AMD Kaveri Docs Reference Quad-Channel Memory Interface, GDDR5 Option
by Anand Lal Shimpi on January 16, 2014 10:51 PM ESTOur own Ryan Smith pointed me at an excellent thread on Beyond3D where forum member yuri ran across a reference to additional memory controllers in AMD's recently released Kaveri APU. AMD's latest BIOS and Kernel Developer's Guide (BKDG 3.00) for Kaveri includes a reference to four DRAM controllers (DCT0 - 3) with only two in use (DCT0 and DCT3). The same guide also references a Gddr5Mode option for each DRAM controller.
Let me be very clear here: there's no chance that the recently launched Kaveri will be capable of GDDR5 or 4 x 64-bit memory operation (Socket-FM2+ pin-out alone would be an obvious limitation), but it's very possible that there were plans for one (or both) of those things in an AMD APU. Memory bandwidth can be a huge limit to scaling processor graphics performance, especially since the GPU has to share its limited bandwidth to main memory with a handful of CPU cores. Intel's workaround with Haswell was to pair it with 128MB of on-package eDRAM. AMD has typically shied away from more exotic solutions, leaving the launched Kaveri looking pretty normal on the memory bandwidth front.
In our Kaveri review, we asked the question whether or not any of you would be interested in a big Kaveri option with 12 - 20 CUs (768 - 1280 SPs) enabled, basically a high-end variant of the Xbox One or PS4 SoC. AMD would need a substantial increase in memory bandwidth to make such a thing feasible, but based on AMD's own docs it looks like that may not be too difficult to get.
There were rumors a while back of Kaveri using GDDR5 on a stick but it looks like nothing ever came of that. The options for a higher end Kaveri APU would have to be:
1) 256-bit wide DDR3 interface with standard DIMM slots, or
2) 256-bit wide GDDR5 interface with memory soldered down on the motherboard
I do wonder if AMD would consider the first option and tossing some high-speed memory on-die (similar to the Xbox One SoC).
All of this is an interesting academic exercise though, which brings me back to our original question from the Kaveri review. If you had the building blocks AMD has (Steamroller cores and GCN CUs) and the potential for a wider memory interface, would you try building a high-end APU for the desktop? If so, what would you build and why?
I know I'd be interested in a 2-module Steamroller core + 20 CUs with a 256-bit wide DDR3 interface, assuming AMD could stick some high-bandwidth memory on-die as well. More or less a high-end version of the Xbox One SoC. Such a thing would interest me but I'm not sure if anyone would buy it. Leave your thoughts in the comments below, I'm sure some important folks will get to read them :)
127 Comments
View All Comments
Schwebbz - Friday, January 17, 2014 - link
Yes. If Kaveri already has four memory controllers, it badly needs a chipset that can take advantage of that fact. And if quad-channel RAM can support even bigger SKUs without bottlenecking, AMD should go for it. I've never liked the idea of my computer needing extra cards for graphics.TechFanatic - Friday, January 17, 2014 - link
For the desktop I would go for a 4-module Excavator core + 14CUs, for a total of 8 Excavator CPU cores and 896 GCN GPU cores.The die size would sit between 300-330mm² (4EModules~ 170mm², 14 CUs ~115mm², other SOC logic~45/30mm²) roughly matching the die size of the FX 8350 at 315mm² on Globalfoundries' 20nm process node.
TDP would likely sit at 100W with 3.0-3.4Ghz clock speeds on the CPU and 720+ on the GPU.
Essentially double the power consumption of an A8 7600 but also more than double the performance.
Quad-channel DDR3/4 memory would have just about enough bandwidth to drive the GPU & CPU.
Also leveraging the powerful 896 GCN core integrated GPU (R7 260X class) in hybrid-crossfire with AMD's Hawaii GPUs (R9 290/290X) and all future AMD discrete graphics cards that will include a CrossfireXDMA engine would improve game performance so significantly that all the CPU IPC deficits the AMD CPUs have would be completely masked and game performance would be significantly higher than any equivalently priced Intel+GPU combination.
