Fudzilla: Bulldozer performance figures are in

[frozentundra123456]

In a budget gaming rig (average gaming machine is $600-700 in most of the world) an extra $100 towards the video card will always give more performance then an extra $100 in the CPU.

For a $1300-1500 budget sure Intel's higher end CPUs are fine, I keep recommending those on forums. So are expensive cases or even RAM. But this was about the bang for the buck. For gaming in full HD once you have a powerful video card the diff is only some 10-15% between the 2500K and the Ph II X4.

Definitely the Intel is not twice as fast in anything and mostly not in gaming.

There are some 1 billion PCs in the world, the enthusiast market is maybe 5%. Same with the cars. Ferrari, Porsche, Lamborghini, Aston Martin, etc don't sell many cars.


IMO the most interesting part of the Bulldozer line are the budget quads. I would like a quad that is close to the i-3 in single threaded apps and has the benefits of the extra cores.

I wont dispute any of what you say. However, even for a budget system, I dont really see why people are so concerned about 100.00 or even more extra for the CPU. If you game on the computer, that is only equal to a couple of games, or a month of two cell phone bill, or a couple of times to go out to eat, etc. My point being that for most people, we could easily cut expenses this much and use the extra to invest in a better CPU. Maybe if someone is a student or really strapped for cash, this wouldnt apply, but in most cases I think it would.

And if you consider that the extra investment could lengthen the interval between upgrades, I think it could be money well spent, as well as making the computer more proficient in other tasks besides gaming.

 

[BlueBlazer]

Timing doesn't work out quite right for it to be a new stepping. Could just be more time needed to get enough chips in hand so as not avoid doing a paper-launch.

That's an extra 2-3 months though, if the last batch was in August. :hmm: Anyway, some side notes confirming the delay? >> AMD aggressive about cloud computing business in Greater China, say executives........

Meanwhile, AMD's server processor codenamed Interlagos will also have difficulty shipping on schedule and is expected to be delayed to November.

 

[jmanny]

Now up to 81 pages of crap

This

 

[mosox]

The fastest growing PC market is in China, also India. We're talking about guys who use Raidmax, Sirtec, Delux, Great Wall PSUs and the cheapest cases you can imagine, made in small shops and who are running old IDE HDDs.

The 2500K is $300 in my country (E Europe) and $100 is a lot of money when you don't have it so the general trend is the Athlon II x3 or X4 and the cheapest mobo.

I just hope AMD keeps producing cheap quads, it seems more and more games need more than 2 cores.

 

[soccerballtux]

This

only 21 pages to me

 

[StrangerGuy]

Oops repost.

 

[Arachnotronic]

Just awesome.

 

[Tuna-Fish]

Sandy Bridge has 3 more functional units than Phenom II so it is 100% higher IPC(Phenom II having 3 functional units functioning at any given time and Sandy Bridge having 6 functional units functioning at any given time is a big boost)

While it's true that Intel counts SB as having 6 functional units, comparing them to phenom like that is probably wrong, mostly because the port 4 is only ever used for store data, which is something that doesn't take an additional execution unit on K8 and derivatives. So for integer code, it's more like 3 vs 5. For SIMD/FP programs, it's even weirder, because Intel uses the same ports for integer and SIMD instructions, while AMD has separate 3 ports for them. So on mixed SSE/integer code, AMD can issue 6 ops, and Intel can only issue 5.

The Intel way is imho better, but it's not 100% better.

 

[BigDH01]

Larrabee was never as far along in the design process as this is. There comes a point in a project where cancelling it would be just a big waist of money. You are better to just release what you have and make whatever money you can.

It doesn't have to be a big waist (). If AMD has strong demand for their server chips with a low supply, it might make sense to cancel Zambezi for the desktop. If you have an inelastic supply, might as well get as much money as possible for it. If I'm AMD right now and have a winner in Llano, a winner in Interlagos, with a very short supply of BD parts, I might stick with selling server parts. I highly doubt Zambezi was going to take the desktop performance crown anyway, and margins are constrained here by Intel. AMD has already positioned themselves to be competitive in highly multi-threaded scenarios and on a price/performance basis that I can't imagine the desktop enthusiast market is really worth it. Nice for fanboys, but won't really help the bottom line.

If Tankguy is accurate and his suppliers say early November for ETA, they really mean early November at the earliest. Judging by AMD's history, you're really talking Dec/Jan timeframe, especially as the first BD chips will be going to servers, and rightfully so.

