YABulldozerT: AMD FX Processor Prices Lower Than Expected

[Riek]

...and yet AMD needs 8 cores and higher clock speeds to compete with a quad-core 2500K

Please explain.

First their are no performance numbers.

Second frequency,core count, complexity and .. is a design choice!! It is not wrong or bad to make that choice!!

Why are people fixated on frequency and on core count...

The fact is BD has 2cores compared to one SB core in terms of transistor count/die area.
The fact is BD has higher frequency ranges in mind compared to SB.

Those are both known design choices since the design was created, made public and discussed.

 

[Abwx]

In absolute performance Bulldozer wins

Sandy Bridge 12 ALU ops/12 Mem ops vs Bulldozer 16 ALU ops/16 Mem ops
Sandy Bridge 12 x 128bit int/fp Bulldozer 8x128bit fp 8x128 int

The FP figure is wrong.
BD can do 64 SP FP ops/cycle while SB reach this rate only
with AVX wich wont be more than 10% in a given code...

If you look at SB diagram , you see that it has only two FP
execution ports ; so your quote of 12 FPU is not accurate,
since there s only 8 FPUs although 256b wide with AVX.

As for a Zambezi , it has 16 128b FP pipe lines.

 

[NostaSeronx]

The FP figure is wrong.
BD can do 64 SP FP ops/cycle while SB reach this rate only
with AVX wich wont be more than 10% in a given code...

I forgot about that slide thanks for showing it to me

If you look at SB diagram , you see that it has only two FP
execution ports ; so your quote of 12 FPU is not accurate,
since there s only 8 FPUs although 256b wide with AVX.

Sandy Bridge Pipes:

1 FADD does FP/Int
1 FMUL does FP/Int
1 FMISC does FP/Int

First section = Core, Second and Third section equals SIMD INT and SIMD FP

Bulldozer Pipes:

1 FMAC does FP
1 FMAC does FP
1 Integer ALU(MMX) does Int
1 Integer ALU(MMX) does Int

(It's the middle green/teal/whatever color that is forest green? forest teal? portion) lol

I also did my research and found what intel has to say

1 256bit FADD
1 256bit FMUL
1 256bit FMISC
and only two of the three can occur per cycle

What happen is it takes up the whole length of the 3 ALUs

So, Doing FP AVX-256b on Sandy Bridge is harmful to single thread performance while in Bulldozer doing FP AVX-256b won't be harmful to single thread performance

You will always have 1 ALU active during the two 256bit Floating Point Ops in Sandy Bridge

While in Bulldozer you will have 4 ALUs(2 per core) active per 256bit Floating Point ops + per 256bit Integer MMX ops

Unless Sandy Bridge ALUs are not 256bit in size but are 320bits in size(then AVX 256 is unharmful)

As for a Zambezi , it has 16 128b FP pipe lines.

Zambezi has 8 128b Floating Point pipelines
2 per module, 4 modules = 8 per CPU

 

[BlueBlazer]

Here's a little interesting titbit I found at BLT, which has also been observed earlier concerning prices of the new Opterons...

OPTERON 6200 SERIES PROCESSORS 2.1G 16MB 115WT G34 TRAY= $849.02
OPTERON X12 6174 G34 2.2G 18MB 80W 6400MHZ 45NM TRAY MAGNYCOURS = $1323.73
OPTERON 6180 SE 12C G34 2.5G 105W TRAY = $1720.82

Again, Interlagos ("16-cores") seems to be cheaper than current Magny-Cours (12-cores)? :hmm:

 

[NostaSeronx]

Again, Interlagos ("16-cores") seems to be cheaper than current Magny-Cours (12-cores)? :hmm:

It's cheaper to make than Thuban and It's small as well(Higher yields)

346mm^2~ x 2 = 12 core
vs
315mm^2~ x 2 = 16 core

:hmm:
So, what you getting at mister!


$1123.42 6172


$849.02 6272

(US$ 1123.42) - (US$ 849.02) = 274.4 U.S. dollars

Seems just fine for a server part aiming for market share that has higher yields and is cheaper to make
​

 

[DirkGently1]

Just confirms what seems to be becoming more and more clear lately; BD isnt going to a very attractive CPU for enthusiasts.

