• Some users have recently had their accounts hijacked. It seems that the now defunct EVGA forums might have compromised your password there and seems many are using the same PW here. We would suggest you UPDATE YOUR PASSWORD and TURN ON 2FA for your account here to further secure it. None of the compromised accounts had 2FA turned on.
    Once you have enabled 2FA, your account will be updated soon to show a badge, letting other members know that you use 2FA to protect your account. This should be beneficial for everyone that uses FSFT.

AMD conference notes: Llano pushed back, BD taped out, not sampling to customers

P

pxc

Extremely [H]
Joined
Oct 22, 2000
Messages
33,063
Yes, that's "not" and not "now" in the title. From past AMD server CPU launches, using the info in the last comment, there are usually 4-5 quarters between this stage and launch. That means BD will likely launch somewhere between Q3'11 and Q4'11. Hope you can wait. ;) I'm more disappoined in the Llano delay. I want to see how the integrated GPU performs in GPGPU apps. It could be a real game changer.

AMD CC (7/15/2010) highlights on launches:
---
Ontario (40nm, TSMC) will launch first, early 2011. It will start shipping in late Q4 (usual for an early January launch).

Llano (32nm SOI) will launch a quarter after Ontario.

BD has taped out, but has not yet sampled yet to customers. It's supposed to start sampling some time in H2 2010. Target launch date is in 2011, with no indication of when.
---
 
I suspect AMD is doing the right thing

This post is not about should or should not

1. AMD pushed hard for 6 to 12 cores Opterons, but reality says due to various reasons, it just take longer than expected.

2. AMD consumer desktop processors are OK. But the ASP is lower and even top bin chips garnered lesser income. You just have to admit lower-ASP if everyone is going to overclock low-cost Athlon II X4. Good for enduser, difficult for AMD bottom line.

So for the CPU-side, it is CRITICAL to respond to customer needs with possible sustainable revenue potential

1. there is no need to waste time on highest end performance mobile right now because consensus says they will stick to Core i-series

2. mid-to-low end is competing critically so pressue is also high. It is known that Intel offerring has higher price but the inertia is there and not necessary competing purely on price.

So we are down to sensible choice belows.

3. The ATOM market, big enough and a lot of potential customers wanting more but could not get products to fully address their next upgrade demand.

4. shift to GPU-focus computing ensures there's value in the ATI solution.

5. The light-power tablet segment (no sensible solution right now) WIDE OPEN! AMD has the benefits of x86 ecosystem, no-smartphone focus so fully fixed on delivering x86 tablet. Do not mistake light-power tablets with minimal-power consumption tablets.

6. For pure standard power-eating x86 tablets, you still face Core-i issues. However, since Intel will not address ATOM, an openning is available if Ontario with ATI graphic integrated on-die combination is designed to address ATOM market.

7. Segments that feel cannot be satisfied by alternative architectures tablets will naturally move here if products are available.
 
Last edited:
TSMC!!!!???!!!

Youve got to be fraking kidding me #1 TSMC has been notoriously unreliable with new stuff on their 40nm process. And #2 That 40nm process is so in demand right now you will be lucky to squeeze in a few waffers a day.

What happened? The talk was to use Gloflos 40nm fab not the crap at TSMC.
 
Zachstar said:
What happened? The talk was to use Gloflos 40nm fab not the crap at TSMC.
Click to expand...
There was one part of the CC I forgot to post above:

"32nm yields are not ramping as expected, which is why Llano is pushed back."

GF is falling further behind on 32nm. AMD went from a year behind with 45nm to GF now approx. 15-18 months behind with 32nm. TSMC isn't the problem here. TSMC is getting the parts out on schedule.
 
Boy, I hope GF is better at ramping 32nm than TSMC was at ramping 40nm when the 5000 GPUs came out.
 
MrC4 said:
Boy, I hope GF is better at ramping 32nm than TSMC was at ramping 40nm when the 5000 GPUs came out.
Click to expand...


GF is doing 32 and 28nm i believe but i know they have put a ton of money and time into prepping for 28nm for the last 2 years.. they are way ahead of anything TSMC is doing..
 
