12 core chip to trump intel?

[georanma]

http://www.engadget.com/2014/01/14/amd-kaveri-pc-processor/

Saw this article tonight. Been quite a long time since I was an AMD fan boy, but that sounds intriguing. They have been a great budget solution consistently, but how will this play in. Still a cheaper chip, but with great built in graphics? How will the onboard play with graphics cards. What could happen?

 

[davedree]

Its not 12 core, its 8 cu gcn shaders + the 4 steamroller cores for the 7850k.
When hsa is developed over the next couple of years, software that can take advantage of it, will make the average 2-8 core cpu redundant. The speed advantage from cpu-gpu compute is huge.

Kaveri is a bit of a disappointment as it potentially could of had a gddr5 memory controller, instead it has to weed itself off by using system ddr3 memory bandwidth.
The 7750 igp instead of offering gddr5 performance is effectively running as a 7750 ddr3. Intel will get very close if not exceed the performance of the amd kaveri with iris pro II.

You can only use the the r240 r250 which feature ddr3 memory to crossfire with the igp.
Pretty pointless might aswell buy a discreet card.

http://wccftech.com/amd-kaveri-dual-graphics-works-ddr3-memory-based-radeon-r7-gpus/

 

[RGone]

Its not 12 core, its 8 cu gcn shaders + the 4 steamroller cores for the 7850k.
When hsa is developed over the next couple of years, software that can take advantage of it, will make the average 2-8 core cpu redundant. The speed advantage from cpu-gpu compute is huge.

Kaveri is a bit of a disappointment as it potentially could of had a gddr5 memory controller, instead it has to weed itself off by using system ddr3 memory bandwidth.
The 7750 igp instead of offering gddr5 performance is effectively running as a 7750 ddr3. Intel will get very close if not exceed the performance of the amd kaveri with iris pro II.

You can only use the the r240 r250 which feature ddr3 memory to crossfire with the igp.
Pretty pointless might aswell buy a discreet card.

http://wccftech.com/amd-kaveri-dual-graphics-works-ddr3-memory-based-radeon-r7-gpus/

So AMD has finally listed the 240 and 250 as able to crossfire with the Kaveri? I looked pretty hard up until a few weeks ago and seemed no one knew for sure just which cards would hybrid crossfire on the Kaveri.

If you have to go discrete card to get some real video power and then disable the video section...heck you might as well get an Intel and discrete card and have some L3 cache for added oomph.
RGone...

 

[ShrimpBrime]

Real gamers don't use IGP and don't tend to buy APU processors.

AMD is just shooting themselves in the foot here.

Now if this was a 12 core desktop (cpu cores not that igp bullsh!t) and even lower clocked say 3.3ghz or something, I'd be hella interested.

Other than that, L3 cpu cache wouldn't make this a gaming rig. The IGP just isn't strong enough..... Yet.

 

[davedree]

Yeah rgone you are right. Altho this applied to Llano and trinity too.Even a dual core haswell, or i3 haswell with a discreet gpu could pack more performance. BUT and a big but, if hsa gains support and Amd are taking a risk on this, seriously software that utilises this architecture will annhiliate any intel 8core8ht future commercial intel.
Furthermore amd wouldn't need to worry about the lack of l3 cache, as huma would help out to readdress memory requests, traffic etc.

Shrimp yeah the enthusiast wouldn't be interested in this, but as I say don't underestimate the strength of opencl and hsa and huma + mantle. Its such a shame that Amd lost the chance to integrate gddr5 with Kaveri. A 7750 gddr5 although not hugely powerful at 1080p would have been a serious upgrade to the recent stagnation of igps.

 

[ShrimpBrime]

Uhg a 7750.... If that's all I had to game on I'd need a 15" monitor and it would be just Godawful.....

No thanks, No IGP for me or anyone I know. Even my Llano rig has a discrete card in it...... cause the 6550D IGP blows chunks on a 19" monitor.

 

[TeknoBug]

Hm isn't that 4 cores + 8 GPU cores? Not the same as 12 cores.

My Motorola Moto X has 8 cores, but it really has 2 CPU cores and 4 GPU cores and the rest handles individual tasks.

 

[davedree]

Hm isn't that 4 cores + 8 GPU cores? Not the same as 12 cores.

My Motorola Moto X has 8 cores, but it really has 2 CPU cores and 4 GPU cores and the rest handles individual tasks.

Yeah 4 steamroller cores and 8 cus (Compute units).

7750 gcn = 8cus = 64Simds x 8cu's = 512 shader cores.

 

[Tyerker]

Ummm... Like over a year old.

http://www.newegg.com/Product/Product.aspx?Item=N82E16819113320

 

[Dolk]

*Sigh* Engineering time noobs

I have been waiting for this type of technology since the announcement of AMD buying ATI. That was a long time ago. AMD has finally done what other manufactures have been trying to do for many years: bring parallel and serial processing together, under one chip. Now you might be saying, "Parallel processing has been around forever in CPUs, Dolky McIdiot! How come your saying its now under the same chip?" Well Mr. Noob, here is the truth.

CPUs have not been able to handle or utilize true parallel processing that we see in GPUs. Their fetch-decode-execute (FDE) cycles have been lacking this capability. Yes multiple threads and multiple cores can allow for a CPU to act very much like a parallel processor, but the scheduling system still has a huge overhead. If anyone has ever worked with tools like Matlab or heavy CAD software, you know what I'm talking about. The CPU has to schedule each process, and than split the parallel process and continuously check in on it to make sure the jumping and branching system doesn't make a mistake. This process costs way too much time for the CPU and creates a large amount of inefficiency.

Now lets talk about Kaveri (awful name isn't it? How am I suppose to pronounce it? Ka-ver-y? Kav-ery? Kavity?) Anyway, this unique chip allows for the CPU and GPU cores to communicate with each other on the same bus. Now the CPU will be able to handle all of the single threaded processes, like it was designed to do since it's conception. The GPU will be able to handle the parallel processing, like it was designed to do since it's conception. This unique architecture will bring new meaning to speed in software. Furthermore, it will allow for much less overhead in heavy parallel processing.

As of right now, most users will not be able to see the full benefits of this type of architecture. Yet, for engineers and programmers, like myself, we are saying WoopEEEEE! No more need for crazy amounts of GPUs and CPUs in order to simulate or solve crazy complex equations. Mantle is probably the best innovation towards software and hardware that I have seen since multi-core become the norm. I think people are not taking this announcement as seriously as they should. Seamless control with little overhead in one chip is just disgustingly beautiful, and has only been achieved with very advanced ASIC and FPGA design.

Here is just the start of what is capable of Mantle:

 

[RGone]

Good write-up Dolk. When AMD gets it tweaked even more...well then we are waiting on software again. Never ending circle for most of us. New Tech. New Software. Oh wait on software. Then Go. Go to begin again.
RGone...

 

[\dev\goat]

*Sigh* Engineering time noobs