• We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.

Llano shipping

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Abwx said:
Your card gdr5 can be as fast it it wants, the GFX will communicate
with the CPU and thus with the MB RAM at only 7GB/s.
Click to expand...


Yes, but it generally doesn't have to go to system memory because it has much faster onboard memory. When we're talking about rendering scenes, there isn't much (if any) advantage to being able to communicate with system memory, provided you have a large and fast enough "cache".


Just because we're talking about bottlenecks doesn't mean workload doesn't matter, because different workloads will have different bottleknecks :biggrin:
 
podspi said:
Yes, but it generally doesn't have to go to system memory because it has much faster onboard memory. When we're talking about rendering scenes, there isn't much (if any) advantage to being able to communicate with system memory, provided you have a large and fast enough "cache".


Just because we're talking about bottlenecks doesn't mean workload doesn't matter, because different workloads will have different bottleknecks :biggrin:
Click to expand...

You re right at some degree, but just think about it :
the datas that must be computed by the GPU are first in the
main memory, then transfered to the gfx wich compute and
drive the display.
The gpu memory (gdr5+) is in fact just the GPU cache
memory that help hide the latencies that would otherwise
completely bottleneck the full logic.
I m not a specialist, but one can figure easily that
there s constant communication between the RAM and
the GFX, otherwise, why would games need a lot of main memory ?...
 
Great thread. Ya wait a bit for Ivy Bridge.

Also the Q9 will run fine, its not a mammoth diff with Core Q and a 2500. You would not tell a difference what CPU your using, cuz they're already so fast..... I wont be upgrading until 2017 maybe,, thx
 
Not really. Sure, that's a bit of a ways off, but if they have good availability, who cares? Sandy Bridge was delayed long enough that it will only have a couple of months head start against Llano. AMD will have a very compelling product for at least 7 or 8 months until IB systems start to ship in volume. Considering that AMD has historically sold a dozen or two laptops per year, I think they will see a lot of growth.

If AMD can keep up in idle battery life, there is no reason they can't compete against IB with this either. Most people don't need the extra performance that IB might offer, and the graphics part of Llano should still be competitive with whatever intel comes up with.
 
I don't know if the schedules have changed, but AMD should have Trinity out to compete against IB. The Stars cores will be replaced with Bulldozer cores and the graphics will be based on Barts or Cayman.

Intel will definitely have an edge with their smaller process and Bulldozer is still a wildcard at this point, so it's hard to definitively say that AMD will stand strong in the long term, but it's a definite possibility.
 
Abwx said:
You re right at some degree, but just think about it :
the datas that must be computed by the GPU are first in the
main memory, then transfered to the gfx wich compute and
drive the display.
The gpu memory (gdr5+) is in fact just the GPU cache
memory that help hide the latencies that would otherwise
completely bottleneck the full logic.
I m not a specialist, but one can figure easily that
there s constant communication between the RAM and
the GFX, otherwise, why would games need a lot of main memory ?...
Click to expand...

It's not - a full stand alone gpu stores all the textures and does most of it's work using the very fast gpu memory, it doesn't use any main memory at all. This can be orders of magnitude faster and blows away anything llano can hope for.

Where llano does better is vs an old integrated gpu (i.e. on motherboard) that didn't have enough discrete memory and used to *share* main memory with the cpu - ie. the real budget end of budget notebooks. In that case they could only share main memory via the 7GB/s link PCI gave, now it has access to the full speed. On the downside llano has no gpu memory at all as far as I know.

Hence when llano is compared too is an old integrated graphics on the motherboard solution.

It used to have: Some, but not enough very fast memory of it's own and some slow access to main memory.

It now has: No very fast memory of it's own but better access to main memory.

Net effect is llano should be about the same speed, it's just cheaper to make as you don't need a separate chip for the motherboard graphics. As I understand it gpu + cpu are both old tech too, so really it's nothing new. You can buy llano performance today with an equivalent AMD cpu + motherboard integrated graphics. Llano will just save a few $$$ as it does away with the need for a separate gpu chip, and a fair amount of power due to the smaller process it's made with.
 
zebrax2 said:
Can't they add a sideport that uses PCI-e lanes and GDDR5?
Click to expand...

In theory, they could add direct GDDR5 lanes, or they could plug a GDDR5 chip under the CPU heatspreader, or they could try and stack memory on the CPU, or they could do many other things to improve memory bandwidth. ATI even has direct experience on using a small (few MB) but very fast memory buffer as a replacement for proper framebuffer, from the XBOX360 `Xenos`cpu.