Are you saying enthusiasts can't be budget conscious? I would say that as enthusiasts we upgrade the most frequently, so buying cheaper parts is right in our wheelhouse!

 

[LOL_Wut_Axel]

Can I quote you on this if the i7 2700K/i7 3820 gets annihilated by the FX-8150/FX-8120 upon release?



Remember I only use benchmarks that are open-sourced and compiled by an open compiler or by a microsoft compiler

But, I'm not buying Zambezi, I'm waiting for Vishera(If that is the name)

That won't happen, even though you want it to. AMD prices their CPUs based on their performance relative to their competition. If it performs around the same speed as the Core i7-2600K, then they'd be pricing anywhere from $310 to $330 depending on if it beats it slightly or is slightly slower. AMD uses multi-threaded benchmarks to compare overall performance, too, which is not always the best choice (the 1090T was somewhat faster than the i7-860 in multi-threaded and slower in single-threaded, so they sold it at $10 more). Since the CPU will be priced at less than $275, then it's because AMD knows it can't keep up with the 2600K. They'll have to fight it out with the i5-2500K and i5-2400.

Also, given how much AMD is taking to release new CPUs that compete with Intel, you'll be having to wait a long time.

And that's the story, repeated again. Bulldozer will have comparable or a tiny bit higher multi-threaded performance, but will be much slower in single-threaded. Same as Lynnfield vs Thuban.

 

[LOL_Wut_Axel]

It's cheaper to make than Thuban and It's small as well(Higher yields)

346mm^2~ x 2 = 12 core
vs
315mm^2~ x 2 = 16 core

:hmm:
So, what you getting at mister!


$1123.42 6172


$849.02 6272

(US$ 1123.42) - (US$ 849.02) = 274.4 U.S. dollars

Seems just fine for a server part aiming for market share that has higher yields and is cheaper to make
​

It may have good yields, but it's not gonna do as good as AMD's extremely mature 45nm process.

 

[frozentundra123456]

What?

Maybe I wasnt clear on what I meant. I was referring to the CPUs on the market now. It is not just an opinion, that for what is on the market now, AMD is clearly inferior in absolute CPU performance, except maybe in a very few highly selective benchmarks. This is a proven fact, which I dont see how anyone can dispute.

I was not referring to Bulldozer. I was only referring to CPUs for which we have reliable benchmarks. My personal opinion is that Bulldozer will be competitive with Intel's last generation chips at best, but that is only my opinion.

 

[sm625]

AMD prices their CPU based on their performance relative to their competition. If it performs around the same speed as the Core i7-2600K, then they'd be pricing anywhere from $310 to $330 depending on if it beats it slightly or is slightly slower.

There is a price chart from 2005 at http://techreport.com/articles.x/8616/1

It shows P4 660 at $605. And a P4 650 at $401. Yet AMD had their 4000+ single core selling for $375. Even though it ker-stomped the $605 P4 660, beating it by 25% in gaming FPS. Hell even the $235 Athlon 3500+ beat the P4 660 by at least 5% in gaming FPS. 6 core BD could outperform a 2500k by 25% in gaming and still be priced lower. That would be a hell of a feat but all I'm saying is that is what AMD does. They've done it many times in the past.

 

[chihlidog]

Are you saying enthusiasts can't be budget conscious? I would say that as enthusiasts we upgrade the most frequently, so buying cheaper parts is right in our wheelhouse!

No, I'm saying it doesnt look like BD is going to be much faster per core, if at all, than PHII, or give us a compelling reason to choose them over intel. 8 cores sounds awesome, but as has been stated to death, so much of what a PC is used for doesnt utilize that many cores/threads.

Some of the more reliable leaked info on BD makes it pretty clear that for most tasks, especially gaming, a $190 i5-2400 is a much better enthusiast part than an 8 core BD.

 

[LOL_Wut_Axel]

There is a price chart from 2005 at http://techreport.com/articles.x/8616/1

It shows P4 660 at $605. And a P4 650 at $401. Yet AMD had their 4000+ single core selling for $375. Even though it ker-stomped the $605 P4 660, beating it by 25% in gaming FPS. Hell even the $235 Athlon 3500+ beat the P4 660 by at least 5% in gaming FPS. 6 core BD could outperform a 2500k by 25% in gaming and still be priced lower. That would be a hell of a feat but all I'm saying is that is what AMD does. They've done it many times in the past.

