedown in Texas in 3 weeks if they are releaseing a "meh" CPU? Phenom 2 performance compared to Intel's is dismell, in this instance AMD takes a beating like a 5 year old VS Mike Tyson...
Do you guys think they are going to ask one of the most critical review sites out there to review thier shiney new CPU if it was not much better than before? Try typing "Hardocp" in google and see if "Infinium labs" doesn't also show up. Why would AMD fuck with the bull when they know they would get the horns?
The performance of Bulldozer is going to be a lot better than this bull shit speculation and way better than the Intel Fanboys are saying, as manifested by thier desire to "put up or shut up" on the 16th of July.
/thread
 
JF-AMD has said 50% more performance for 33% more cores. If we take him at his word it doesn't say much for IPC improvement (+13%). We do know Bulldozer has disproportionately more FP power, so it could translate to more than a 50% total increase for HPC code like Folding. That's the part I'm looking for mostly, my internet browsing and video watching doesn't take much at all.
 
JF-AMD has said 50% more performance for 33% more cores. If we take him at his word it doesn't say much for IPC improvement (+13%).
And that's exactly why I think Bulldozer will not catch up to even Nehalem in single threaded performance. The performance disparity between a Phenom II and a Nehalem is much greater than 13% at the same clock speed. There was also large performance gap between Phenom II and Penryn before it.

Multi-threading will be where the bulldozer architecture shines. However, the performance disparity clock for clock, core to core between a Phenom II and Nehalem is still large. This disparity has only shrunk as AMD released Thubans with 6 cores. In some very rare cases, they are on par in heavily multi-threaded benchmarks.
The real question is, did JF-AMD mean the performance improvement of 50% was at the same clocks as the previous generation? If so, that could close the gap on multi-threaded performance close to current gen Intel chips. On the desktop, that won't mean much since very few desktop applications can even scale beyond 4 threads. All they can hope for is a CPU that is powerful enough to not bottleneck their current multi-gpu setups using their own GPU's. But on the server/workstation side, this chip has the potential to be a monster.

As for the whole performance disparity core for core, people need to get over it. The same thing happens on the GPU side. Granted the architectures are completely different, but the situation is analogous. Since the HD2000 series (perfected in the 4000 series) AMD has thrown as many simple shader processors into a GPU core that was physically possible. Nvidia increased the amount of shaders but made fewer of them that were much more complex. I believe AMD is shooting for the same goal on the CPU side as long as they can create enough cores to increase performance without going overbudget on transistors and die size.

I honestly believe that Bulldozer will be the budget performance per dollar chip on the desktop side, just like Phenom II before it. The performance and price disparity won't be as large as before, and it will be "just enough" performance increase and overclockability to justify a higher price within range of Intel's core series.

On the server side, I expect the Bulldozers will make significant gains in performance and will be worth their hefty premium.
 
Last edited:
JF-AMD has said 50% more performance for 33% more cores. If we take him at his word it doesn't say much for IPC improvement (+13%). We do know Bulldozer has disproportionately more FP power, so it could translate to more than a 50% total increase for HPC code like Folding. That's the part I'm looking for mostly, my internet browsing and video watching doesn't take much at all.
I don't know... In the case of Magny-Cours vs Interlagos (what JF-AMD was referring to), 16 cores at similar clocks nets you 50% higher performance that 12 cores. Basically, at similar clocks, it would take 18 'Stars' cores to equal 16 'Bulldozer' cores, a pretty nice bump in IPC. Example: At 3.3GHz it would take a 9-core Phenom II to equal a 3.3GHz FX 8-core chip (gross oversimplification, I know).

For browsing and videos, a lowly Athlon II or a new Llano system is more than enough.
 
i do browsing and video on a 5 year old laptop with an Athlon64 3000+ under linux just fine :)
 
JF-AMD has said 50% more performance for 33% more cores. If we take him at his word it doesn't say much for IPC improvement (+13%). We do know Bulldozer has disproportionately more FP power, so it could translate to more than a 50% total increase for HPC code like Folding. That's the part I'm looking for mostly, my internet browsing and video watching doesn't take much at all.
Performance increases aren't linear to the number of cores (e.g. a dual core does not net twice the performance of a single core). This isn't a perfectly efficient system you know.
 
Since this thread is about release date, AMD doesn't have much in the way of excuses left. We know they hauled the release date back due to initial stepping performance issues but they won't pull the release completely, they hyped too much this time. Their Fusion chips are doing well and Q4 will be focused on Southern Islands graphics chips for the holiday season. Even if Bulldozer is the same as the current Phenom II core for core, it'll be good enough. They can work on shaking out the bugs on Bulldozer II next year, like how Nvidia survived through the first Fermi. There's nothing wrong with continuing to hit performance/price ratios for another year if the chip is usable and improvements would mean delaying past the summer shopping season.

