AMD Launches "Kaveri" A-Series APUs

Page 2 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Status
Not open for further replies.
Not if the L2 cache is large enough.I don't know if 4MB of L2 cache is enough, but as long as it's large enough, there's no need for L3.
For x86, it seems 4MB of L2 cache is optimium for AMD? implementation. However, if they ditch a couple of cores in place of 12MB of eRAM, it might be faster and consume a bit more power. It was hinted that Intel? eDRAM in Iris PRO took 20 watts constant power and they had 128MB. This means 2 watts extra plus another 3 watts for enhance memory controller to manage this cache. This design has been hinted on the Apple A7 SoC design where a suspected 4MB sRam was seen on a die-shot. It may account for its speed and no need for 128bit memory bus there.In Xbox One design 32MB eSRAM front caches DDR3 ram to make it almost as fast as DDR5 on the PS4. It may be the reason MicroSoft had to cut down on gpu sizes to make room for such enhancements yet not tax the power consumption so much.
 
The future of computing is more in "True audio" block than in any GPU.I don't want to comment performance un till i read a really good review, in the best case it will be on pair with last FX gen (per core). I am hoping that Tom's review is coming!
 
Yup, it also used unified ram.
Except that's only one part of HSA/hUMA. The second (and new) part is that CPU and GPU have access to the same data while both can work with said same data and since nobody marketed the technology for Jaguar, there is no reason whatsoever to believe it's there.

As for the review that could (should?) have been here yesterday, I hope Tom's are also planning to compare just the CPU part, e.g. FX-9370 with R9 290 against A10-7850K with R9 290 (and yes, I'm quite sure that the Vishera would beat Kaveri in most things, I just want to see how much of a beating would that be)
 
As per usual, the GPU half of AMD's APU is great and the CPU half is subpar and holding the entire package back. Sure, they made decent IPC gains, but with lowered clock speeds, the net gain is basically zero in purely CPU applications. HSA will be what makes or breaks the bulldozer family of chips going forward, but we won't really find out how that plays out for at least a couple years.
 
As per usual, the GPU half of AMD's APU is great and the CPU half is subpar and holding the entire package back. Sure, they made decent IPC gains, but with lowered clock speeds, the net gain is basically zero in purely CPU applications. HSA will be what makes or breaks the bulldozer family of chips going forward, but we won't really find out how that plays out for at least a couple years.
 
its an improvement all around, except for price. a $90 intel cpu and a $90 discrete amd gpu will beat the APU in both power consumption and performance, and now the APU doesnt have any price advatage at all....very sad. AMD, this chip neds to be $140, the same as the richland a10, or it will not sell. at least, not as well as the previous a10.
 

I would agree that they weren't impressive in terms of CPU benchmarks, but I knew they wouldn't be the second we had an idea of clock speeds. However, even on the CPU side performance per watt is up a bit, especially if you compare the 7600 @ 45 watts vs previous 45 watt APUs. That will make a great HTPC chip. On the gaming side, Mantle and TrueAudio support may prove to be a boon to not only their latest APUs, but to all GCN 1.1+ hardware.

However I'd say that with this generation the main thing is the HSA capabilities they are enabling with hUMA. Developers will turn to HSA more as it becomes easier to enable (via language, compiler, VM, API support). Unfortunetely for AMD, this doesn't provide a lot of immediate shiny benchmark results. You have to push out the hardware first. Having HSAIL up and running in the next year or so will really give these APUs (and their successors) a big boost.
 

Jaguar is a CPU core. Not an APU. Jaguar is a PART of the larger APUs used in the newer consoles. In fact, by the same token, you can't even compare these chips to other APUs using Jaguar - because they're different. Everything I've read pins both consoles as having memory access capabilities equivalent to hUMA, regardless of what they call them. I'd also recommend looking at some block diagrams for these chips - I think you'll come to the same conclusion.
 
Interesting move by AMD. They're putting a lot of faith in Mantle's eventual ability to take it to Intel and Nvidia. I hope they do well, we really need them to produce top results with this because it may shake this sector up a bit and that can only be good for us consumers :)
 

They're talking about channels, not number of speakers. You can have multiple channels per speaker. Even ancient soundblasters had support for this many channels. The improvements in audio capabilities IMO primarily involve offloading and potential for increased spatial precision/quality/additional effects. Any improvements made to positional audio will benefit any setup. You can better convert a game's 3D positional audio to suit any setup with these kinds of resources at your disposal, even stereo. The more free audio processing (and channels) you have at your disposal without sucking up CPU time, the better.
 


While true, let's face it: that would be so awesome! hahaha

One day you'll be sitting on comfortable speakers while watching a wrap-around screen causing you to dodge bullets and try to get a different look at the main actress' butt. Can't wait for those days 😛
 
Status
Not open for further replies.