http://img697.imageshack.us/img697/6632/fusion.jpg
After Intel's "Sandy Bridge" leak, here is AMD's CPU/GPU mix up: four K10 cores w/ 1MB L2, no L3 and a side strapped GPU core.
More here.
Printable View
http://img697.imageshack.us/img697/6632/fusion.jpg
After Intel's "Sandy Bridge" leak, here is AMD's CPU/GPU mix up: four K10 cores w/ 1MB L2, no L3 and a side strapped GPU core.
More here.
sweet! now we need Hans to decipher building blocks and size :D
That looks to be the 2011 Llanos APU product.
presentation from Chekib Akrout, General Manager, Technology Group could be very interesting!
What's up with the constant "32nm is 2011 thing"?:confused:Quote:
Originally Posted by terrace215
It's been a known fact for over a year now...
BTW,thanks for sharing the pic and news fellix_bg. Interesting approach used by AMD,make a monolithic APU with "older" core but with a new gpu, on a new node.This could be a training of sorts for bulldozer parts that will follow soon after.
Nice shot.
That dark "thing" in the lower part must be the GPU/IGP, but looks so strange...
AMD+ATI chip starting to become real (long way until real product, I know).
what do you mean "older core"?Quote:
Originally Posted by informal
I wasn't aware of AMD quad-core CPU with 1MB L2 per core (and probably some more greenish squares on it :) )
I mean 10h and not bd/ontario ;). L2 increase is a bonus but not design change that meant huge investments in redesign(look at Regor vs Propus ;))Quote:
Originally Posted by Nedjo
Soooo in 2011, the bulldozer core is going to be 8 & 6 cores version, and the interlagos platform is going to be bulldozer MCMs?. Seems really promising if they can deliver!
Hope that bulldozer core kicks some serious ass!
Well, if you think about the potential benefits compared to the risks on the engineering side it makes sense. By using the existing core they should really only have to deal with shrinking it to the smaller node size and not worry so much about potentially changing specs, but more on the overall picture; verification should be easier at least.Quote:
Originally Posted by informal
Also, you have to remember the segment that it's targeting, ie. it's more of a mainstream part that doesn't need as high performance of a core.
What I'm more interested in is how much it shares in common with Ontario and the actual structure, ie. the shared memory controller/etc. Since Akrout went through the subject rather quickly, hopefully they go in depth during the APU breakout session :)
How many SP(ATI GPU) in the APU ?
@vietthanhpro: Hm, round 240 id say.
They are saying its 480.
http://www.xbitlabs.com/news/cpu/dis...rocessors.html
And Anand is saying AMD is saying it will be GigaFLOP APU :D
In other words HD4850 ALU performance .:up:
Quote:
Originally Posted by vietthanhpro
Quote:
Originally Posted by w0mbat
Quote:
Originally Posted by SimBy
for some reason I trust w0mbat most of all ;)Quote:
Originally Posted by Lightman
but could we move this conversation in to the "official" thread: http://www.xtremesystems.org/forums/...d.php?t=238702
i see 6 SIMDs on the die so thats 480 alu's. looks like they changed the layout of the SIMDs to make it fit nicely and they still have a lot of space left for moar cores.
Quote:
Originally Posted by Nedjo
this time it seems 480 is the correct figure.
LOLz....you better refer to source before you trust/believe in anyone
And now we can see the complete die shot of 32nm Llano core,not the cropped version :):
Complete slide
http://img22.imageshack.us/img22/491...rmsslide03.jpg
And re-sized:
http://img233.imageshack.us/img233/6...d2010small.jpg
640SP ? lol
Where did you find it?:)Quote:
Originally Posted by informal
Teh internetz is full of nice stuff :D.Quote:
Originally Posted by Oliverda
Actually i got a link to the slide via e-mail :). The person who sent it knows a thing or two :).
PS The thing at the bottom does look like a "bit" bigger GPU,or could also be northbridge-related for all we know.
According to PC Watch:
Attachment 98552
http://pc.watch.impress.co.jp/docs/c...12_328392.html
4770/4850/57x0 cards had mem bandwidth about 64Gb/s, 4670 had twice less.
How they gonna provide at least 32Gb/s to their IGP ?
