• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

AMD Launches Three Kaveri APU SKUs in February 2014 – Feature Set For A10 and A8 APUs Detailed

It won't be all that useful benching it, one of the biggest features of Kaveri is HSA and the unified memory. Until you can get likely a windows update and a public driver for it, final performance wouldn't be known. Likely needs an updated 290x driver to go along with it.

They may ship a working HSA/unified mem driver with the chip, or it may be working in windows already for all I know. However with the chip not out I absolutely wouldn't be surprised to see these things implemented/released nearer launch, I also wouldn't underestimate how big a difference unified memory(hUMA) and HQ will make to gaming.
 
An APU is expensive if you don't use the graphics part. e.g. a 6800K is over £100, the same price as an 8320 basically with nowhere near the processor performance.

With Kaveri the performance may be a lot higher, making it good as a CPU-only part.
 
Unified memory will bring with it some significant performance benefits, however I can't honestly say in which situations it will/won't work.

Will they be able to make(eventually) an 8+ core with no gpu HSA chip, or is having a GPU on die fundamental to HSA working. IE a newer FX chip with 8 steamrollers can't in fact be HSA enabled. Second, if you have a HSA enabled chip and it does require a gpu on die, how does this work with a discrete gpu. I honestly don't even know the situation today with a richland + discrete gpu. Can the discrete go into zero power mode and on die gpu be used for desktop, there are loads of situations I can think of where I'm not quite sure if it works.

ultimately the push for HSA is for on die acceleration via the gpu of MANY applications. With Java getting HSA support soon, where almost any Java application will be incredibly easily made parallel "aware" essentially and will be capable of using the on die gpu.

Where the industry is going is, you may have a Java app that, be it a game or anything else, runs at 1/10th of the cpu power you have. But look over what Intel/Arm are doing. Power savings and efficiency comes not from having the lowest load power usage possible, but by being as fast as possible, getting the work done, and going back into idle mode.

So rather than a chip taking 3 seconds to do some work at 12W then going to idle at 1W. It's far better to use 24W at load, for 1 second then getting back to 1W idle. Arm cores are all about accelerating everything as much as possible to get back to idle. This is one of the reasons that HSA is so important, every wasted memory copy(unified memory removes craploads of these) every second less the cpu can be powered up the better.

The other issue is there are things you can offload on die, that make it worthwhile, but both the power to send a signal off die(one of the biggest power savings SOC's have made and why everything is a "soc" now, is because off die communication uses magnitudes more power and for magnitudes longer) and the latency hit. If it takes 0.002ms to send it to the on die gpu and saves 0.003ms, win. If it takes 2ms to send it to the discrete gpu and saves 0.1ms(faster off die gpu), then you lose, plus you've powered everything up for 4ms longer(it's got to get back to the cpu), and you've woken up the discrete GPU, etc, etc.

On die gpu's are becoming necessary rather than being thought of as extra or expensive for budget gaming. They are for acceleration, they speed things up and reduce power usage. The software stack is finally approaching, the industry is pretty much on board, some of the biggest/widest useds languages like Java are getting simple to implement support.

Overall Kaveri + mantle + unified memory(hUMA) + HQ(letting software access the gpu directly rather than always through the CPU) could all very easily add up to Kaveri being the fastest gaming platform, and fastest in many other things.

It won't have the raw cpu power to beat Intel in very simple compute, but it could beat it out in many places where gpu acceleration/unified memory make a huge difference.

What I really want is a 3 module Kaveri :( probably be looking at 3-4module + gpu chips from AMD and hexcore mainstream from Intel in 2015 now, but they should be good at least.
 
There can (and will) be HSA enabled CPUs that lack an iGPU. The chip needs to have hardware IOMMU and be supported by the motherboard/BIOS (AMD have been ahead of Intel in this respect for yonks). This coupled with a HSA enabled dGPU will allow the kind of fancy load balancing and unified memory access.

As far as APU + dGPU power, ZeroCore is (iirc) a feature of GCN so hasn't worked on Hybrid Crossfire setups as of yet (discounting the bug/unintentional support of HD8670+HD7750), but I see no reason why it would not work with Kaveri+GCN, drivers willing.

iGPUs are essential today more so than ever in the consumer/OEM space. The vast majority of off the shelf boxes and boring office machines will never see a dGPU, ever. Having a capable iGPU (relatively) has so far been of little importance to enthusiasts like us. If HSA can leverage all GPU cores to make life better, regardless of where they reside in your system, having a decent iGPU could become an important metric in CPU buying choice soon enough
 
drunkenmaster said:
Will they be able to make(eventually) an 8+ core with no gpu HSA chip, or is having a GPU on die fundamental to HSA working.
Click to expand...

Have a quick read of this http://www.amd.com/us/products/technologies/hsa/Pages/hsa.aspx#2

What AMD seem to mean by HSA is having a single chip containing conventional CPU cores (good at sequential, bad at parallel tasks) and GPU-like cores (bad at sequential, great at parallel tasks). Look at the chip in the PS4 for a concrete example (see Gazb's link above).

HSA (Heterogeneous Systems Architecture) is an intelligent computing architecture that enables CPU, GPU and other processors to work together in harmony on a single piece of silicon by seamlessly moving the right tasks to the best suited processing element.

HSA will empower software developers to easily innovate and unleash new levels of performance and functionality on modern computing devices, leading to powerful new experiences such as visually rich, intuitive, human-like interactivity.
Click to expand...

I suspect this isn't (yet) going to affect anybody who needs a dedicated GPU, rather it's aimed at "modern computing devices" (read, all machines not for gamers or scientific computing).

d_brennen seems to know more than me on this though.

d_brennen said:
This coupled with a HSA enabled dGPU will allow the kind of fancy load balancing and unified memory access.
Click to expand...
 
Last edited:
Having local GPU cores can only be a good thing, and AMD's focus on APU designs of late have pushed all the HSA materials we've been shown in this direction. I can imagine a generation or two of Opterons that lack an iGPU, but my guess is as good as yours, joeyjojo
 
You must log in or register to reply here.
Back
Top Bottom