Monday, May 23rd 2011

AMD Bulldozer, Llano Pricing Surface

Here are the first figures made public of the market prices of AMD's upcoming two lines of desktop processors. AMD will approach the desktop PC market with two platforms, the A-Series "Llano" accelerated processing units (APUs), and the FX-series "Zambezi" processors (CPUs). APUs are functionally similar to Intel's Sandy Bridge processors, in having processor cores, a graphics processor, memory controller, and PCI-Express switch packed into a single piece of silicon. AMD is apparently relying on its powerful GPU architecture to make Llano a more wholesome product. Zambezi functionally resembles Intel Westmere/Bloomfield, in having a number of processing cores, a high-bandwidth memory controller, and a large cache packed into a single die, making up for a performance part.

By mid-June, AMD will launch the FX-Series with two a 4-core, a 6-core, and two 8-core parts. The series will be led by eight-core AMD FX-8130P priced at US $320, trailed by FX-8130 at US $290. The former probably is a "unlocked" part. Next up is the six-core FX-6110, priced at $240. Lastly there's the quad-core FX-4110, going for $220. You will notice that the price per core isn't as linear as it was in the previous generation.

Around the same time as the FX-Series, AMD will launch its A-Series APUs, based on the brand new FM1 socket and single-chip chipset. The series is capped off by A8-3550P, which is an unlocked quad-core part priced at $170. Its "locked" variant, the A8-3550, will be priced $20 less, at $150. The A8 sub-series consists of quad-core parts with 400 stream processors enabled in the iGPU. Next up is the unlocked A6-3450P quad-core priced at $130, its locked counterpart, the A6-3450, is priced at $110. With A6 sub-series, the iGPU has 320 stream processors. At the bottom of the pile are dual-core parts, A4-3350P priced at $80, and E2-3350 at $70. The E2 sub-series has 240 stream processors on the iGPU. All prices in 1000-unit tray quantities.

Source: DigiTimes
Add your own comment

80 Comments on AMD Bulldozer, Llano Pricing Surface

#1
inferKNOX
I know this isn't Monopoly, but if the performance is within acceptable margins, "I'm buying!":rockout:
Come on 8130P, don't you dare disappoint!!
Posted on Reply
#2
Widjaja
IF the initial pricing is at stated, this is good news.
I was expecting to see stupid crazy prices similar to the last FX series with the 939 sockets.
Posted on Reply
#3
rem82
Jonap_1st said:
can't wait to see these octo-cores running side by side with 2600k :toast:
Bulldozer module has:

one FETCH
one DECODE
one FPU
two Integer scheduler
one L2 Cache for module.
one L1 instruction cache


Same number of transistors with sandy 2600K

Yes , 8-core Bulldozer is a true 4-core chip with excellent HYPER TREADING technology !!! Not true 8-core !!

Bulldozer architecture is very elastic !! That is the power and secret for bulldozer ... 2x128bit FMAC or 1x256bit FMAC or 4x64bit !!!

Posted on Reply
#4
Imsochobo
rem82 said:
Bulldozer module has:

one FETCH
one DECODE
one FPU
two Integer scheduler
one L2 Cache for module.
one L1 instruction cache


Same number of transistors with sandy 2600K

Yes , 8-core Bulldozer is a true 4-core chip with excellent HYPER TRANSPORT technology !!! Not true 8-core !!

Bulldozer architecture is very elastic !! That is the power and secret for bulldozer ... 2x128bit FMAC or 1x256bit FMAC or 4x64bit !!!

http://img35.imageshack.us/img35/3954/filephpt.jpg
Hypertransport.. I think you confusing hyperthreading, its almost a true 8 core but its not almost hyperthreading alike, inbetween, great aproach anyhow.
Excited to see if it delivers, I don't want to replace all my servers in my datacenter, they are currently running amd cpu spec for clustering.
I use all my old gaming rigs for servers, and if the new is compatible with the old it'd be marvelous. :D

Like reasoning behind the Bulldozer core. Especially the fact that they share the FP pipe between two int pipes. It is also good to relieve most of the FP against many small cores in the future rather than running it in FAT-cores.

But again fetch decode can lead to greater latencies...

We all have to be patient and wait... and see..
there is alot amd can improve with this design in 2nd gen bulldozer when AM4 comes around. this design can probably do alot of magic!

less than a month!
Posted on Reply
#5
Jonap_1st
rem82 said:
Bulldozer module has:

one FETCH
one DECODE
one FPU
two Integer scheduler
one L2 Cache for module.
one L1 instruction cache


Same number of transistors with sandy 2600K

Yes , 8-core Bulldozer is a true 4-core chip with excellent HYPER TRANSPORT technology !!! Not true 8-core !!