@ Face
That picture of Llano die is actually not complete ;),look up a few posts.
The not-cropped picture shows much greater section dedicated to GPU,approx. ~480SP.Total die size est. 200+mm2,est. IMC is 192b wide supporting 1600Mz DDR3 memory.
So Llano is the second Fusion stage, combining SPs with X86 arch K10 cores. The final Fusion stage (full integration) will fusion (so appropiate) with Bulldozer arch? Or it will implicate a full new architecture? Talking about 2012 i supose...
quad core +Quote:
Originally Posted by informal
4750-4770(ish) gpu
and only ~200mm2, sounds pretty damn sweet
wonder what kind of clocks it can reach? and how much power it will take. im thinking it could do 3ghz core for about 50W and 900mhz on the gpu could be under 50W aswell. would be a killer package for nettop pcs
The APU will succeed for these reasons:
1) Software support already exists and is supported by major software vendors; Microsoft's DirectCompute and Khronos Group's OpenCL.
2) Not only AMD, but Intel are making APU products.
3) AVX works on AMD and Intel
AMD will simply be in an amazing position with their future products. Also new AMD APUs will come out yearly, instead of every 4 years (CPU) or 6 to 12 months (GPU) thanks to the type of design AMD will be using. The AMD haters will be silenced in 2011.
Thankx for the picture :^)Quote:
Originally Posted by informal
It was clear that it was cropped from the DRAM IO paths, so I had
already extended that. Also the other (left) side shows a truncated
HT IO module (It seems there are four half HT units)
I had expected a 64 bit GDDR5 "side buffer" below the GPU. I wouldn't
know where they would otherwise get the memory bandwidth from...
The IMC width is 128 bit . It's exactly like the regor one. Maybe it
runs faster but it seems more logical to have a side bus with a minimum
of two GDDR5 chips with could fit in the package itself providing much
more bandwidth as the standard IMC.
The foils say it has an high frequency memory interface in a such a
way that it strongly suggest that it's GDDR5 compatible.
Regards, Hans
Oh nice. The new OpenMM (OpenCL enabled) Folding@Home client will LOVE Llano! I wonder how the GPU cores will work with a discrete GPU installed too. Some form of Crossfire? Or better yet, have a Lucid Hydra managing the load balancing between the two :shocked:.Quote:
Originally Posted by SimBy
Either way, maybe we can expect some good PPD in 2011.
i was thinking they should offload bullet physics to itQuote:
Originally Posted by Mechromancer
I don't think lucid hydra can interface with a GPU integrated to the NB or CPU. Hydra is meant to be to be placed between the pcie slots and the pcie controller, and is designed accordingly. Moving the GPU to the wrong side of the system probably breaks compatibility with hydra.
I can't make sense out of this, so has anyone seen some info floating around
the net on this "news", if it is and not some kind of fake.
Original Link
Google Translation
Got the link from this post over at B3D Forums.
Quote:
Originally Posted by Kej
2010 H1 would be too early but I can imagine a 2010 Q4 date for it. The offical date is ~2011 but nowadays the AMD likes to repeat the "ahead of schedule" sentence in its presentations. :)
Maybe he was thinking the year is Jan 2010 and meant that the product will come in q1 2011 ... lolQuote:
Originally Posted by Kej
Or maybe his Chinese is messed up "if he gave the speech in Chinese"
Or maybe its true APC in 2010, but who am i kidding i want the bulldozer :D
But almost a year ahead :)Quote:
Originally Posted by Oliverda
Wow, well done indeed....
The tape out has done (theoretically), and the 32nm technology will be ready for production in 2010 H2. So we will see what happen.:)Quote:
Originally Posted by ajaidev
yeah AMD said that, but gigaflop performance doesn't mean HD4850 ALU performance, it would have at least a HD4850 performance if they said "it will be a Teraflop APU"Quote:
Originally Posted by Lightman
Current rumors say that it will have the same amount of crunching power (per clock) as a Radeon 46X0, although it will be based on the 5000 series architecture and it basically is a 56X0 with 320 SPs. Also note that I didn't say performance, but crunching power. I've no idea what kind of performance to expect. This is all just rumor so far anyway.Quote:
Originally Posted by Lokinhow
Yeah, I know... I've made late night mistake :hitself:Quote:
Originally Posted by Lokinhow
Anyway as Helmore says it should be HD46xx crunching power with HD5xxx or even 6xxx capabilities in that regard.