Last edited:
pxc said:
There was one part of the CC I forgot to post above:

"32nm yields are not ramping as expected, which is why Llano is pushed back."

GF is falling further behind on 32nm. AMD went from a year behind with 45nm to GF now approx. 15-18 months behind with 32nm. TSMC isn't the problem here. TSMC is getting the parts out on schedule.
Click to expand...

Gloflo 28nm is almost completely different from its 32nm fabs. For instance they talk about perfect wafer test patterns for 28nm and it going well but 32nm delayed. Different processes different goals. 40nm tho was supposed to be on track. What happened?
 
Zachstar said:
Gloflo 28nm is almost completely different from its 32nm fabs. For instance they talk about perfect wafer test patterns for 28nm and it going well but 32nm delayed. Different processes different goals. 40nm tho was supposed to be on track. What happened?
Click to expand...

Glofab dumped 40nm a long time ago.. said they were just going to stick with 28nm as their primary.. guess at the last minute they decided to add 32nm to their fab which is probably why its delayed..
 
Proof on 40nm dumping. Because I saw multiple articles talking about various customers using its 40nm process.

But assuming you are right Its doomed. TSMC is filled to the brim with orders already. ATI 5x series price drops have been delayed due to tightness in supply and then fermi and massive demand for the GF104 Ion Tesla, Heck there are hardly any Tegra 2s about and they were supposed to be able to fit a boatload of chips into one wafer.

So what is this going to be? Another Neo now? Will end up in some random netbook and no desktop motherboard combos?
 
Zachstar said:
Proof on 40nm dumping. Because I saw multiple articles talking about various customers using its 40nm process.

But assuming you are right Its doomed. TSMC is filled to the brim with orders already. ATI 5x series price drops have been delayed due to tightness in supply and then fermi and massive demand for the GF104 Ion Tesla, Heck there are hardly any Tegra 2s about and they were supposed to be able to fit a boatload of chips into one wafer.

So what is this going to be? Another Neo now? Will end up in some random netbook and no desktop motherboard combos?
Click to expand...


on the ATI side TSMC has solved pretty much all their 40nm issues and have been at 100% production for a while.. but with the ATI HD6k series coming the 5k series should hit the discontinued list pretty soon which means TSMC would move to the HD6k series production so i dont see much of a problem for TSMC making the transition.. TSMC has put a lot of money into 40nm fab/production and are moving directly to 22nm production from their last report a few months ago which pretty much gives away the fact that Glofab is taking over the 28nm production..

the lack of 5k series price drops has more to do with the lack of competition then the actual supply and demand.. ATI hasnt bothered to really drop the 5k series prices because there hasnt been any pricing competition.. but that will probably change now that the GTX 460 is out since its in direct competition with ATI's main stream market products.. if the rumors are correct that ATI is actually working on making a driver that supports triple DVI eyefinity running multiple cards then we will probably see a price drop on the 5770, 5830 and 5850 pretty soon.. but i have a feeling that wont happen until a month or so before the 5800 series refresh or 6000 series is planned to be released..

as far as glofab dropping 40nm there was an article early in the year talking about the future plans of Glofab's production.. when i sober up in the morning ill see if i can find it..

anything on the nvidia side of supply and demand from TSMC can be completely blamed on nvidia for not actually testing their shit before hand like ATI did with the 4770.. as much as anyone wants to say that nvidia's 210/220/240 were their test products, they were a completely failure and didnt actually prove anything nor help with the fermi developement or fab production since they were a complete waste and didnt have any fermi architecture in them.. they were just a die shrink of the 55nm process that was already used in the G200 chips which didnt even out perform the discontinued and out dated G92 GPU's..
 
#1 TSMC is not skipping 28nm And on top of that they are skipping 22nm and going to 20nm instead.

#2 Multiple recent articles talk about Gloflo 40nm all the talk is them skipping 32nm

As for 100 percent production that means different things. Take a look at this http://www.digitimes.com/news/a20100701PD207.html

Since TSMC is likely to give its major clients, Nvidia and Qualcomm, supply priority, the sources believe AMD may not be able to share much of TSMC's capacity in the second half of 2010.
Click to expand...