No, and no. AMD wants high ASP, but Bulldozer has lower IPC. It's not gonna be 25% faster in gaming, so get it out of your head. In fact, it'll be the other way around. Games in general don't care for Six-Core processors and would rather have a CPU with two-four strong cores, and Bulldozer has weak cores. Also, arguing over gaming performance even in 2005 by running games at resolutions people would've never played at like 640x480 and 800x600 is worthless. 1024x768 was the lower-end at the time, with 1280x1024 being mainstream. We want to know gaming performance at normal resolutions, which would right now mean 1680x1050, 1920x1080, and 1920x1200.

And no, it's not what AMD does. AMD needs high ASP, and the best way to get that is by pricing your CPUs the highest relative to their nearest competition. Price wars are something AMD tries to avoid at all costs, since for them it's much more devastating than Intel. In 2005 the ASPs were high enough that AMD could undercut Intel a bit in price, but now they simply cannot (and even then, Intel was still competitive overall). The i5-2400 is only $190.

 

[sm625]

I'm not saying what will happen because I wont make baseless assumptions. I am just saying what can happen, based on historical precedent. That precedent says that a 6 core BD could outperform a 2500k by 25% in gaming and still be priced lower. It also means we cannot infer anything based on ASP or MSRP. Prices change. This is 2011, the new normal, not 2005 which was the peak of a bubble. Neither company is going to get prices back to where they were. The fact that AMD thinks they can get ~$300 for a chip is impressive.

 

[RussianSensation]

Can I quote you on this if the i7 2700K/i7 3820 gets annihilated by the FX-8150/FX-8120 upon release?

Intel is now rumoured to release a $340-350 2700k. Things are looking even worse. It appears that Intel may not discontinue the 2600k and replace it with the 2700k at the same price, but they are going to keep the 2600k and release a higher priced 2700k with only 100mhz clock speed increase.

Besides the overclocking world record, not a single positive leak of Bulldozer has materialized in the last 9 months. All we are hearing is that BD has slow IPC and it needs 8 cores to compete with Intel's 4. One of these companies is completely out to lunch. Either Intel is vastly underestimating its adversary, or AMD is all smoke and mirrors with claims that its 8-core FX processor will bring the "FX legend" back.

The only logical explanation I can think of is that AMD is going "all-in" on multi-threaded performance: content creation user superiority in video encoding, rendering apps, encryption, etc. Otherwise, if we are to believe that its single core is as good as a single SB core, they'll deliver a fast processor in 1-4 threaded apps and an unbeatable processor in 6-8 threaded apps - for $100 less than Intel top 2700k chip? That sounds like something I'd ask from Santa!

6 core BD could outperform a 2500k by 25% in gaming and still be priced lower.

Had a good lunch today?

2 core in BD design gets at least a 10%-20% penalty vs. 2 full fledged cores.
SB is at least 40% faster per clock than Phenom II is.
For BD to be 25% faster than SB, it would need to have an IPC increase of:

0.85 (mid-point module penalty) x 1.18% to normalize to 2 full fledged cores
--> 1.0 IPC Base clock for Phenom II x 1.4x to normalize for SB advantage
--> x 1.25x to beat SB by 25%
=====================
1.18*1.4*1.25 = 2.07x IPC increase over Phenom II to beat SB by 25%!

 

[The J]

Has this BLT website been reliable with pricing info during past releases? It seems that their pricing info has been taken as absolute fact on multple forums, and I can't help but wonder how wise that is.

 

[Idontcare]

Has this BLT website been reliable with pricing info during past releases? It seems that their pricing info has been taken as absolute fact on multple forums, and I can't help but wonder how wise that is.

:hmm: about as wise as seeking information on pre-release hardware from a troll-laden forum in the firstplace? :|

We all get what we pay for here, the info is free and it's worth every penny

 

[LOL_Wut_Axel]

I'm not saying what will happen because I wont make baseless assumptions. I am just saying what can happen, based on historical precedent. That precedent says that a 6 core BD could outperform a 2500k by 25% in gaming and still be priced lower. It also means we cannot infer anything based on ASP or MSRP. Prices change. This is 2011, the new normal, not 2005 which was the peak of a bubble. Neither company is going to get prices back to where they were. The fact that AMD thinks they can get ~$300 for a chip is impressive.