I don't know, at this point, if I'm AMD, I'm surveying the situation (and I'll throw out the most realistic scenario here). I've got a new chip that everyone has been waiting for for years. I've missed the release date over and over now, and a very vocal minority of my customers are quite upset. My new chip can't really compete with Intel's best in gaming (the primary task undertaken by this vocal minority), but shines in heavily multithreaded scenarios. I've got a limited supply of this chip which I can sell to the vocal minority for small margins or sell to the server market for high margins where the strengths of this chip are better utilized. Do I really need to be concerned about that vocal minority? In all honesty, the wind has already been knocked out of my marketing sails for any sort of major launch on the desktop anyway, and my cheaper desktop option is doing very well, in fact, I can't keep up with demand. If I sat down and looked at the numbers, I might cancel Zambezi for the desktop. Supply is simply too constrained and the chip can now never live up to the expectations after repeated delays. And by not releasing it on the desktop, I lose what, 1% of potential revenue next quarter (just a guess)? More than made up for by the fact that I can keep my higher margin channels supplied.

 

[Despoiler]

I wonder how the November launch rumblings squares with the exec video confirming an October launch. I'm guessing Oct 31 launch or some such late October silliness. It's that or a mid October quasi paper launch with parts available, but in very low quantities. At this point AMD really needs to get on top of this. Letting the rumor mill run is bad for everyone, but Intel.

 

[Troll Trolling]

The Intel way is imho better, but it's not 100% better.

Do you think it is better because it is faster or because it is more efficient per W and mm²?

Serious question.

 

[trollolo]

Do you think it is better because it is faster or because it is more efficient per W and mm²?

Serious question.

both

 

[LOL_Wut_Axel]

Uh, nice fallacy.
Make a claim, get a counter claim, ask for documentation for the counterclaim.
It called reversal of proof.

You made an argument.
You document that argument.
Or do you think people have to prove their innocence too in order to not be guilty?

I provided evidence for my claims in the last page. He didn't.

That's the difference.

 

[LOL_Wut_Axel]

The basic facts, as I said in a previous page:

Intel had mostly dominated the CPU market ever since 2006:

• Core 2 was introduced. While it had much lower clock speeds than the Pentium D, it brought along with it a ~100% IPC boost.
• 25% higher IPC than the Athlon 64 X2
• Introduced using a 65nm process node; AMD's Athlon 64 X2 used a 90nm process node
• Much smaller die: 143mm^2 vs 183mm^2
• Much better performance/watt: >50% higher
• Aggressive pricing: E6600 and E6400 were priced comparably to 4600+ and 4200+

And afterwards:

• K8 had headroom for high clock speeds
• K10 brought IPC improvements, but at the expense of extremely high power consumption and low clock speeds
• K10.5 had ~23% higher IPC than K8; Penryn had ~5% higher IPC than Conroe
• Intel retained their manufacturing process lead, coming to market with the 45nm Penryn in 2007 and AMD taking until 2009 to come with the 45nm Phenom II/Stars
• A year after this, Intel already had 32nm Westmere

The current situation:

• Sandy Bridge enjoys >100% better performance/watt than Phenom II
• Intel is moving to 22nm early next year; AMD still hasn't come to Performance market with 32nm

Look at what I wrote, and now apply it to NVIDIA and AMD since 2008. You'll see a lot of similarities when it comes to aggressive pricing, die sizes, performance/watt, and manufacturing leads. AMD hasn't sold as many GPUs as NVIDIA, but they've definitely had the best GPUs overall.

 

[Tuna-Fish]

Do you think it is better because it is faster or because it is more efficient per W and mm²?

Whether it is more efficient per W and mm² is an extremely hard question to answer without the kind of data only the CPU manufacturers have, essentially because it's extremely hard to isolate the effects of the different approaches from the other pieces of relevant architecture and the manufacturing process. We don't even conclusively know that it would be faster on identical process tech -- it's obviously faster per clock, but if AMD had access to Intel fabs, who's to say it wouldn't clock more to offset that?

Based on what little we do know, I think that the Intel approach of segregated memory units is better because they can skimp on forwarding. To elaborate more, the units themselves are mostly uninteresting -- a fast adder is a fast adder, and they haven't changed much in decades. What's interesting is keeping the units fed and getting data from unit to another one as fast as possible.

On the K8 approach, all the units need to be able to forward results to each other -- they all might have the result that's needed for the next uop at a different unit. As adding just one more unit more than doubles the amount of connections needed by forwarding network, and also increases the physical length of those connections, you are seriously limited on the amount of units you can support.

On the Intel approach, they can cut away some forwarding paths -- an AGU is never going to forward to an ALU, so you don't need those paths. This is why Intel can put the 5 units to it's cores, and AMD can't.

AMD has more or less validated this idea with their BD and bobcat cores -- which use segregated AGU's and ALU's.