And I am QUITE sure they would rather have ATI 6x series and remaining 5x series cranking out than an extra wafer of bobcat. Am I the only one noticing just how messed up a situation this is? Where is Tegra 2? Already they said how fab shortages have caused massive delays in it. And bobcat is much bigger.
 
Hmm this doesn't look good for AMD.

They will only deliver notebook cpu that will be used for low cost notebooks (low cost multi core cpu and gpu with some performance probably makes makers of 400-500$ notebooks drool from enjoyment) but chances that it can compete with intel culv+nvidia optimus technology for low power subnotebooks is rather small and it won't be able to outmatch i3/i5/i7 for performance notebooks.

In desktop they are living in a borrowed time as the intel could make their cpus pointless if it weren't oh so comfortable situation for them to keep producing 45nm cpus using existing lines and selling them for 200+ $. And then there will be those few months when Intel has Sandy Bridge on market while Bulldozer is still unreleased.
 
Zachstar said:
#1 TSMC is not skipping 28nm And on top of that they are skipping 22nm and going to 20nm instead.

#2 Multiple recent articles talk about Gloflo 40nm all the talk is them skipping 32nm

As for 100 percent production that means different things. Take a look at this http://www.digitimes.com/news/a20100701PD207.html



And I am QUITE sure they would rather have ATI 6x series and remaining 5x series cranking out than an extra wafer of bobcat. Am I the only one noticing just how messed up a situation this is? Where is Tegra 2? Already they said how fab shortages have caused massive delays in it. And bobcat is much bigger.
Click to expand...

hmm.. ok thanks for the corrected info.
 
pxc said:
Yes, that's "not" and not "now" in the title. From past AMD server CPU launches, using the info in the last comment, there are usually 4-5 quarters between this stage and launch. That means BD will likely launch somewhere between Q3'11 and Q4'11. Hope you can wait. ;) I'm more disappoined in the Llano delay. I want to see how the integrated GPU performs in GPGPU apps. It could be a real game changer.

AMD CC (7/15/2010) highlights on launches:
---
Ontario (40nm, TSMC) will launch first, early 2011. It will start shipping in late Q4 (usual for an early January launch).

Llano (32nm SOI) will launch a quarter after Ontario.

BD has taped out, but has not yet sampled yet to customers. It's supposed to start sampling some time in H2 2010. Target launch date is in 2011, with no indication of when.
---
Click to expand...

Arghhh.........

This sucks as it means that my next folding rig will likely have to have yet another Intel processor in it. As much as I want to put an AMD in it, AMD just cannot compete on the high end for the few applications that actually need that kind of multithreaded performance. My gaming rig has an AMD, but that is because games don't really need that kind of processing power as they are still always GPU limited first.
 
intel pretty much ownz the roost right now, you can't compete with their clocks and ipc.
 
You cant compete with ATOM and Ion if you are reliant on TSMC either.

This is all so depressing. You cant just throw a product into a fab. It takes time to start the process so when It launches and all they sudden they happen to realize they can get but a few wafers of product a day due to 40nm demand. It will take months to get GloGlo producing them. Basically guaranteeing this will be a flop at the start. Likely no motherboards with it on newegg and OEMs will be very hesitant to use this if supply is not more guaranteed. And with Liano delays... Sigh...
 
unless AMD redoes it's arch to fix things they won't be able to compete with the IPC power of the intel chips. It's rediculous how efficient and powerful those chips from intel are now days. I'm rooting for the men in green so that we don't have a monopoly, and competition makes us consumers winners.
 
Except to get that efficiency the cost is absurd. 115 for a starter i3 90 fir a below starter pentium before mobo costs. Which are terrible themselves.

Also I do not consider their onboard graphics efficient. That may change with Sandy Bridge but for what little effect it has it sure takes up plenty of power.
 
Zachstar said:
Except to get that efficiency the cost is absurd. 115 for a starter i3 90 fir a below starter pentium before mobo costs. Which are terrible themselves.

Also I do not consider their onboard graphics efficient. That may change with Sandy Bridge but for what little effect it has it sure takes up plenty of power.
Click to expand...

yes, but for me the perf is worth the cost. I mean if I needed the perf I'd go intel, but for gaming I figured quadcore amd for cheap and that's what I did.
 