Bulldozer architecture is very elastic !! That is the power and secret for bulldozer ... 2x128bit FMAC or 1x256bit FMAC or 4x64bit !!!

http://img35.imageshack.us/img35/3954/filephpt.jpg
calm down brother :toast:

i said octo-cores doesn't mean i really care about core counts. i'm just repeated the lines from what i've been read..

heck, since i already knew the prices. i'm sure i dont have a right anymore to complain about it :)
Posted on Reply
#6
rem82
Imsochobo said:
Hypertransport.. I think you confusing hyperthreading
Hyperthreading I wanted to write.
Posted on Reply
#7
mcloughj
if these chips had reverse hyperthreading they would have made mega gaming chips... oh well!
Posted on Reply
#8
Damn_Smooth
mcloughj said:
if these chips had reverse hyperthreading they would have made mega gaming chips... oh well!
What do you mean?
Posted on Reply
#9
Fourstaff
mcloughj said:
if these chips had reverse hyperthreading they would have made mega gaming chips... oh well!
I was discussing about "reverse hyperthreading" some time ago, long story short its not going to work. Uses too many cycles in moving the data around.

If people takes full advantage of the stream processors of Llano, will that cause Llano to take a massive lead in encoding and other massively parallel computing tasks compared to Bulldozer? I have a feeling that Bulldozer's days in consumer market are numbered even before launch due to Llano and Sandy Bridge.
Posted on Reply
#10
mcloughj
Damn_Smooth said:
What do you mean?
long story short: make a multicore chip act like a single core, with all the cores focused on the one task. Since most games were until recently not multi-core optimised, having a quad or octo core acting as a single core would be very fast for some apps.

But only theoretically as Fourstaff pointed out. :)
Posted on Reply
#11
heky
Anyone care to share the release date?
Posted on Reply
#12
Imsochobo
heky said:
Anyone care to share the release date?
~20th june

Midjune, mobo's come 19th-21th somewhere there. not long till ;)
Posted on Reply
#13
TheGrapist
Imsochobo said:
~20th june

Midjune, mobo's come 19th-21th somewhere there. not long till ;)
yay,hopefully i'll have all my stuff sold by then so i can afford it :o
Posted on Reply
#14
Steevo
Fourstaff said:
I was discussing about "reverse hyperthreading" some time ago, long story short its not going to work. Uses too many cycles in moving the data around.

If people takes full advantage of the stream processors of Llano, will that cause Llano to take a massive lead in encoding and other massively parallel computing tasks compared to Bulldozer? I have a feeling that Bulldozer's days in consumer market are numbered even before launch due to Llano and Sandy Bridge.
Look at how long it has taken us to get open CL, and yet so few real world applications use it. I don't think that X86 and the small amount of X64 computing is going away soon. It will take a large leap of whole operating systems supporting it, applications, and even consoles supporting it before it becomes mature.


It is only the leading edge of where this tech is going though, as we hit the limit of how small we can make a transistor and how many are required I believe it will take a large change in software and coding to take advantage of the next few major milestones in hardware progression. Memory, interface speed and fetching from datastores are the limiting factor. What good is 8 cores if we can't keep them fed data, what good are they when four have stalled? Higher clock speed will only help so much, and more cores does not directly relate to actual speed.


I truly hope AMD is working on the next, next tech and is using the CPU's they haven't sold to simulate the CPU and iron out the wrinkles.
Posted on Reply
#15
_JP_
Alright, so it's the FX-6110 for me.
Posted on Reply
#16
wahdangun
mcloughj said:
long story short: make a multicore chip act like a single core, with all the cores focused on the one task. Since most games were until recently not multi-core optimised, having a quad or octo core acting as a single core would be very fast for some apps.

But only theoretically as Fourstaff pointed out. :)
its impossible to do that, it will just the same as multi threading single application.

@ fourstaff: but SNB doesn't support openCL. So it will take a while
Posted on Reply
#17
bucketface
Fourstaff said:
I was discussing about "reverse hyperthreading" some time ago, long story short its not going to work. Uses too many cycles in moving the data around.
but isn't that with tradional cores not the highly integrated cores of bulldozer. I would have thought that it should be relativly easy to get a single threaded app to run on a "module" as if it were a single "core" utilising both integer clusers as a single "large core", similar to how the fpu can apparently function as a single 256b, 2 128b, or 4 64b.
ps.
it would be great if i could get an explanation as to why this is undesirable, if that is the case.
Posted on Reply
#18
largon
Wouldn't it be nice if there was a quadcore part that has all four modules enabled but without SMT, instead of two modules with SMT.
Posted on Reply
#19
faramir
Fourstaff said:
If people takes full advantage of the stream processors of Llano, will that cause Llano to take a massive lead in encoding and other massively parallel computing tasks compared to Bulldozer? I have a feeling that Bulldozer's days in consumer market are numbered even before launch due to Llano and Sandy Bridge.
You mean via OpenCL and similar interfaces ? I'm positive enthusiasts who decide to go the Bulldozer route are more than likely to pair their new CPU with AMD's graphics (something vastly more powerful than Llano's 400 shaders, of course) which is going to have this same parallel computing functionality but on a far larger scale.

