Fusion die-shot - 2009 Analyst Day
- Thread starter AnarchX
- Start date
I truly hope so. Vendor agnostic APIs are far better for the industry, IMHO.
UMA in SoCs is a headache even for a TBDR. While I don't think that the above is wrong, bandwidth could become a bottleneck for something as anal like multiple HD video streams for example. Overall I don't think though that these platforms will run into any particular problem. For those that have higher demands than such SoCs will deliver it might be better for them to make a bigger hw investment from the get go.
Blazkowicz
Legend
I love how you're not afraid of using "anal" and "video streams" so close in a sentence.
itsmydamnation
Veteran
"multiple HD video streams " 
"multiple HD video streams "
Those things we really need to do, watching multiple videos at the same time
There is one web comic strip about this:
http://www.hs.fi/fingerpori/1135250316182
Last edited by a moderator:
Silent_Buddha
Legend
Well it may become an issue with HTML 5 if they go with h.264 for encoding of video with video advertisements that are loaded. Sometimes multiples on a page. This wasn't an issue with flash when it was just SD video.
But it can become a problem if you are like me and forget to turn off flash (yet to hear if you can turn off things like this in HTML 5 which is why I hate it) and then open multiple windows with HD video playing like say Youtube or ESPN.
Regards,
SB
But it can become a problem if you are like me and forget to turn off flash (yet to hear if you can turn off things like this in HTML 5 which is why I hate it) and then open multiple windows with HD video playing like say Youtube or ESPN.
Regards,
SB
Uhhhm there's a reason why I labeled a multi-stream HD video scenario as 'anal' in the first place. I can't imagine why anyone would want multiple streams for mainstream use anyway, unless of course the same PC is being used in parallel by two (or even more) different users.
It was merely an example how bandwidth on a UMA (or alike platform) can become a bottleneck. Something older along that line: http://www.anandtech.com/tradeshows/showdoc.aspx?i=3494&p=7
It was merely an example how bandwidth on a UMA (or alike platform) can become a bottleneck. Something older along that line: http://www.anandtech.com/tradeshows/showdoc.aspx?i=3494&p=7
If you think that is anal, just try developing simple IP-TV via multicast...I don't even want to think about multiple HD streams at the same IP right now
(Info: Work for an ISP)
Email Sasha Grey, she's probably the world's leading expert on the merits of multiple streams for any use, be it mainstream or not
Maybe not the first gen of Fusion, but it isn't impossible for later iterations. It certainly fits well into the theme of massively integrating system components and would give better performance to boot.
Blazkowicz
Legend
the boring, easy way is waiting for DDR4, it's due for 2012 (and better interesting in 2013), from a quick check on wikipedia.
your idea is good but it might be easier and more profitable for AMD to just sell an additionnal GPU rather than incur platform and memory costs.
your idea is good but it might be easier and more profitable for AMD to just sell an additionnal GPU rather than incur platform and memory costs.
One benefit from Fusion might be to boost sales of a lagging product w/ this scheme (as AMD's CPUs lag behind Intel's) w/ the halo from a superior product (which are the ATI's GPUs) though it might work the other way around too w/ the CPU dragging down the GPU... and making the package less attractive. (It's unlikely though since the relative importance of the CPU is getting smaller for the average consumer).
Selling an additional GPU (or CPU depending on how you look at it) automatically is what they're trying to accomplish. I'd say they're trying to milk their superior GPUs using Fusion (so if it's working for them, they're effectively selling an additional AMD CPU w/ each GPU instead of having the customer go Intel), and a faster closed memory subsystem could really help them leverage the advantages of integration even more. They've certainly done this before w/ the xbox 360 which uses only GDDR3 for both the CPU and GPU and achieves better far CPU bandwidth relative to PCs.
There's just no need for commodity DIMMs on most consumer computers. If they can fit the entire system onto a graphics board sized card, any upgrade would just involve the card and would be much more dramatic than upgrading only one component. The total cost would probably be around that of a mid-to high end ranged graphics board which is perfectly reasonable for a basic consumer system.
Last edited by a moderator:
My guess, 2012. In laptop market, perf/W matters more than raw power, so that should help compensate for the lame cpu.
Before GPU + Bulldozer fusion we will see a GPU + Bobcat Fusion.
But the K10 core in Llano is not just the "same" core as 2008 Barcelona, it has many small improvements, the most important being changing L1 cache type which should allow much higher clock speeds while still keeping voltage down.
But yes, sandy bridge will also have many improvements, so it will definetely be faster. But Llano will have much better GPU, and Llano is not a high-end product anyway.
Yea thats true but AMD loses on Perf/W as well. Phenom cannot compete with Lynnfield or Bloomfield in perf/W. In terms of dual cores, right now the comparison is skewed as Clarkdale is on 32 nm compared to Athlon II/Phenom X2 on 45 nm but even Core 2 duo on 45 nm beat AMD easily in perf/W.
Anyway what is AMD's share in the notebook segment? Is it even 10%?
Bobcat seems to be a Atom competitor and not a mainstream laptop chip. http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=3674
Those small improvements might bring what a few % improvement? Clock speeds should increase though and AMD gets power gating as well so it should be a decent part. GPU wise im sure Sandy bridge will have a beefier GPU compared to Clarkdale but if AMD has in fact incorporated a Redwood class GPU In Llano then they should walk all over it