 

[-Slacker-]

To whom it may concern:




Updating the thread with 100 posts per day makes it look like someone finally posted some genuine leaks, which is inconvenient for posters who just want to see some actual numbers and are not interested in foraging through pages of text only to find a wall of highly speculative debates.

Yes, I know you're all very smart with thinking, and all this technical jargon and math problem solving is all very impressive, but ... there are a billion other threads on AT that are catching dust, and billions more to be made ... just sayin'

 

[Troll Trolling]

Whether it is more efficient per W and mm² is an extremely hard question to answer without the kind of data only the CPU manufacturers have, essentially because it's extremely hard to isolate the effects of the different approaches from the other pieces of relevant architecture and the manufacturing process. We don't even conclusively know that it would be faster on identical process tech -- it's obviously faster per clock, but if AMD had access to Intel fabs, who's to say it wouldn't clock more to offset that?

Based on what little we do know, I think that the Intel approach of segregated memory units is better because they can skimp on forwarding. To elaborate more, the units themselves are mostly uninteresting -- a fast adder is a fast adder, and they haven't changed much in decades. What's interesting is keeping the units fed and getting data from unit to another one as fast as possible.

On the K8 approach, all the units need to be able to forward results to each other -- they all might have the result that's needed for the next uop at a different unit. As adding just one more unit more than doubles the amount of connections needed by forwarding network, and also increases the physical length of those connections, you are seriously limited on the amount of units you can support.

On the Intel approach, they can cut away some forwarding paths -- an AGU is never going to forward to an ALU, so you don't need those paths. This is why Intel can put the 5 units to it's cores, and AMD can't.

AMD has more or less validated this idea with their BD and bobcat cores -- which use segregated AGU's and ALU's.

Enlightening response, thank you.

 

[Vesku]

When hard BD results come in there will be more than just this thread around. You may be incredibly smart but I'm just pointing out that actual BD release information won't be hard to find ONCE IT EXISTS.

Just sayin' ;p

To whom it may concern:




Updating the thread with 100 posts per day makes it look like someone finally posted some genuine leaks, which is inconvenient for posters who just want to see some actual numbers and are not interested in foraging through pages of text only to find a wall of highly speculative debates.

Yes, I know you're all very smart with thinking, and all this technical jargon and math problem solving is all very impressive, but ... there are a billion other threads on AT that are catching dust, and billions more to be made ... just sayin'

 

[Vesku]

http://www.canardpc.com/pdf/

More evidence of an Oct 12th launch. Haven't read their no numbers review in their #10 issue but 2nd hand reports say they report SB gets the better gaming results but it was easy to OC the FX.

 

[inf64]

Whether it is more efficient per W and mm² is an extremely hard question to answer without the kind of data only the CPU manufacturers have, essentially because it's extremely hard to isolate the effects of the different approaches from the other pieces of relevant architecture and the manufacturing process. We don't even conclusively know that it would be faster on identical process tech -- it's obviously faster per clock, but if AMD had access to Intel fabs, who's to say it wouldn't clock more to offset that?

Based on what little we do know, I think that the Intel approach of segregated memory units is better because they can skimp on forwarding. To elaborate more, the units themselves are mostly uninteresting -- a fast adder is a fast adder, and they haven't changed much in decades. What's interesting is keeping the units fed and getting data from unit to another one as fast as possible.

On the K8 approach, all the units need to be able to forward results to each other -- they all might have the result that's needed for the next uop at a different unit. As adding just one more unit more than doubles the amount of connections needed by forwarding network, and also increases the physical length of those connections, you are seriously limited on the amount of units you can support.

On the Intel approach, they can cut away some forwarding paths -- an AGU is never going to forward to an ALU, so you don't need those paths. This is why Intel can put the 5 units to it's cores, and AMD can't.

AMD has more or less validated this idea with their BD and bobcat cores -- which use segregated AGU's and ALU's.

For your information, SB has 3 ALU ports and 2 "AGU" ports.

The scheduler will issue the oldest, ready to execute uops each cycle to the six available ports, depending on the type of uop. Ports 0, 1 and 5 are used for executing computational uops. There are three types of computational uops and execution units or execution stacks: integer, SIMD integer (which we will refer to as SIMD) and FP (either scalar or SIMD). Each of the 3 execution ports has hardware for the 3 different types of uops.

K10 has symmetric ALU/AGU organization with 3 ALU and 3 AGU units and separate FP coprocessor unit. It can execute up to 9 uops but retire only 3 macro ops .