This sort of product at $229 would be hyper-competitive.
Markus2umasi - Friday, January 17, 2014 - link
Oh lord that chip would be a gamers dream.With HSA this could also become a very successful workstation and server chip.
I personally don't think fitting this in notebook form factors would be an issue, we already have 150W+ TDP gaming laptops in very thin and light configurations.
After all the thinnest laptop in the world the 14" Razer Blade uses a 37W+75W CPU+GPU combo.
WoodenMarker - Friday, January 17, 2014 - link
Ahh I loved the eight core chips, with the new 28nm SHP process AMD managed to keep 4 cores running at 3.1Ghz in a 45W TDP envelope they can theoretically give us 3ghz 8 cores under 90W right now, I would LOVE to get my hands on such a chip and they can fit the 8 cores in a die size smaller than Kaveri, so essentially giving you an 8 core Steamroller CPU in a sub 230mm die area for less than 150$.DROOOOOOOOOOOOL
You know what I would take a six core Excavator + 7790 out of a single processor any day, it doesn't have to even be 8 cores, HSA can handle any parallel workloads, AMD just needs to bring up the per core performance slightly.
TechFanatic - Friday, January 17, 2014 - link
You know what I actually agree, I think a 6 core CPU would be enough, also would help keep the die size nice and low for improved yields and lower prices.BushwayGopher - Friday, January 17, 2014 - link
What about cooling though. You'd need a pretty beefy cooler for a 300mm chip for laptops.Also I think your configuration would be very much possible once AMD introduces die stacking.
Put HBM memory on there, 4 serial compute units, a thousand GCN cores, asics, memory controller and you have a very nice package.
WoodenMarker - Friday, January 17, 2014 - link
I don't think cooling is going to be a problem, I mean look at AMD's 45 7600. If we essentially double the components and power you're going to get 90W on a 28nm process, 20nm will consume even less power.I think die stacking is still a ways off, likely 3 years ahead.
raja_jagadeesan - Friday, January 17, 2014 - link
I might vote for a 2-module Steamroller core + 20CUs with a DDR5 interface (the memory would likely have to be included). Basically a PS4+. Expandibility of course would be an issue with DDR5, but the performance would be great for the price, and would not need any high-speed memory on-die.But either way - DDR3 or DDR5, a large Kauveri could be *very* interesting; what a great price-performance box that could be as a gaming rig.
Runadumb - Friday, January 17, 2014 - link
Yes I would very much be interested in that. I built an HTPC using an A10 5800K.Its main purpose is an emulation station (I run hyperspin on it) but I've used it for suitable PC games too. Having a bit more GPU grunt would give me more gaming options (I don't want to use a discrete card).
AMD need to sort out their IPC though. Until they have a chip that can handle full speed Wii/PS2 emulation I won't be upgrading. No matter what the GPU gains.
yannigr - Friday, January 17, 2014 - link
With the current Kaveri you can start from A8-7600 and latter add a 240/250 in dual graphics mode. I am not considering 7700/7850k because they are expensive for that job, which is, making a low cost entry gaming PC with the possibility of easy upgrade with a cheap graphics card using dual graphics.If 7850K had 896 stream processors and 7700 had 768 on the other hand, things could be very different, assuming of course we didn't had that huge bandwidth limitations like now. In that case the price would be more than justified to start that entry level gaming PC buying just one of those processors and have the option latter to upgrade 7850K with 260X or 7700 with a card like 7770.
But as long as we have that limitation with memory bandwidth AMD can't do much more than concentrating on the cpu part. Giving a model with just 128 stream processors and 3 or 4 modules on the FM2+ socket with prices comparable to those of 63X0/83X0 would trigger the upgrade path for many people like me, who are still happy with their AM3(+) platforms.
So while Anand you would be interested in a chip with a huge GPU on it, me and I think most AMD customers who are with a Thuban, or with a 3 or 4 module AM3+, would be more interesting for an APU with more Steamroller modules.