Performance increases aren't linear to the number of cores (e.g. a dual core does not net twice the performance of a single core). This isn't a perfectly efficient system you know.

JF is a server guy so he probably wasn't talking about multi-threaded programs.
 
JF is a server guy so he probably wasn't talking about multi-threaded programs.

JF also didn't do the tests (probably), and I don't remember if he said what it was measured by. Basically the point is that there's not enough information out to be able to really figure out anything. We don't know if it was single threaded to test server loads or massively threaded to test HPC application. Hell, we don't even know what clock speed the Interlagos chip was running at (especially given that it was pre-release hw)
 
I don't know... In the case of Magny-Cours vs Interlagos (what JF-AMD was referring to), 16 cores at similar clocks nets you 50% higher performance that 12 cores
Similar clocks was not mentioned in JF's quote. ;) At the time that was written, 12 core Magny Cours topped out at 2.3GHz. A 2.5GHz 12 core version was released 6 months later. The often quoted 13% per core improvement is nothing more than a guess. It's based on a throwaway estimate comparing MC @ ?.?GHz to a guestimated BD target frequency. Handful of salt. If there is one performance estimate to ignore for launch parts at least, it would be "we estimate that customers will see up to 50% more performance from 33% more cores."

http://blogs.amd.com/work/2010/08/02/what-is-bulldozer/
From a performance standpoint, if you compare our 16-core Interlagos to our current 12-core AMD Opteron™ 6100 Series processors (code named “Magny Cours”) we estimate that customers will see up to 50% more performance from 33% more cores. This means we expect the per core performance to go in the right direction — up.​

BD clock speeds for Interlagos have not been released yet. The estimate may have been overzealous. What seems to be most important right now is just showing that it exists and works: http://news.softpedia.com/news/AMD-Demos-Working-16-Core-Interlagos-CPU-at-ISC-2011-207521.shtml ;)
 
Confirmed from a couple of sources.....September 6th at this time. Surely it could move again. So not to say, "I told you so," but yeah, I did already. Not happy about it either.
 
I guess we can safely say that the "we're delaying this for strategic reasons" response was a load of crock. Clearly, things aren't going smoothly with the processor. 3-5 months with AM3+ boards, and no product. Jawsome!


I am not so sure that AMD is keeping the Llano production priority right now. If you can sell every one of those you can produce, you might want to keep doing that....that is just a guess on my part though, but I have heard this theory batted around by AMDers.
 
True.

Well at least that date is better than the "end of September." Jeez, this thing better be good.
 
Well, shit.

At least Llanos are dribbling out. I got a couple of HTPC builds for myself and some family members to do.
 
lol this new chip better blow sandybridge out of the water.

and also perform closely enough to Ivy Bridge to mater...


oh, shit, this means I just started a Rumor that Bulldozer is competitive with Ivy Bridge!!
 
I guess we can safely say that the "we're delaying this for strategic reasons" response was a load of crock. Clearly, things aren't going smoothly with the processor. 3-5 months with AM3+ boards, and no product. Jawsome!

Zambezi sells to the enthusiast market, which is <10% of market. Llano sells to the mainstream which is probably 70%+ of the market.

Zambezi sells only into desktops (that I am aware of) which is a shrinking market and Llano sells into desktop and mobile (which is a growing market.

To me that is the essence of strategy - prioritize products around a large, growing market and over a small, shrinking market. But, then what do I know, I'm a server guy.
 
I think they are delaying wayyy too much and most of the users are getting tired of waiting and have already made the switch to sandy bridge
 
I think AMD is trying to release too much at the same time and with all these push backs, I don't see them releasing their new GPU's anytime soon (this year:mad:).
 
Luckily, AMD's GPU division doesn't use the same fabs for their primary products (so far).
 
True, I hope that I'm wrong. They've only just got 32nm out the door, now I don't see them releasing a new process (28nm) within 6 months.
 
Zambezi sells to the enthusiast market, which is <10% of market. Llano sells to the mainstream which is probably 70%+ of the market.

Zambezi sells only into desktops (that I am aware of) which is a shrinking market and Llano sells into desktop and mobile (which is a growing market.

To me that is the essence of strategy - prioritize products around a large, growing market and over a small, shrinking market. But, then what do I know, I'm a server guy.

So you're saying BD is fine and dandy, but you keep it under wraps to fulfill Llano demand ?