BD has separate ALU and AGU pipelines and again a fp coprocessor which can execute 4 "cops". For BD AMD said it is executing and retiring "Cops" or complex ops (analog to macro op term in K10). BD module can execute at any given cycle (in theory) 2x4 cops(2x2 in 2 integer cores and 4 in FlexFP) and retire 8 cops (or 4 per core in the module). The retired cops are a summed retired c0ps by both FlexFP and each integer core (achieved throughput per core is 4 cops according to Mike Butler chief architect of BD).This is up from K10 by 33%.

 

[Vesku]

So apparently the FX-8150 and FX-6100 are somewhere on these charts compared to some Sandybridge CPUs. Anybody willing to play pin the CPU to it's percent bars prior to Oct 12th?

 

[frozentundra123456]

It doesn't have to be a big waist (). If AMD has strong demand for their server chips with a low supply, it might make sense to cancel Zambezi for the desktop. If you have an inelastic supply, might as well get as much money as possible for it. If I'm AMD right now and have a winner in Llano, a winner in Interlagos, with a very short supply of BD parts, I might stick with selling server parts. I highly doubt Zambezi was going to take the desktop performance crown anyway, and margins are constrained here by Intel. AMD has already positioned themselves to be competitive in highly multi-threaded scenarios and on a price/performance basis that I can't imagine the desktop enthusiast market is really worth it. Nice for fanboys, but won't really help the bottom line.

If Tankguy is accurate and his suppliers say early November for ETA, the really mean early November at the earliest. Judging by AMD's history, you're really talking Dec/Jan timeframe, especially as the first BD chips will be going to servers, and rightfully so.

I don't know, at this point, if I'm AMD, I'm surveying the situation (and I'll throw out the most realistic scenario here). I've got a new chip that everyone has been waiting for for years. I've missed the release date over and over now, and a very vocal minority of my customers are quite upset. My new chip can't really compete with Intel's best in gaming (the primary task undertaken by this vocal minority), but shines in heavily multithreaded scenarios. I've got a limited supply of this chip which I can sell to the vocal minority for small margins or sell to the server market for high margins where the strengths of this chip are better utilized. Do I really need to be concerned about that vocal minority? In all honesty, the wind has already been knocked out of my marketing sails for any sort of major launch on the desktop anyway, and my cheaper desktop option is doing very well, in fact, I can't keep up with demand. If I sat down and looked at the numbers, I might cancel Zambezi for the desktop. Supply is simply too constrained and the chip can now never live up to the expectations after repeated delays. And by not releasing it on the desktop, I lose what, 1% of potential revenue next quarter (just a guess)? More than made up for by the fact that I can keep my higher margin channels supplied.

From a purely financial standpoint, what you say probably makes sense. However, in my opinion it would be a public relations disaster and severely inpact what positive image the company has left, especially after they way they have been touting Bulldozer as the next great thing. I mean they would then be resigned to being the low end provider for who knows how long.

And I am somewhat impressed with Llano, especially for laptops, but I am not sure it is the rampant success that some AMD fans suggest. Yes, they sell all they can make, but that is because they are having production issues as well as high (relatively for AMD) demand. Plus Ivy Bridge will be out soon and will will have much improved graphics and of course much better CPU performance than the stars core in Llano. So AMD really needs to get out a better preforming CPU instead of relying on APU graphics performance alone as a selling point.

 

[Despoiler]

So apparently the FX-8150 and FX-6100 are somewhere on these charts compared to some Sandybridge CPUs. Anybody willing to play pin the CPU to it's percent bars prior to Oct 12th?

My french is pretty rusty, but I believe they are saying something to the effect that benches with FX are all over the place and they removed some that lead too much.

 

[Vesku]

My french is pretty rusty, but I believe they are saying something to the effect that benches with FX are all over the place and they removed some that lead too much.

Hmmm, did they trim off the ones that seemed too low as well? I stopped taking French after Year 11 in High School.

 

[LOL_Wut_Axel]

From a purely financial standpoint, what you say probably makes sense. However, in my opinion it would be a public relations disaster and severely inpact what positive image the company has left, especially after they way they have been touting Bulldozer as the next great thing. I mean they would then be resigned to being the low end provider for who knows how long.

And I am somewhat impressed with Llano, especially for laptops, but I am not sure it is the rampant success that some AMD fans suggest. Yes, they sell all they can make, but that is because they are having production issues as well as high (relatively for AMD) demand. Plus Ivy Bridge will be out soon and will will have much improved graphics and of course much better CPU performance than the stars core in Llano. So AMD really needs to get out a better preforming CPU instead of relying on APU graphics performance alone as a selling point.

Trinity is supposed to be launched in Q1 2012, so we'll see. Maybe it's delayed to Q2 and is finally launched in Q3.

It should bring higher CPU and IGP performance along with lower power consumption than Llano, naturally.