Quote:
Originally Posted by Lightman
Man, just stop and think about that!
If within two years we can buy a single CPU + Motherboards and have a complete system, that delivers incredible gaming performance TO ALL, for relatively cheap. Then this platform will take the industry by storm.
AMD Fusion 2nd gen... think dx11 system, AMD Black 955 performance + ATI Radeon HD5750 graphics by 2011? Or even close to that will be impressive. Who'd buy anything else? (relatively speaking, of coarse!)
Stuff that into a small box and Microsoft has the new 360!
all AMD needs for 1TFLOP is 400 Stream Processors@1250MHz... and that's absolutely achievable! ;) My guestimation would be 1300 MHz 'cos of appropriate multiplier - 6.5x for APU, 10.5x for NB, and let's say 15.5x for cores ;) and yeah Q4 2010 is in line with GloFo roadmap... but Computex 2010 demo does sound plausible... remember that first DX11 accelerator (and Ll(el)iano) will be DX11 APU was demoed at Computex!Quote:
Originally Posted by Lokinhow
umm dual channel DDR3 + on chip cache...Quote:
Originally Posted by netkas
...or two 32bit GDDR5 chip close to the socket. Freq and size can depend on the mobo manufacturer.Quote:
Originally Posted by xlink
Well this is from AMDs Analysts Day 2009...
Quote:
Journalist: Rick you mentioned that you will be sampling Bulldozer and Bobcat in 32nm, Fusion products in the first half of 2010. That would suggest that you either have taped out or is close to taping out all those products. Can you give us some sense off where you are in terms of the development on those cores?
Rick Bergman: We don't really give out inner milestones, but we, obviously you know how the semiconductor industry works. You know, June of 2010 is 7 months from now, it takes a few months to build these things so I don't think your comments are off-target at all we are, we're loading up the gun getting ready with our next wave of these fantastic products
Journalist: You mentioned, I think the roadmap for Global Foundries showed that you had, that they were doing 32nm in the second half of 2010. So how do you actually get those built in the first half?
Dirk Meyer: That was a production question.
Emilio Ghiraldi: That's manufacturing ramp, which is consistent with the...
Journalist: So they are doing the tape-outs now?
Rick Bergman: Oh yeah, absolutely and we have running material, we have material.
"The AMD executives said the company will begin shipping its Fusion line of integrated CPU-GPU products late in 2010. The integrated Fusion products will compete for market share with the discrete graphics processors, Meyer said. AMD is expected to ship Fusion first on a silicon-on-insulator (SOI) technology, and follow that with a bulk Fusion product."
source
^is that the 32nm Llano we saw with 400SPs? due late this year?
Yes it is. (with ~480SP;))Quote:
Originally Posted by Manicdan
I have a feeling that the evergreen refresh and fusion release will be together...
Any reason for that? Graphics are bulk, Fusion SOI completely...it usually has different cycles. Could be close, bot not exactly same time.Quote:
Originally Posted by ajaidev
The process is probably not that interesting from a marketing perspective.. To release it together with a new platform seems like a good idea to me.Quote:
Originally Posted by Behemot
You'd have all of their partners up in arms. These seem like hdtv/laptop/lowed laptop type deals to compete against the ion platforms for the next couple years. Maybe if these chips had cache like a 955 worth looking into. But for now I'm not sure why anyone is excited to be honest?Quote:
Originally Posted by Xoulz
Think about it is right, MSI, Gigabyte, XFX, ASUS, ect ect ect video card sales would drop. The motherboard industry for them would practically cease. If all you need to do is keep upgrading your chip w/gpu then occasionally upgrade your mobo you sure wouldn't need to buy a motherboard much. I think the only way AMD would make these guys happy if they change sockets every 2 years to keep them in some sort of business. Most of these companies will die off if this is the case.
Core clock at 1250Mhz for integrated gpu?Quote:
Originally Posted by Nedjo
You say that like it's nothing while the HD58xx series are topped at 850Mhz core clock.