Michaelius said:
So amd will use 32nm for cpus and 28nm for gpus?
Click to expand...

AMD is lagging ... I still love them...

My friend who is a boy and I went to a tech demo and Intel was already belting out 22nm.
Wouldn't be surprised that 11nm would be the next with a jump over 16.
 
sushiserv said:
My friend who is a boy and I went to a tech demo and Intel was already belting out 22nm.
Click to expand...
Yeah, Intel showed a test wafer with each chip having 2.9 billion transistors, built on a 22nm process in September 2009. Supposedly those were working chips. Intel will be shipping 22nm CPUs within 9 months of AMD's first 32nm chips. Intel and Samsung already make 25nm flash memory.

One thing that wasn't really discussed during the CC is the state of GF's HKMG process. It's tied to GF's 32nm process for AMD CPUs. The problems ramping up yields may be related to the transition to HKMG. New core, new process, new manufacturing steps. Not exactly low risk, and exactly the kind of thing that has caused AMD huge manufacturing problems in the past. If they pull it off, they will be heroes. If not, well...
 
Intel is not going to be quick going to 22nm. What they have done lately is try to get more out of existing processes. AMD the same. They will make far more money by using 32nm as a mature process with new features and new layers than going to 22nm for a good but expensive performance increase but potential yield issues.

From what I understand at 20nm and lower test waffers will be far less able to determine issues in yield. And therefore it is going to start getting REAL diffucult to produce CPUs after 22nm. There is no rush.
 
Zachstar said:
Except to get that efficiency the cost is absurd. 115 for a starter i3 90 fir a below starter pentium before mobo costs. Which are terrible themselves.
Click to expand...


How is it absurd?

On the low end products you pay an extra $20 or $30. That is little, even to a person at the bottom of the economic ladder. Maybe if you lived in a country where the average income was $150 US a month. And this is for a product that can last several years.
 
It is not "a little" You can get quality AMD dual cores for 60-70 USD and Intel mobos lately are vastly more expensive than normal AMD counterparts.
 
Zachstar said:
It is not "a little" You can get quality AMD dual cores for 60-70 USD and Intel mobos lately are vastly more expensive than normal AMD counterparts.
Click to expand...
Yeah, but your example using the Core i3 is a processor that has the speed of an Athlon II X4, the i3 is a bit faster in gaming performance and a bit slower in encoding performance ($25 difference between retail i3 540 and retail X4 630). Intel makes much cheaper dual core processors too, which are performance competitive with the Athlon II X2 ($17 difference between retail X2 250 and retail E6500).

The motherboard difference isn't that much either, definitely not "vastly more expensive" either. What used to be a premium chipset, CrossfireX supporting P45 boards are around $80 now. Overclockable major brand boards for budget LGA775 systems are $45 on newegg, pretty close to the price of similar spec AM2 boards. Many LGA 1156 boards are available for $80 or less. An EVGA P55 CFX/SLI board is available for $70AR on newegg.

Going off newegg prices, I'm not seeing how a i3 540 + P55 SLI/CFX board for $200 is "vastly more expensive" than a X4 630 + SLI/CFX board for... $200. :p Or how a X2 250 + cheap board for $107 is vastly cheaper than an E6500 + cheap board for $124. It's a $17 difference.
 
Zachstar said:
It is not "a little" You can get quality AMD dual cores for 60-70 USD and Intel mobos lately are vastly more expensive than normal AMD counterparts.
Click to expand...


You can get quality Intel cpus for the same price (E5X00). I said a $20~30 difference, not "a little". You can call it tiny if you want though.

And Vastly? Are you a drama major?

There is a large selection of sub $100 motherboards and even a small selection of sub $50 motherboards. Seriously, even if you make minimum wage, the differences are very small.
 
Man, another delay. I've been looking forward to upgrading my rig but I just keep hearing of push-backs.
 
Comparing 775 to a modern AM3 chipset.. Gotcha... :p

I have to be careful with what I say less I draw out the Intel fanboys on this topic. Trying to use costs of a quad to justify their duo.. Ya ok..

