SB's have a lot of analogue in them which is why they are often on an older process. They are painful things relative to massive logic dominated processors.
Fusion die-shot - 2009 Analyst Day
- Thread starter AnarchX
- Start date
Here's a video from Fusion accelerating the IE9 quite dramatically:
http://www.pcgameshardware.de/aid,7...emo-eines-DX11-Spiels-auf-Llano-APU/CPU/News/
Typing this from a netbook, I am pretty sure, that kind of enhancement will make much more of a difference to the end user than AvP
I wish my netbooks GPU could do this with a new driver. *sigh*
http://www.pcgameshardware.de/aid,7...emo-eines-DX11-Spiels-auf-Llano-APU/CPU/News/
Typing this from a netbook, I am pretty sure, that kind of enhancement will make much more of a difference to the end user than AvP
I wish my netbooks GPU could do this with a new driver. *sigh*Hmm interesting. Maybe GPU isn't cut down all that much from Llano then?
The smartphones SoC's seem to do just fine with I/O on latest processes.
They are not dealing with SATA, PCI, PCI Express, etc., etc.
Well, the PCI is dead. But PCIe would prolly be necessary to provide.
I didn't know SATA speeds were also troublesome.
SATA runs at 3Gbps and 6Gbps.
The other problem is that getting enough pins on a single chip is challenging.
CPU needs power, ground, memory pins
I/O needs pins for USB, PCIe, GPIO, etc. etc. etc.
It all adds up, and pins on the die are expensive.
DK
I guess they'll implement something like Intel's FDI interface which is used to output video in the current H55 chipset. So that means a new southbridge(ie which is now effectively the northbridge) for both Llano and Ontario(hopefully with both SATA 6 Gbps and USB 3)
I really dont understand why they didnt plan Ontario at 32 nm from the start. Would have given them considerable die size, power and clock speed advantages. Maybe 32nm SOI was too expensive for this sort of chip right now :/
But anyway i'd speculate that Ontario probably has a graphics core on par with Cedar, though that would make the die size fairly large. Cedar is 59 mm2. According to AMD the cpu core will offer 90% of today's mainstream performance at less than half the die size. If we take mainstream to mean Athlon II class, it has a die size of 117 mm2 at 45 nm. Less than half and @40 nm would mean ~50 mm2. The memory controller and cache might be common between the CPU and GPU so there's some die area saved there. So i think somewhere around 90-100 mm2 could be possible. Jawed's estimate of 77 mm2 now throws a spanner in the works of my logic, maybe they've cut the graphics core even more
For reference, Pineview has a die size of 66mm2 on 45 nm(and this includes a single core atom core, memory controller and graphics. Now a dual core Pineview is 87mm2 so it means each atom core is 21mm2.
Jawed
Legend
I think I saw something about Ontario being built at TSMC, so could that be due to capacity? Or timing? 32nm at TSMC was cancelled.
Will AMD be offering Ontario as a synthesisable core for third parties to use how they want?
aaronspink
Veteran
Except they aren't. Most of the I/O/analog is sitting on separate chips. Most of the so called SoC solutions actually consist of 3-5 different packages.
Time to market? GPU is already on 40 nm and the CPU core is synthesizable, so TSMC 40 nm would be faster? Also, bulk wafers are cheaper and Ontario is targeted at lower-end.
I don't think so.
Well, 32 nm got canned and GF 28nm was sometime away, so I guess that's why the TSMC 40 nm.I guess they'll implement something like Intel's FDI interface which is used to output video in the current H55 chipset. So that means a new southbridge(ie which is now effectively the northbridge) for both Llano and Ontario(hopefully with both SATA 6 Gbps and USB 3)
I really dont understand why they didnt plan Ontario at 32 nm from the start. Would have given them considerable die size, power and clock speed advantages. Maybe 32nm SOI was too expensive for this sort of chip right now :/
But anyway i'd speculate that Ontario probably has a graphics core on par with Cedar, though that would make the die size fairly large. Cedar is 59 mm2. According to AMD the cpu core will offer 90% of today's mainstream performance at less than half the die size. If we take mainstream to mean Athlon II class, it has a die size of 117 mm2 at 45 nm. Less than half and @40 nm would mean ~50 mm2. The memory controller and cache might be common between the CPU and GPU so there's some die area saved there. So i think somewhere around 90-100 mm2 could be possible. Jawed's estimate of 77 mm2 now throws a spanner in the works of my logic, maybe they've cut the graphics core even more
For reference, Pineview has a die size of 66mm2 on 45 nm(and this includes a single core atom core, memory controller and graphics. Now a dual core Pineview is 87mm2 so it means each atom core is 21mm2.
IMHO, unless something radical happens, Ontario 2 should move to GF.
Yep Ontario is apparently on TSMC 40 nm process. GF's 32nm SOI process was available though, and the timing also fit since Ontario is scheduled for Q1 2011. Ease of design on a known process and cost reasons are probably why they chose TSMC 40nm. Though the benefits of 32nm would have accounted for some of those drawbacks. Ahh well im sure they had their reasons, we'll learn more in due time
AMD have stated that Ontario is synthesisable and easy to resuse but i dont think they'll be offering it to third parties.
Well that maybe the reason but 32nm is also on schedule for Q4 2010 and Ontario is slated to release in Q1 2011. It would have made it in time. Here is an interesting piece of info i came across, i have no idea if its true or not
Source is http://tech.icrontic.com/news/amd-samples-llano-to-vendors-tests-ontario/
As i stated there was the possibility of GF 32 nm, but Ontario 2 should most definitely be manufactured at GF and likely on the 28nm process
Couple of other questions i had are, any news on when Fusion 2 is expected/planned? ie a fusion product with a GPU+Bulldozer cores. And does AMD get preferential pricing on wafers at GF or do they pay the same rate as any other customer?
A 77mm² die that has two cores and an IGP (and both are no garbage performance) on 40nm process! That's freaking amazing if true
http://en.wikipedia.org/wiki/GlobalFoundries#Merger_with_Chartered_SemiconductorI think they get royal prices, seeing as they are the only customers.
Probably more like 25, as it's at most 2 cores. And then the 77mm2 total seems somewhat reasonable.
I think they get royal prices, seeing as they are the only customers.
Well they are the only major customers right now but they have a lot of potential customers lined up for 28nm. This is what is posted on their website. "GLOBALFOUNDRIES is driving the global standard for new technologies such as High K Metal Gate (HKMG) with seven world-leading semiconductor product companies including Freescale, IBM, Infineon, NEC, ST, Samsung, and Toshiba." And with the acquistion of chartered they probably inherited a few more customers
Maybe 25 mm2 per core, theres no way it can be 25 mm2 for both cores. Atom itself is 21mm2 for each core and bobcat is supposed to be far superior to Atom