This is AMD-lingo for "Products some time in Q1 2011", consistent with their recent Analyst Day slides, with all Llano products in early 2011.Quote:
Originally Posted by Oliverda
http://phx.corporate-ir.net/phoenix....rol-analystday
See Rick Bergan's presentation, pages labeled 27 & 28.
As Intel mentioned on their CC, the industry does not care for big Q4 product intros, they much prefer Q1. I have no doubt AMD will align to this, and launch in Jan if they are ready, later if they have 32nm hiccups.
what I really looking forward for Fusion is that I hope AMD could merge GPU to act/behave like FPU in CPU
Different process (32 nm SOI High K Metal gate vs 40 nm bulk), different manufacturer (GloFo vs TSMC), so why can't it happen the way Nedjo expected ? The clock might be on the upper range, and might not all that feasible from the TDP standpoint, but i'm quite optimistic regarding this new baby.Quote:
Originally Posted by bladerash
when can we expect this on an ultra thin notebook? i'm itching to pull the trigger on a dell vostro v13
Anyone have any TDP numbers?
I would imagine it would have to be around 25-35w...
Also, I thought it was either 240 or 480SPs.
Someone care to chime in? *wink* *wink* *nudge* *nudge*
Its 480SP's and the TDP most likely will be around 55w for the whole chip.
Well, a 55 w TDP quadcore fusion chip specced at 2.4 GHz CPU 800 MHz GPU WILL sound very nice to my ear, epecially if the chip can maintain its TDP's threshold even when loaded simultaneously with Prime 95 & Furmark. :D
It's an interesting question. ATI/AMD GPU has never manufactured with SOI or HKMG technology. Theoretically the SOI(+HKMG) is capable for higher frequencies than the simple bulk (but it's a more expensive process technology). Just compare the present day CPU and GPU frequencies.Quote:
Originally Posted by spursindonesia
I know that roadmap very well but recently the AMD likes to play the "ahead of schedule" game. :)Quote:
Originally Posted by terrace215
Intel is a different company. For example the first Phenom CPUs were launched on November 19th, 2007.
Thats kinda weird.. considering Q4 is where the holydays are.. that's when ppl spend the most. So why would they aim to launch it when the holydays are over?
If they launched it in Q4 then that would mean it will end up in store shelves in Q1 of the next year. Unless they launch it at the end of Q3 or beginning of Q4 then they may be able to get some products on store shelves for the holiday shopping season.Quote:
Originally Posted by bladerash
Usually what's launched in Q1 will end up in the laptops sold during the back to school season. You will see products featuring the new chips earlier than that, but then volume is a bit lower and there are still plenty of 'previous generation' chips in laptops at that time.
What?Quote:
Originally Posted by Helmore
Back to school season is on Q3.
You are making up timelines....
PHII launched in late Q4 2008 and made it to the stores in early december 2008 ;)Quote:
Originally Posted by Helmore
I can't completely deny this :p:Quote:
Originally Posted by v_rr
I was only talking about notebooks here. It's very common in notebooks that whatever gets launched in the Q2 and Q3 time frame does not end up in the back to school notebook lineup. That's just an observation I made though, nothing factual and I could thus very well be wrong.Quote:
Originally Posted by generics_user
Because it takes a looong time for the mobile systems to get finalized.Quote:
Originally Posted by bladerash
Since we are already on the topic of early silicon can someone tell me what is the die shot on the left?
http://pl.tinypic.com/r/2hezu5v/6
http://i48.tinypic.com/2hezu5v.jpg
Others are Deneb, Fusion and GPU's but I cant figure out the first one.
AMD Hammer aka K8Quote:
Originally Posted by Shadov
Yes, barcelona is the second one from left but first?Quote:
Originally Posted by generics_user
Thanks - now I see it! :)
It's a K8 single core CPUQuote:
Originally Posted by Shadov
http://gecko54000.free.fr/documentat...er_anatomy.jpg
edit: generics_user was faster :D
There's that pesky 8 core cpu again.
8 core?Quote:
Originally Posted by flippin_waffles
I see only a single core K8 and a quad core barcelona on the left side :confused:
That's not a quad core barcelona is it? I thought it looked like the die from that other thread. Aren't the quad core dies square not rectangular? Its pretty small, but that sure looks like that other die shot that was apparantly a photoshop job.
