Ontario 9w and 18w versions + pics - Page 45

Amiga500 · #**441** 09-14-2010, 07:37 AM

Quote:

Originally Posted by krumme

Until single module BD +gfx is ready, there is a very slim market for Llano.

I dunno.

I am revising my thinking on that.

A bit of hypothetical arguing:

1. If we assume Llano has graphics performance roughly equal to a 5650 & compute performance roughly equal to a Phenom II.

What percentage of the GPU market uses cards stronger than a 5650?
What percentage of the CPU market contains CPUs stronger than a 3.4 GHz PhII?

2. If we assume Llano has graphics performance roughly equal to a 5670 & compute performance roughly equal to a Phenom II + 10%.

Same two questions.

I'm coming round to the line of thought that the market that exists beyond Llano is (relatively) very small, and would be where enthusiasts do their encoding etc. In other words, the Llano market is GPU limited, always has been - and SB is weak in the area that actually matters.

Another way of looking at it would be:

What do the customers in that market ask?
(a) Which has the higher FPS in Dirt2/WoW/whatever
(b) Which can encode my media files quicker?

I can see a future where SB has very little traction in any area below the high end. Even there, BD is a threat on multi-threading.

Intel might be about to get Athlon'd.

reader · #**442** 09-14-2010, 07:39 AM

Quote:

Originally Posted by CRoland

I'm not sure what you mean by the last part. Are you trying to say that Onatario (9W?) will be slower than N550?

No the reverse in fact. I'm assuming the BOINC and other demos are of Zakate (some we know are). I'm saying even if the performance of Ontario is hugely reduced for lack of power I expect it to still beat N550 handsomely.

Amiga500 · #**443** 09-14-2010, 07:40 AM

Quote:

Originally Posted by James

This causes a problem for me, because at 225mm2, I'd expect more than 400 sp's. A lot more.

A quad core Athlon 2 is 169 mm2, a 5770 with 800 sp's is 170mm2. So that's 339mm2 on 45+40nm. With Llano being on 32nm, a 50% shrink would be...226mm2.

Unless the high speed bus is huge, I can't see Llano "only" having 400 sp's.

oooofffft.

**If** it were to have 800 SPs.... SB is confined to the upper market only.

stef · #**444** 09-14-2010, 07:42 AM

Quote:

Originally Posted by James

A quad core Athlon 2 is 169 mm2, a 5770 with 800 sp's is 170mm2. So that's 339mm2 on 45+40nm. With Llano being on 32nm, a 50% shrink would be...226mm2.

There's also additional 2 megs of L2 to take into consideration. It's not just shrinked Propus.

Beowulf · #**445** 09-14-2010, 08:05 AM

Quote:

Originally Posted by stef

There's also additional 2 megs of L2 to take into consideration. It's not just shrinked Propus.

Right, that usually does take up a lot of die-space.
Also, a 800sp part would definately be memory bandwidth starved.

reader · #**446** 09-14-2010, 08:09 AM

Quote:

Originally Posted by Amiga500

I dunno.

I am revising my thinking on that.

A bit of hypothetical arguing:

1. If we assume Llano has graphics performance roughly equal to a 5650 & compute performance roughly equal to a Phenom II.

What percentage of the GPU market uses cards stronger than a 5650?
What percentage of the CPU market contains CPUs stronger than a 3.4 GHz PhII?

Let me add a little to that. Apologies if this put the comment out of the scope of this thread.

The performance prediction about Llano is probably the wildest game in business at the moment I think. People are generally assuming Llano with a performance very similar to PhII which might very well be the case. But here are some fact.

Its a 32nm process and SOI+HKMG at that. We have seen PhII based 6 cores at 115W TDP and PhII reaching 3.2Ghz at 45nm. We all know what a disaster Ph was originally and how much AMD has improved it incrementally at every revision.

So what would be the increase in clocks due to the new process? Zero because they need the power for GPU or significant?

Clock gating, clock grid design, digital APM and other changes. Mostly they all might hugely change the power equation for Llano core. This might either go to GPU or a real good turbo. Following Charlie's article the clock grid by itself saves 15% power usage straight. The improved with clock buffer/gating whatever they are, I have zero clue to how much their impacts are. The digital APM also makes fine grained and faster turbo on/off possible I would assume which sure is good for a responsive system.

The L1 has been changed for good, L2s are double. It sounds as though their speed and higher clock potential are there. There is suppose to prefetcher improvement and TLB.

So here's the deal. Llano has almost nothing drastic. The clock gating has been done by Intel, the clock grid and digital APM might be a new thing. But as far as a super layman like me understand every single thing has been tweaked a little here and there to squeeze a little more. Understand that its a seriously old core, AMD engineer has worked on this for long years and should know it very well. This might all come down to nothing but I am waiting to be pleasantly surprised and the VP marketing's "Whoa" quote might actually have some substance to it.

James · #**447** 09-14-2010, 08:10 AM

Quote:

Originally Posted by stef

There's also additional 2 megs of L2 to take into consideration. It's not just shrinked Propus.

Maybe add 12-15mm2 for that.

You can sorta see that 400 sp's (assuming a 50% shrink that is), is not going to get this cpu near the 225mm2 suggested.

Maybe it's somewhere in-between, say 640?

Also looking at TDP values, with HKMG on top, it's clear a quad Athlon and 400 sp's is going to have a small TDP.

example - http://www.amd.com/us/products/noteb...870-specs.aspx

50 watts for the mobility 5870, which is slightly slower that the desktop 5770 (both have 800 sp's). If you clock it up to even 65 watts, 32nm and HKMG should easily knock that down a lot.

http://www.amd.com/us/products/noteb...770-specs.aspx

400 sp's comes in at 30w, albeit with lower clocks again but even if you say 40w that's 40nm, no SOI, no HKMG. That could easily be 15w we're talking about on GF's 32nm.....

I'm not saying it's true, but some stuff isn't adding up here. Why have we seen die shots of the cpu but not the gpu? There's been a lot of secrecy there.

Ratatosk · #**448** 09-14-2010, 08:19 AM

Quote:

Originally Posted by James

...

I'm not saying it's true, but some stuff isn't adding up here. Why have we seen die shots of the cpu but not the gpu? There's been a lot of secrecy there.

Redundancy for better yelds, maybe?