They don't. The flagship will be $250-275. It's cheaper than Intel by a good amount, and for obvious reasons: it's slower overall. If they could price it at $300-320, they would. High margins is what you always want to be looking at, but if the other product is better you need to lower prices.

RussianSensation:

That new 2700K is based on newer samples and will probably overclock 200MHz or so higher at the same voltage, which is only good for Intel.

The fact that some AMD fanboys think Bulldozer will have similar IPC to Sandy Bridge and therefore they'll be giving you a CPU that's faster than the Core i7-3930K for less than half what it costs is just laughable.

 

[RussianSensation]

Has this BLT website been reliable with pricing info during past releases? It seems that their pricing info has been taken as absolute fact on multple forums, and I can't help but wonder how wise that is.

Even if we ignore all the pricing rumors, if BD is so competitive, why was it delayed so many times, why did it require so many revisions to get higher frequencies, why is it 9 months late since SB arrived, why hasn't there been a single positive leak from anyone?

 

[LOL_Wut_Axel]

Even if we ignore all the pricing rumors, if BD is so competitive, why was it delayed so many times, why did it require so many revisions to get higher frequencies, why is it 9 months late since SB arrived, why hasn't there been a single positive leak from anyone?

Why does it need eight cores when the price comparable Intel CPU has four, etc.

 

[Crap Daddy]

I think that by this time, even with nothing official, everybody who uses it's brains understands that it's impossible for the AMD top dog 8150 to surpass Intel's 2600K. The price says it all. It might be able to compete in some multithreaded tasks but overall it will be under the 2600k. I'm afraid that in gaming this FX series will be surpassed even by the 2500K.

 

[BlueBlazer]

Has this BLT website been reliable with pricing info during past releases? It seems that their pricing info has been taken as absolute fact on multple forums, and I can't help but wonder how wise that is.

Its not to say that the pricing is reliable but any bit of information brings more clues and answers. I was also referring to the price of Opteron 6272 being lower than Opteron 6176 SE earlier from another site. :hmm:

It may have good yields, but it's not gonna do as good as AMD's extremely mature 45nm process.

Furthermore there's the need to recoup the costs of development. Not forgetting that the Interlagos has been touted (by Thomas Siefert) to deliver "up to 35%" performance improvements.

Even if we ignore all the pricing rumors, if BD is so competitive, why was it delayed so many times, why did it require so many revisions to get higher frequencies, why is it 9 months late since SB arrived, why hasn't there been a single positive leak from anyone?

Most of those "positive leaks" has been either debunked or found/proven to be fan-made material. Quite a number of sites fell for them. :hmm:

 

[LOL_Wut_Axel]

Its not to say that the pricing is reliable but any bit of information brings more clues and answers. I was also referring to the price of Opteron 6272 being lower than Opteron 6176 SE earlier from another site. :hmm:

Furthermore there's the need to recoup the costs of development. Not forgetting that the Interlagos has been touted (by Thomas Siefert) to be "up to 35%" better performance.

Most of those "positive leaks" has been either debunked or found/proven to be fan-made material. Quite a number of sites fell for them. :hmm:

But then again, it also has 33% more cores than Magny Cours, and HPC has great scaling as you add more cores.

 

[BlueBlazer]

But then again, it also has 33% more cores than Magny Cours, and HPC has great scaling as you add more cores.

That's the oddity. More "cores", more performance(?) and newer 32nm generation selling for less. Undercutting oneself seems illogical. :hmm:

 

[GaiaHunter]

But then again, it also has 33% more cores than Magny Cours, and HPC has great scaling as you add more cores.

But those cores have the module penalty, right?

So those 16 cores, at 80% are only 12.8 cores - so those 35% would have to come from IPC and clock speed.

But since the clock speed is the same either the IPC is way higher or the performance won't be 35% up.

 

[sm625]

Even if we ignore all the pricing rumors, if BD is so competitive, why was it delayed so many times, why did it require so many revisions to get higher frequencies, why is it 9 months late since SB arrived, why hasn't there been a single positive leak from anyone?

You could say the same thing about the Apple A6. How many transistors does it have? How much does it cost to make? We know nothing about it, yet there must be 10 million of those chips somewhere.