Edit:
Ahh the stretched version. Ok
i need fusion.......now! lol ugh i hate waiting for things.
Like this?:)Quote:
Originally Posted by Hans de Vries
http://prohardver.hu/dl/upc/2010-01/...600photo_p.jpg
It's a Radeon E4690 GPU (128-bit, 512MB GDDR3).
I remember when you had to add L2 daughter boards to your motherboard...
I wouldnt mind doing that again for extra AGU performance.. but say comes stock with GDDR3 and put GDDR5 in.. lol ;)
Was this with the original Celerons that didn't have L2 cache?Quote:
Originally Posted by STEvil
More like the 386/486/Pentium days.Quote:
Originally Posted by FlawleZWas this with the original Celerons that didn't have L2 cache?Quote:
Originally Posted by STEvil
--Matt
celeron was actually first to have on die L2 cache. it was to save money and ended up being really fast compared to pentium.
That wasn't true on the original Covington cellies, which were just Deschutes PIIs without their L2 cache. Only the Mendocino and later Celerons had it.Quote:
Originally Posted by Chumbucket843
--Matt
That's exactly what I was thinking. Celerons originally appeared minus L2 cache vs the PII's which did have L2. The performance was very poor without L2 cache and thus why it was later added.Quote:
Originally Posted by mattkosem
I think there were a couple P2/Xeon boards you could put daughter boards for L2/L3 in but mostly it was just 386/486/P1.
[OT]I have lot of 'em here, if you want a photo of my collection ;) Most PI boards are running :D[/OT]
PIC
The brown slot is for the cache sram extension.:)
looks like production starts in H2!
http://babelfish.yahoo.com/translate...rUrl=Translate
Chekib Akrout disclosed that AMD was developing with all one's strength first section of APU named LIano, started the quantity to produce in the second half of the year, and will go on the market officially in next year.Quote:
Originally Posted by god_43
no Llano till 2011? bahhhhh!!
Quote:
Originally Posted by Manicdan
Llano is not as revolutionary as BullDozer
It is just simply a dual core K10.5 with a GPU build in; no major micro architecture change
no thats right, gpu and cpu in one isnt revolutionary.Quote:
Originally Posted by haylui
1st of all AMD disclosed Llano as a quad core product. 2nd,AMD reworked some parts of Deneb design(as IMC,power management ).Quote:
Originally Posted by haylui
and it will be 32nm, and phenom 3 based parts (zoazma right?)
quaddy Llano? that's interestingQuote:
Originally Posted by informal
btw, hopefully AMD could get their power management really do the job;
rumours that Intel SB will be delay to Q1 2011, which give AMD some room to breath
Quote:
Originally Posted by god_43
I am looking forward to Zosma; want to upgrade both my AM2+ machine to 6 cores
currently I am running Athlon II x2 240 and Phenom II x3 720BE @ quad cores
vietthanhpro: a bit ot, but why is in your signature this fake screen Zosma?:confused:
just for fun i guess, but i really like the direct connect links, it makes MCM not suck!Quote:
Originally Posted by FlanK3r
whoa thia has a lot of info (i have never seen some of these slides before). more info comming monday...w00t!
http://translate.google.ca/translate...%26tbs%3Dqdr:d
Quote:
Originally Posted by god_43
Good read
the CTO gives a clear explanation over the comparison of Intel Clarkdale and AMD LIano.
Looking forward to AMD LIano for its fused CPU & GPU architecture
btw, god_43, are you a chinese?
heh no i have to say i am not, i understand why you are asking though, but i swear that every time i look for info on amd.....i get some Asian website...lol, damn Asians and their secrets! i think intel and amd like Asians better than Europeans or north Americans???Quote:
Originally Posted by haylui
:shocked:
ps. Google translation really is my savior ;p.
Quote:
Originally Posted by god_43
ic
most of the times, those mobo makers leak the news
don't forget, almost all mobo makers are in Taiwan.
this time, a little bit different, it is came from China
almost all mobo makers are taiwan basedQuote:
Originally Posted by haylui
but almost all of them make boards in china
p.s. i am a chinese
err i dont know about this one; but if they are right...fusion coming first Q or 2010???? they must mean 2011 right?
http://www.ibtimes.com.au/articles/2...processors.htm