I'm assuming that the hybrid Crossfire will use the memory on the discrete chip instead of the main system memory. At that point, if anyone is looking for a Llano system that can also be used for gaming, they can just pair it with a low-end or mid-range card with a sizable amount of GDDR5. Those who aren't can forgo the discrete GPU and still get decent graphics.
I'm assuming that the hybrid Crossfire will use the memory on the discrete chip instead of the main system memory. At that point, if anyone is looking for a Llano system that can also be used for gaming, they can just pair it with a low-end or mid-range card with a sizable amount of GDDR5. Those who aren't can forgo the discrete GPU and still get decent graphics.
Is AMD going to sell these as stand-alone chips, or are they going the route of laptops and soldered-in motherboards only?
I believe they will be stand alone chips... however you ll need a new motherboard reguardless because they ll be made for a new socket type.
The Llano APU are suppose to have their own socket. Socket FM1 for the current Llano APU notebooks. The Llano APU for desktop should be using a socket as well but I suppose it won't be using the AM3+ or similar sockets due to the on die GPU.
formulav8
Diamond Member
- Sep 18, 2000
- 7,004
- 523
- 126
Not clear cut, I think the doubling of L2 cache will make it very close in certain area's compared to phII. Look at Athlon II dual cores compared to dual core phII with the L3 cache, they come quite close in a few area's because of having the bigger L2 cache compared to phII.
Llano has 4 MB's of L2 cache compared to quad AthlonII/PhII having just 2 MB. Doubling the bigger-faster L2 will be more important in some area's than the huge-slow L3. So I wouldn't be surprised if Llano comes very close to phII in a few area's. I would somewhat think they would lower the L2 latency with this new, much improved 32nm process. Current L2 is 15 cycle cache, so they have room to make it perform even better.
Should know in a couple months I guess.
Jason
formulav8
Diamond Member
- Sep 18, 2000
- 7,004
- 523
- 126
To bad they didn't keep on pursuing that htx slot. Back before/around the AM2 release they were talking about the htx slot which would be given a direct link to the cpu (Kinda like a co-processor). They could have mapped that for the gpu and make it a dual channel memory bank or something. :thumbsup:
Actually FM1 is the Desktop socket. FS1 is the Laptop Socket. And FT1 is the pin configuration used for embedded (soldered) APU's which seem to be restricted to the Zacate line as even the E series Llano appears to use the FM1/FS1 sockets.
You're right. I referred to a chart that showed 4 sockets for Llano and I was wondering why FM1 has a IHS which is odd because I have not seen a laptop CPU with an IHS so far. The FS1 does not have a IHS so that means it should be with the notebooks.
From Toms:
A 5550 can do ~30 fps @1680x1050 low details, 4xAF (with a i5-2500k).
The Llano will probably be able to do something like that, so... will it play metro2033? yes in 1680x resolution and low details
That's my point is everyone complains about console ports and such, but the reality is that they have to. Consoles are actually much higher in performance then the lowest common denominator that people bring up. To many PC's are shipping with Intel GMA or AMD 6100 integrated parts. Even computers with Discrete graphic its usually an offshoot of the 6300 series from AMD. All of this is lower performing then the adapters available in the X360 and PS3.
We want games made for PC's then we need hardware in everyone else's machines to be reasonable. That way Consoles get the port and we can have our High Rez texutures and let the console ports deal with have lower poly models. But if the market for a playable game is only 500k-2 million, while the consoles have 2 million as a base of probable buyers and its into the 20-30 million from there, its where the development will lay. You have developers spending 30-100 million on a game, specially the ones that PC users want to play.
Llano forcing better development on the GMA from Intel, or Nvidia with no solution, pushing Discrete as hard as possible, will get the bottom end on desktops up high enough and eventually outpace consoles. It might be to late for the next gen, but I wouldn't be surprised if all three of them go with an off the shelf "APU" in the future. Bringing it all back to the PC end in the future. Well, it least its the only way it will happen otherwise games not console friendly (RTS's and other God View games) will die off and hardware will stagnate as the demands for PC hardware die off.
busydude
Diamond Member
- Feb 5, 2010
- 8,793
- 5
- 76
Hehe.. best answer evar.
Do you know the impact of overclocking CPU-NB for L3 less Athlons? Does it improve the performance at all?
formulav8
Diamond Member
- Sep 18, 2000
- 7,004
- 523
- 126
There are a few threads about that. But yes it helps in a few area's. My own testing showed about 14% lowering of memory latency and 14% increase in memory read/write throughput (from 1410 mhz to 2820, 2256mhz to 2820 mhz saw 6%) and it translated to real improvements. Gaming saw the biggest increases. I actually made a small article on one of my sites. (But I never link to it). I basically checked performance at 1410mhz, 2256mhz, and 2820 mhz northbridge speeds. I simply wanted to see the memory controller response (I didn't check typical stock 2000 mhz nb oced to 2800mhz nb because I was looking for a trend and was using very low speed ram anyways.). But as the nb clock went up, memory controller latency went down and thus, memory throughput went up. Anandtech also ran some nb oc benches on PII and there were good speed-ups as well increasing the l3 cache and mem-controller speed.
