Our 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 :)

Comments Locked

127 Comments

View All Comments

  • losergamer04 - Friday, January 17, 2014 - link

    I see three markets for this:
    1) The ITX that was already mentioned. However, the market is small.
    2) Laptops that were mentioned but getting people to buy it would be hard and a custom chip for that small of a market is going to lose money.
    3) R&D. I think AMD has a real chance here. HSA and GDDR5 would offer a really high performance solution for researchers needing real-time processing.

    Combind all three of those markets and you might have enough market to sell the CPU and not lose money.
  • pattycake0147 - Friday, January 17, 2014 - link

    I fully agree with Anand here an APU with 20 CUs would be very welcome provided memory bandwidth wasn't an issue. Give me four channel DDR3 or see what DDR4 can bring as some others in the comments have said. I would love to ditch my discreet card (HD7850) for a single chip solution, but until that kind of performance is reached or surpassed I won't be spending up on a half baked solution.

    I got into the PC market during the Althon XP days and it pains me to see where AMD is today. I would love to justify a new AMD PC but as is stands I'll be sticking with my Intel CPU and AMD graphics card.
  • nfriedly - Friday, January 17, 2014 - link

    Heck yes I'd be interested in any of those options. I think I recall seeing someone show 5-15% performance bump on Kaveri by just switching from DDR3 1600 to DDR3 2133.

    I'm seriously considering at a Kaveri chip for a HTPC / light gaming machine, but I'm concerned that it will be a little too light for the games I play. I think more memory bandwidth would be enough to fix that.

    If the performance was about equal, I'd prefer a 256-bit interface to regular DDR over the other options, but I'd be happy with any of them.
  • Flunk - Friday, January 17, 2014 - link

    That high-end APU could be a good option for saving a few dollars on mid-tier gaming PCs. If it performs like a $200 CPU and a $200 GPU if they sold it for $300 I could see buying one. Kitted out with all the right components you could end up with a half-decent gaming PC or Steambox for $500-$600 which would be great for most things.

    Some of my less affluent friends could really use something like that and you don't need to lose out on upgradablity because they could easily include a PCI-E slot on the board. Even on an mITX form factor.
  • Borkmier - Friday, January 17, 2014 - link

    I would still prefer a 4 module part if possible, as at least in my scenarios 2 module parts choke when running more than 5 applications at once. Something in the range of 20-28 CU's seems about right for what you could hope to run off a 256bit DDR3 interface. If such a part existed for say $300.00 or less it would find its into many of the systems I sell to others at this point, not to mention a couple of my own.
  • davidpanz - Friday, January 17, 2014 - link

    I personally would love to have that option. AMD would have to go the Intel Iris pro route, and sell it on a pre-soldered motherboard. The difference would be they could sell to OEMs and the retailers, unlike Intel. I was actually hoping for a beefier gpu on Kaveri, I want a Project Spark box for building, I have a Xbox One for playing. AMD could have made and itx motherboard with both DDR3 slots and soldier GDDR5 on the back of the board and use the itx case as a heat sink. We have itx motherboards with mpcie slots on the back already. If you a wondering, HSA is enabled on/for Hawaii CGN gpus on pcie slots so it already works across different memory types/controllers, mixing both looks like what Kaveri had planned. Iris pro and Xbox one SoC already do, it's just another cache, GPU and CPU would both be taken care of.
  • Rocket321 - Friday, January 17, 2014 - link

    I'd say a high end APU as you mention would be the perfect for a mid-range steam box, making for a near perfect "console" competitor.

    Personally I'd want one for a steam + HTPC box in the living room, and I'd love one inside a premium laptop with a 1080p+ LCD.
  • eanazag - Friday, January 17, 2014 - link

    I think those extra mem controllers come into play on the server variant. Or a future revision. As to a high speed variant, I would be interested in a dual processor system with more CPU cores and GPU cores with configurable TDP for each side of the equation (CPU\GPU). It would be great if turbo would automatically figure out which set of cores would be better served with more TDP headroom on the fly, so applications that are processor intensive get the extra juice that they need. 8-12 CPU cores. I had commented on AMD releasing and 200W TDP processor and they did with the FX processor, but I was thinking of a 200W TDP APU, because that would make more sense as in a case you would just skip discrete cards. Those higher GPU cores only make sense if you can have CPU that push the FPS in games. If the CPU is too weak, than that GPU goes to waste. This was something that was apparent in a laptop review that was done here on Anandtech. It was a review with the 8970.

    I'd like to see a system that can run a 4K monitor at 30 FPS with settings at medium to high. I think that is realistically possible given AMDs situation. That could support 64GB of RAM. Throw in the GPU virtualization that works in VDI solutions like VMware and Citrix and I would be really interested.
  • willis936 - Friday, January 17, 2014 - link

    While a one chip solution is elegant and powerful by some metrics I think now more than ever keeping the GPU and CPU separate is important for high end performance for the same reason overclocking is dying. As processes get smaller the thermal overhead shrinks because it's no longer "how much power can you dissipate" but "how much power can you dissipate on a a given size die" and there's a difficult reality to accept to the latter question. If I want a powerful GPU it needs a big heatsink and big fans (a la R9 290, 780, etc.) There simply isn't room to put a CPU on GPUs of that size without the chip literally melting. If they can then they could always separate them and get more thermal breathing room and get higher clocks so I'd like to see them stay separate.
  • kwrzesien - Friday, January 17, 2014 - link

    +1 insightful

Log in

Don't have an account? Sign up now