The problem is that all these would very considerably increase both cost and power, which is something they really do not want. I hope they have some tricks up their sleeve for the higher-end Llano's, but you can be pretty confident that at least the low end will be very seriously memory bandwidth limited.
 
Can't they add a sideport that uses PCI-e lanes and GDDR5?
Click to expand...
GPU memory bandwidth, is a issue.

a few solutions:
1) Sideport memory (either Embedded, or a extra socket, for GDDR5 ram to be put into)
2) Embedded DRam on chip. (like IMB does with its Power line, or Xbox360 ect ect)
3) Faster DDR system memory, more channels (tripple or quad).


DDR4 is right around the corner too... DDR4 in tripple channel or quad channel, could be enough for next generation APUs. I believe the Embedded DRam on Chip solution is what is gonna be adopted, AMD will ask IMB for a little help and volla... 16-32mb Embedded Dram will be added onchip for the APUs (which will be plently memory bandwidth for the next gen APUs).

I dont expect next gen top APUs to be 6970's in performance.


http://citavia.blog.de/2011/04/05/a...onstruction-work-at-globalfoundries-10934679/
So in single-threaded benchmarks (sorted to the left) Llano sometimes achieves higher scores than the 2.2 GHz Athlon II, sometimes comparable to a 2.4-2.5 GHz chip
Click to expand...
Lmao... 800mhz Llano with Turbo on, matching the 2.4-2.5ghz Athlon II, in single threaded Benchmarks.
 
Last edited:
Mopetar said:
I don't know if the schedules have changed, but AMD should have Trinity out to compete against IB. The Stars cores will be replaced with Bulldozer cores and the graphics will be based on Barts or Cayman.

Intel will definitely have an edge with their smaller process and Bulldozer is still a wildcard at this point, so it's hard to definitively say that AMD will stand strong in the long term, but it's a definite possibility.
Click to expand...

I have quite a doubt that Intel will ever be able to produce anything decent in the department of graphics.

I have trouble imagining Intel coming up with anything like Barts/Cayman within the next 5 years, and even if their CPU part is stronger it will probably make no difference as overall the chip will be much weaker.

Smaller process= good but I'd expect GF to have something new within 6 months at least --
 
Do we know what the estimated MSRPs are going to be for Llano?

Will the dual core Llano be cheaper than its presumed competitor, the Core i3 2100? Or is AMD going to try to say that because it has faster graphics, it is a "superior" CPU?
 
Having double the l2 cache will probably give some decent gains. 10% in some area's may be possible? Quad core will probably respond better to a bigger cache than amd's typical dualcores did in the past.

So it appears that llana will simply use system ram as their frame buffer. No sideport or nothing I guess?

I still think amd should have released llana as a dualcore chip using 45nm process. They have given Intel way to much r&d time and thus could have hit intel harder if they released it a year or so ago.

Also Intel will get better and better gpu's since they are using nvidia tech already. They're fusion chips already has alot of nvidia tech in them and will continue on that path...
 
formulav8 said:
Also Intel will get better and better gpu's since they are using nvidia tech already. They're fusion chips already has alot of nvidia tech in them and will continue on that path...
Click to expand...

...

What are you smoking? nV and Intel aren't even on speaking terms, let alone share R&D. The SNB GPU's don't have any nV tech, and there won't be any in future Intel processors.
 
I'm mostly curious about how well the hybrid crossfire will perform. It may be possible to to build a cheap, yet reasonable powerful gaming rig that will get decent 1680 x 1050 performance with high settings. I know a few friends who would like to get into PC gaming, but don't want to drop a lot of money to get a decent gaming rig.
 
The 6790 was born for Llano. 250$ for a 6790 and a quadcore Llano makes a good match, provided that Llano clocks well and it will since Turbo Core 2.0 can run on top of manual overclocking.
 
grimpr said:
The 6790 was born for Llano. 250$ for a 6790 and a quadcore Llano makes a good match, provided that Llano clocks well and it will since Turbo Core 2.0 can run on top of manual overclocking.
Click to expand...

Can TurboCORE 2.0 run on top of manual overclocking? We know that TurboCORE 1 can, but TurboCORE 1 is pretty simplistic -- it just lowers the multipliers for three cores, and raises the multipliers for three other ones. I was under the impression that TurboCORE 2.0 calculates how much power is actually being used at each moment (so different workloads will have different TurboCORE speeds, independent of thread count), and sets the turbo accordingly. Wouldn't that make TurboCORE speeds the ceiling for TurboCORE in general, and if you go over that you're on your own?
 
You must log in or register to reply here.

TRENDING THREADS

Back
Top