AMD Launches New Mobile APU Lineup, Kabini Gets Tested

An anonymous reader writes "While everyone was glued to the Xbox One announcement, Nvidia GeForce GTX 780 launch, and Intel's pre-Haswell frenzy, it seems that AMD's launch was overlooked. On Wednesday, AMD launched its latest line of mobile APUs, codenamed Temash, Kabini, and Richland. Temash is targeted towards smaller touchscreen-based devices such as tablets and the various Windows 8 hybrid devices, and comes in dual-core A4 and A6 flavors. Kabini chips are intended for the low-end notebook market, and come in quad-core A4 and A6 models along with a dual-core E2. Richland includes quad-core A8 and A10 models, and is meant for higher-end notebooks — MSI is already on-board for the A10-5750M in their GX series of gaming notebooks. All three new APUs feature AMD HD 8000-series graphics. Tom's Hardware got a prototype notebook featuring the new quad-core A4-5000 with Radeon HD 8300 graphics, and benchmarked it versus a Pentium B960-based Acer Aspire V3 and a Core-i3-based HP Pavillion Sleekbook 15. While Kabini proves more efficient, and features more powerful graphics than the Pentium, it comes up short in CPU-heavy tasks. What's more, the Core-i3 matches the A4-5000 in power efficiency while its HD 4000 graphics completely outpace the APU."

Comments

hUMA

[Anonymous Coward]

heterogeneous Uniform Memory Access is really what one should be paying attention to. With that tech in both of the upcoming consoles and major support from the same, Intel better watch out.

Re:hUMA

[VortexCortex]

I can give you an example in gaming: TWICE THE WORLD GEOMETRY. The data has to be loaded from persistent storage or network into main RAM, then that same exact data must be shoved over into the GPU in batches to be rendered on demand. With hUMA I don't have to have a copy on the GPU and a copy in main memory -- just one copy. That means TWICE the geometry with the same amount of total RAM.

Furthermore, physics is great on the GPU I can parallelize the hell out of that. However, triggering sound effects and updating network state via read-back buffer is a horrible slow hack. hUMA means the GPU can actually be used to update gamestate that actually matters -- instead of just non-gameplay affecting things like particle effects. Logic can be triggered much more easily and course grain physics data can be read back at will for network synchronization. Client side prediction (latency compensation) also becomes a lot cheaper.

I can get a crap load of fine structural detail rendering and acting to physics right now on discrete GPUs, but the problem is when I want any of that to actually mean anything in terms of gameplay, I have to read back the data to the CPU side. hUMA utterly destroys the barriers preventing all sorts of RAM intensive gameplay. Hell, even weighted logic trees for AI can be processed on the GPU instead of only on the CPU, and we'll have the RAM budget to spare because we don't need two copies of EVERYTHING in memory all of a sudden. That means larger more complex (read: smarter) AI, and lots more of them.

Folks really don't realize how horrible the current bottleneck is. You want world that's fully destructible down to the pixel (atomic voxel), with models that actually have meat under the skin, and rebar in the walls, and with different physical properties so that you can freeze a door then shatter it, or pour corrosive acid on the hinge or create reactive armored structures on the fly by throwing some metal plate atop explosives atop the concrete bunker... Yeah, we can do all that on the GPU right now. However, without hUMA, on the CPU logic side of things the GPU is seen as a huge powerful black box -- We put the equations and bits of inputs in, amazing stuff happens, but we can't actually tell what's going on except for through a very tiny output signal -- the RAM transfer bottleneck; So, we can't really act on all the cool stuff going on. Right now that means we have to just make all the cool GPU stuff not important for gameplay, like embers that burn and blow about but can't burn you, or drapes that flutter in the breeze but can't be used to strangle someone with, or tied together to make an escape rope; Unless we planned all that out in advance.

For crying out loud

[Anonymous Coward]

On Wednesday, AMD launched it's latest line of mobile APUs, codenamed Temash, Kabini, and Richland.
.
Should be:
.
On Wednesday, AMD launched its latest line of mobile APUs, codenamed Temash, Kabini, and Richland.
.
.

Re:Price & power consumption

[strata8]

Under $70. The highest spec embedded Kabini part is $72 so we can expect retail to be a bit below that.

Intel officially prices the i3 3217U at $225 but somehow I think that's not the actual price it's sold at.

Re:Price & power consumption

[WaroDaBeast]

amd cant compete on power consumption

... and that's exactly why AMD's CPU's power consumption in this article is lower. Now tell me, were you always this bad at math, or did it occur after an accident?

matches power consumption?

[Luke_22]

What's more, the Core-i3 matches the A4-5000 in power efficiency while its HD 4000 graphics completely outpace the APU.

has anyone bothered looking at the benchmarks? The overall system power consumption when games were run was 20watts for AMD and 35watts for the Core i3.
To my calculation, that's a 75% more power consumption then AMD. Intel hardly "matches" anything...

AMD was still at least 3 watts less power hungry in any other benchmark, too...