It would definitely be interesting to see how the gpu in Llano will respond to nb speeds since the gpu could use all of the bandwidth increases, and latency reductions it can get. :thumbsup:
I might buy it just for that experiment....
It would definitely be interesting to see how the gpu in Llano will respond to nb speeds since the gpu could use all of the bandwidth increases, and latency reductions it can get. :thumbsup:
I might buy it just for that experiment....
Here's two videos from AMD showcasing Llanos GPU with multitasking various graphical workloads (Starcraft II, Big Buck Bunny, a picture facial recognition program) and an OpenCL computation running some fluids/physics program bench simultanousley with SuperPi and WMP playing back Avatar probably at 1080p on a 1.8ghz/2.5ghz Turbo Llano vs a desktop 2600 Sandy Bridge.
http://www.youtube.com/watch?v=1YfRh1FBkI4&feature=player_embedded
http://www.youtube.com/watch?v=_H9PUgT4XpA&feature=player_embedded
http://www.youtube.com/watch?v=1YfRh1FBkI4&feature=player_embedded
http://www.youtube.com/watch?v=_H9PUgT4XpA&feature=player_embedded
Last edited:
podspi
Golden Member
- Jan 11, 2011
- 1,982
- 102
- 106
I have been saying over and over again we would be surprised at how high Llano clocks. The 'M' stands for mobile right? I can still not be wrong, right?
Hopefully we get some more model numbers and specs soon!
Phynaz
Lifer
- Mar 13, 2006
- 10,140
- 819
- 126
I guarantee the Intel system has been gimped. Most likely APU code is running only on the AMD system.
I guessed so too, it just screams that it runs on the x86 cores due to non functional or non shipping OpenCL implementation in Intels HD3000 drivers. We can wait although, in a year they may have a solution.
The first video that showed the A8-3510MX able to balance 3 workloads on its CPU and GPU but the load test is a little bit unrealistic for real life use.
The second video is impressive as SCII looks very playable even with additional workloads with it. Also a little bit unrealistic for real life use.
I guess the key feature of Llano is the capability to handle multiple workloads with ease. Can't wait to get my hands on a Llano APU notebook. :thumbsup:
I hate to call foul without more proof. For a certainty AMD has picked a test that puts them in a good light, even if its not a good representation of actual use. The Intel systems are already in the wild sp it shouldn't be terribly hard to reproduce (or disprove) the suggested results.
If AMD were that fast and loose with the truth, one would think that Intel would call them out on it.
podspi
Golden Member
- Jan 11, 2011
- 1,982
- 102
- 106
Intel has an CPU only OpenCL implementation and has done so according to Apples wishes imho. Check out the software portofolio of Apple and see if you can spot the apps that make use of the GPU with OpenCL for filters and various acceleration, Aperture is one of them.
Joseph F
Diamond Member
- Jul 12, 2010
- 3,522
- 2
- 0
Wikipedia says that the quad core models will have 4MB of L2 cache. This would make them a hell of a lot faster than Athlon II and possibly a good bit faster than Phenom II. AFAIK more L2 cache is more important than more L3 cache and therefor should make up for the lack of L3.
Yeah there have been many examples in the past (including PII and launch Athlon) that the more cache closer to the core is better.
L1>L2>L3>Memory. Problem is always in the die size, and impact on clock speeds. Easier to clock up L3 than L2, L2 over L1. So yeah my guess is 4MB L2 would be better per clock then having the L3, or at least keep them close since L3 didn't seem "that" important in most benchmarks.
Charlie D. has an interesting blurb, seems ARM joins the OpenCL bandwagon and makes an appearance in AMDs Fusion Developer Summit.
http://semiaccurate.com/2011/04/26/arm-jumps-in-to-opencl-with-both-cpu-and-gpus/
http://semiaccurate.com/2011/04/26/arm-jumps-in-to-opencl-with-both-cpu-and-gpus/
Does this mean there could be APUs in the future with ARM CPUs instead of x86?
ARM supporting OpenCL means now AMD/Nvidia/IMB/ARM all support it, top to bottom. So the future og GPGPU on OpenCL looks bright.
Heterogeneous computing from ARM+AMD anyone?
I get what GPGPU brings to the table, taking advantage of massive TLP that is not cost-effective to pursue with x86 cores, but what would ARM bring to the table in such a heterogenous processing environment?
Maybe some kind of x86 and ARM hybrid wherein the ARM side is used when the work load is low enough? Wild speculation i know
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 24K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
-
Facebook
Twitter