32nm delays were expected. Late 2010 was too good to be true for a new process. And the 32nm CPUs are meant to be an OEM dream CPU with vastly superior onboard graphics but I doubt it is built for a power user in mind. The Bobcats are far more important and by going TSMC they are doomed to be unable to compete with ATOM.
 
I thought TSMC is only gonna make some of them, the "bulk" stuff? Rest of it will be either 32nm or eventually 22nm, particularly for the faster variants right?

Personally I'm pretty disappointed in AMD. They keep screwing up and they can ill afford to do that. Sandy Bridge is gonna come out late this year too, and supposedly with have an ~80% IPC single thread performance improvement... My WAG is when Bulldozer finally does launch it'll have a very hard time meeting much less beating that sort of performance. I hate to say it but there might be a real chance that AMD will go under in a few years unless they pull it together and stop making mistakes because I somehow doubt Intel is going to screw up this time and hand them the lead a la the K7 vs P4 days.
 
mesyn191 said:
I thought TSMC is only gonna make some of them, the "bulk" stuff?
Click to expand...
Yeah, like those low cost Ontario processors. In the past, AMD used Chartered to produce some, and later more, of its K8 chips.

Sandy Bridge is expected to have a 30% performance boost over Nehalem, according to Intel. AVX will bring very significant SIMD performance improvements. But like I wrote in the OP, the integrated IGP in Fusion processors (Llano) for OpenCL/DirectCompute performance could be very interesting and possibly a real game changer. Whether it has the rumored 360 or 480 SPs, that's still a lot of computing performance. Caveat: with memory bandwidth bottlenecks, some loads may not get much of an improvement. I wouldn't be surprised if Socket FM1 had pins for 2 x 32-bit DDR5 channels, or some other local memory interface.
 
I thought it was still using a PGA of some sort for the socket? Makes it unlikely that it'll have the pins for off package dedicated RAM for the GPU. Especially since GDDR5 is pin intensive and AMD is already pushing the limits of what is practical given the numbers of pins they're using now. e: For example, DDR2/3 dual channel uses something like 400 pins alone.

e: can't find the article anymore but its speculative anyways, 80% IPC improvement would be insane, probably special case only at best. 30% over i7 would still probably be tough for AMD to meet or beat though. Fusion could be a game changer...if the software is there to make use of it. GPGPU development is still lagging heavily, its your typical chicken/egg scenario. AMD really should've been pushing more on the front a year ago if they wanted to have a whole software ecology ready for launch. As it is now even the stuff that responds stupidly well to massively parallel processors like GPUs (ie. video encoding) isn't really ready. Closest thing there is are distributed computing apps like Milkway, but few use them.

Either which way you look at it AMD can't afford to keep screwing the pooch much longer, they'll become irrelevant fast if Bulldozer fails to deliver the performance without further delays too.
 
Last edited:
mesyn191 said:
For example, DDR2/3 dual channel uses something like 400 pins alone.
Click to expand...
That's a 128-bit memory interface. ;) A single 32-bit GDDR5 chip has about 67 signal pins and the rest are power and grounds in the 170 BGA package. An on board graphics style memory controller for 1 or 2 channels would use much fewer than 400 pins.

Even a fast, single GDDR5 device would help: a 32-bit chip @ 4GT/s = 16GB/s bandwidth, or a single 5GT/s chip = 20GB/s bandwith. Not only that, but it would also improve the performance of the IGP when it is used HyperMemory style.
 
Zachstar said:
Comparing 775 to a modern AM3 chipset.. Gotcha... :p
Click to expand...


Yeah. Your view is clouded because you aren't considering that for the large majority of buyers (of these level of computers), they don't know the difference between 775 and AM3. Maybe 98% of them don't even know what a chipset is. And if they did, do you think that an AM3 board would make their Facebook run better or faster than if they used a S775 board? The prices on the cpus are almost the same drama boy. The only advantage I see to using AMD is when you want to work with integrated graphics. It won't save you a lot of money but it will save and its a lot nicer than Intels (solutions on S775 boards).
 
You must log in or register to reply here.
Back
Top