Of course, with your enviable position in servers and high end segment, it is better to sell one 230mm^2 Llano for $60-80 than a 290mm^2 BD for $ 250-1000. Amazing strategy....a case study in strategic management for years to come.
 
So you're saying BD is fine and dandy, but you keep it under wraps to fulfill Llano demand ?

Of course, with your enviable position in servers and high end segment, it is better to sell one 230mm^2 Llano for $60-80 than a 290mm^2 BD for $ 250-1000. Amazing strategy....a case study in strategic management for years to come.

I'd guess fab yields and market demand have more to do with it*
You have to remember, high end is a miniscule segment of the market, and HPC servers are another market entirely (which was already slated for Q3 2011 by AMD, so nothing lost there). If AMD can sell a metric shit-ton more 60-80usd 230mm^2 chips, enough to make more than fewer 290mm^2 (which would, assuming same amount of redundancy in deisgn, have lower yields, therein raising the fab price up) processors, then... why not? Speculated pricing based off of AMD presentations indicate nothing over 400usd for the desktop market, anyhow.

Just my take at this.



*along with the more likely performance fixes
 
^That's why I'm interested in BD, I'm in the market to buy two dual socket servers this summer, Intel's duel socket offerings are all EOL'd (I refuse invest in sockets/ processors with zero upgrade options), so AMD has an excellent opportunity here! But it looks as though I am going to have to buy some cheap E3 xeons (bang for buck is excellent on those) to tie me over till next year when I price war breaks out.

I was looking forward to some 32 core goodness..
 
So you're saying BD is fine and dandy, but you keep it under wraps to fulfill Llano demand ?

Of course, with your enviable position in servers and high end segment, it is better to sell one 230mm^2 Llano for $60-80 than a 290mm^2 BD for $ 250-1000. Amazing strategy....a case study in strategic management for years to come.
Where are you getting this $250-$1000 figure from?
 
So you're saying BD is fine and dandy, but you keep it under wraps to fulfill Llano demand ?

Of course, with your enviable position in servers and high end segment, it is better to sell one 230mm^2 Llano for $60-80 than a 290mm^2 BD for $ 250-1000. Amazing strategy....a case study in strategic management for years to come.
You are looking at this from the wrong angle. Even if the markup on an FX chip makes it highly profitable, the sheer demand (70% of market) from a low markup part like Llano more than makes up for the temporary loss of the high end customers who make up 10% of the market. They also gain from having an AMD Vision sticker on each product sold, which helps a small bit to win over the hearts and minds of Average joe who is 90% of the customer base.
The only real x factor in all of this is the insanely marked up server parts which make up a much larger portion of the market (remaining 20%) than the high end desktop market. As long as AMD can produce enough of those to meet demand also, they should be fine.
 
I'd guess fab yields and market demand have more to do with it*
You have to remember, high end is a miniscule segment of the market, and HPC servers are another market entirely (which was already slated for Q3 2011 by AMD, so nothing lost there). If AMD can sell a metric shit-ton more 60-80usd 230mm^2 chips, enough to make more than fewer 290mm^2 (which would, assuming same amount of redundancy in deisgn, have lower yields, therein raising the fab price up) processors, then... why not? Speculated pricing based off of AMD presentations indicate nothing over 400usd for the desktop market, anyhow.

Just my take at this.

Instead of selling 1000 Llanos, don't you think AMD would be better off with 900 Llanos and 100 BDs if it is ready ? Short answer : it is damn obvious. Why aren't they doing it ? Because BD is still not ready contrary to what JF said and the overall BS theory of favouring Llano.




Where are you getting this $250-$1000 figure from?

Desktops are around $300 and workstation/server CPU span from $250 to around $1300 last I've checked for Magny Cours. Do you expect BD not to fit in these price ranges ?


You are looking at this from the wrong angle. Even if the markup on an FX chip makes it highly profitable, the sheer demand (70% of market) from a low markup part like Llano more than makes up for the temporary loss of the high end customers who make up 10% of the market. They also gain from having an AMD Vision sticker on each product sold, which helps a small bit to win over the hearts and minds of Average joe who is 90% of the customer base.
The only real x factor in all of this is the insanely marked up server parts which make up a much larger portion of the market (remaining 20%) than the high end desktop market. As long as AMD can produce enough of those to meet demand also, they should be fine.

Like I've said, if Zambezi is ready, why not sell 900 Llanos +100 BDs instead of 1000 Llanos ?
 