Llano is about bringing adequate performance for money, conveniently tucked into a nice single package (well and the PCH ;) ). It isn't nearly as powerful nor flexible as the combination of Bulldozer + whatever GPU you want is going to be.
Posted on Reply
#20
Wile E
Power User
Strider said:
That's not exactly true. A thread is not a core, not even close. I will take cores over threads any day of the week, especially when it comes to what I do the most, and that's high-end gaming. At any rate, 8 real cores will beat a comparable 4 core 8 thread processor in most all cases.

Also, Bulldozers are multi-threaded anyway. Even though multi-threading (Hyper-threading in an Intel trademark) has been around a long time, since the late P4 days. There are still far more applications that do not support it than do.

So it's not something I put much stock in. heh
You must be missing what I am saying. The pricing seems to suggest that 8130 is going to be close in performance to the 2600. 8130 is an 8 core, the 2600 is 4 cores with HT. This tells me that Bulldozer has less performance per core than Intel, which, in turn, tells me AMD most likely won't have anything to compete with Intel when they release their new 6 and 8 cores.

Again, I hope I'm wrong. I want an all out war in the $1000 uber-cpu market.
Posted on Reply
#21
xenocide
Wile E said:
You must be missing what I am saying. The pricing seems to suggest that 8130 is going to be close in performance to the 2600. 8130 is an 8 core, the 2600 is 4 cores with HT. This tells me that Bulldozer has less performance per core than Intel, which, in turn, tells me AMD most likely won't have anything to compete with Intel when they release their new 6 and 8 cores.

Again, I hope I'm wrong. I want an all out war in the $1000 uber-cpu market.
Refer to rem82's post. The 2600k and 8130 are pretty damn similar on paper, so depending on how well the architecture works, they should be about equal in term's of price\performance.
Posted on Reply
#22
Wile E
Power User
xenocide said:
Refer to rem82's post. The 2600k and 8130 are pretty damn similar on paper, so depending on how well the architecture works, they should be about equal in term's of price\performance.
If that post is truth, then it does fair much better, but then how many modules can they put on a chip (and why call them true cores if they are not)? Intel plans 8 cores with hyper threading, iirc. Will there be 16 "core" Bulldozers?

But price/performance is not what I am concerned with. I am only concerned with performance per core per clock (and OC ability).

Guess we'll have to wait and see.
Posted on Reply
#23
[H]@RD5TUFF
Ocoto cores are over priced, should be $260-275, especially since they are not real octo cores and simply hexa cores pretending to be 8 cores. But it will all depend on performance, if they are anything less than faster than a 2600K, and not capable of 5ghz on air then AMD really hasn't succeeded, also there is something to be said when it takes 8"cores" to compete with a 4 core chip. :roll:
Posted on Reply
#24
a_ump
[H]@RD5TUFF said:
Ocoto cores are over priced, should be $260-275, especially since they are not real octo cores and simply hexa cores pretending to be 8 cores. But it will all depend on performance, if they are anything less than faster than a 2600K, and not capable of 5ghz on air then AMD really hasn't succeeded, also there is something to be said when it takes 8"cores" to compete with a 4 core chip. :roll:
TRUE, but there is also that old argument used back when C2Q came and the Phenom's were on their way. C2Q's weren't true quads, phenom's were, so AMD used that for marketing(till the flop :P). Now to the average joe that just looks at boxes comparing numbers, AMD could say "we prefer to use 8 cores, a core for each task so to speak, whereas Intel uses 4 core's to do those 8 tasks, which isn't as efficient, eh ehm."

either way, it can be twisted to benefit the other company. Also, is the integer unit the main component in a CPU or something? if it is, they could pull some crazy twisted saying like "each of our core's is like 75% of intel's, however if you add it up, we have 600% performance and intel only 400%, that's 200% more processing power". the retard lines i can pull outa me ash, yet to the general public it'd fly:banghead:
Posted on Reply
#25
Platibus
Two questions:

Is the 8130P the highest end model of Bulldozer? And, will the release of these APUs lower the prices of current CPUs, Phenom and Athlon?
Posted on Reply
Add your own comment