Desktops are around $300 and workstation/server CPU span from $250 to around $1300 last I've checked for Magny Cours. Do you expect BD not to fit in these price ranges ?
Since nobody here is talking about Interlagos being delayed (except apparently you) why are you throwing that into a discussion about whether Zambezi or Llano is a higher priority for AMD? Server side roadmaps work independently from desktop as they have for years now. This has been discussed ad nauseum already in this forum.
 
Since nobody here is talking about Interlagos being delayed (except apparently you) why are you throwing that into a discussion about whether Zambezi or Llano is a higher priority for AMD? Server side roadmaps work independently from desktop as they have for years now. This has been discussed ad nauseum already in this forum.

What is Interlagos if not two Zambezi in the same package ? :p Since AMD is using the same design, if one is flawed and delayed, how exactly doesn't it also affect the server side ?

But let's not go off topic. My point was that I do not believe Zambezi got delayed to make room for Llano as implied by AMD-backed rumour mill. It's simply smoke and mirrors to hide another execution screwup.
 
What is Interlagos if not two Zambezi in the same package ? :p Since AMD is using the same design, if one is flawed and delayed, how exactly doesn't it also affect the server side ?

But let's not go off topic. My point was that I do not believe Zambezi got delayed to make room for Llano as implied by AMD-backed rumour mill. It's simply smoke and mirrors to hide another execution screwup.
You do realize the leaked/unofficial/official/whathaveyou roadmaps had Interlagos in Q3 right? So again, please explain why/how you think Interlagos is delayed so that I can understand the relevance of Interlagos pricing and profit to the topic at hand. If Zambezi got delayed from Q2 to Q3 and Interlagos had the same exact problem but was still released in Q3 that wouldn't be a delay now would it? Have you still not noticed that your post with the "$ 250-1000." figure was directly quoting JF-AMD: "Zambezi" & "desktop".
 
You do realize the leaked/unofficial/official/whathaveyou roadmaps had Interlagos in Q3 right?

True. But It could be July or late September. Since Interlagos is using 2 8 core dies, wouldn't you expect issues with one to affect the other ? The only case in which this won't apply is if whatever problems Zamebezi has ( as speed path issues / thermals in getting to high frequency ) do not apply to Interlagos ( being a lower speed product ).
So again, please explain why/how you think Interlagos is delayed so that I can understand the relevance of Interlagos pricing and profit to the topic at hand. If Zambezi got delayed from Q2 to Q3 and Interlagos had the same exact problem but was still released in Q3 that wouldn't be a delay now would it?

No it wouldn't if we're looking at Q granularity. However Zambezi did not simply get delayed from Q2 to Q3, it got delayed to late Q3 and most likely availability in Q4. Even more interesting is that the higher speed models got pushed all the way to Q1 2012 from the Digitimes article. Could that imply an ever longer delay for the high speed models ?
Have you still not noticed that your post with the "$ 250-1000." figure was directly quoting JF-AMD: "Zambezi" & "desktop".

Yes, I did. But he's a server guy, so I pushed it into his direction. He suggested that Zambezi is delayed not because of problems, but to make room for Llano. Wouldn't that mean servers are also delayed even if it will still launch in the same Q ? I fail to see the logic in doing that, thus I implied the original premise is wrong.
 
Instead of selling 1000 Llanos, don't you think AMD would be better off with 900 Llanos and 100 BDs if it is ready ? Short answer : it is damn obvious. Why aren't they doing it ? Because BD is still not ready contrary to what JF said and the overall BS theory of favouring Llano.

You're assuming that AMD has customers lined up for those "100" BDs. You're also not applying certain properties of microeconomics that are relevant to this type of supply decision. I find that a reading on the the [ame="http://en.wikipedia.org/wiki/Guns_versus_butter_model"]Guns versus butter model - Wikipedia, the free encyclopedia[/ame] would be apropos here (even though it's technically applies to macroeconomics.)

Let's use your numbers for a simple example. Say AMD can only crank out 1000 32nm chips. If there is demand for Llano that is >= to 1000 then they can sell everyone of those Llanos. But maybe demand for BD chips isn't so high. Maybe it's only 50 BD chips. So, if they made 100 BD chips they would not only not sell 50 BDs but there is also an opportunity cost in the form of Llano chips that could have been made (and also were not sold).

So if it takes, I dunno, $30 to make the chip and AMD sells Llano chips at $100 a piece then the total cost to AMD is $130 for every BD chip not sold (since we're assuming they can sell every Llano they can conceivably make). Technically the opportunity cost still applies to all those chips AMD made BDs instead of Llanos, but I thought is was easier to comprehend this way.

I dunno what AMD is up to, but this would be a scenario where not releasing BD in conjunction with Llano would make sense.
 
Back
Top