Kaveri = BF4 w/o graphics card..

gman97005

Honorable
Oct 31, 2012
190
0
10,710
Well, Kaveri will run BF4 at medium settings on 1080p without a graphics card @ 35fps, absolutely outstanding performance for just the APU, show me an Intel CPU that can do that without a dedicated GPU..
Keep in mind this is without the HSA enabled and without Mantle, both the HSA and Mantle will greatly boost performance far above and beyond what we are seeing today and adding a dedicated GPU in Hybrid CrossfireX will only make it more powerful, talk about price/performance you just can't beat Kaveri..
January 14th release date and I am ready to buy it the moment it hits the shelf at Newegg, in the meantime i am laughing all the way to the bank..
 
Solution
from separate leaks it's looking like Steamroller is pushing out similar single cored IPC to Sandybridge... granted Haswell is about 15% faster then SB. Now to be clear, i said "similar"... the numbers seem to indicate Steamroller is probably 5% slower then sandybridge at the same clock speed. That's pretty close though and marks the first significant improvement in AMD's IPC in 3 years.

It also cuts the intel IPC advantage in half.

The Kaveri gpu performance puts it inline with the lower end results of a hd7750... depends, as you know benches of gpus are influenced by the cpu, and there is no bench of a hd7750 and a quad core steamroller cpu to compare it to. That said, with the igpu using GCN 1.1 cores, and sporting...

theagentd

Honorable
Jun 23, 2012
14
0
10,510
Mantle does nothing if you're GPU bottlenecked. At most you'll get a few extra FPS from Mantle. Hybrid Crossfire suffers from heavy microstuttering.

Move along, people. Nothing to see here.
 


If anything, the GPU will bottleneck the CPU, which Mantle will fix. Nobody said anything about crossfire. Please read the thread before you tell people to ignore it. This could actually develop into a pretty good discussion. Thanks for the link Squid. It should do pretty well with mantle it seems, assuming it's as great as AMD is saying, I'm sure it'll stay around 35 fps with mantle in a rocket laser tank battle.
 
from separate leaks it's looking like Steamroller is pushing out similar single cored IPC to Sandybridge... granted Haswell is about 15% faster then SB. Now to be clear, i said "similar"... the numbers seem to indicate Steamroller is probably 5% slower then sandybridge at the same clock speed. That's pretty close though and marks the first significant improvement in AMD's IPC in 3 years.

It also cuts the intel IPC advantage in half.

The Kaveri gpu performance puts it inline with the lower end results of a hd7750... depends, as you know benches of gpus are influenced by the cpu, and there is no bench of a hd7750 and a quad core steamroller cpu to compare it to. That said, with the igpu using GCN 1.1 cores, and sporting about the same specs as a 7750, we probably should be expecting better performance from it then this if everything was equal.

Everything is not equal. We're getting low end gddr5 7750 performance from a igpu with ddr3 ram. Which means this igpu will really benifit from fast ram. Frankly i didn't think this type of performance was possible on a ddr3 system, which means AMD worked some serious magic with their memory controller. Which of course whispers at the secrets to steamroller's improved IPC.

Overall there is a lot here to get excited over. If amd can fix dual graphics, this cpu should xfire/dg with a 7770/7750 and possibly with a r7-260x. A 7750+7770 in xfire is around a 7870xt in performance. That would be nothing to laugh at.

 
Solution


It was demonstrated at AMD's APU13 developer's conference. Here is a link to PcPer.com which has video demoing the game for less than 1 minute. The game is mostly set to medium with light quality set to low and no anti-aliasing. The settings can be seen in video at about 0:48 into the video.

http://www.pcper.com/news/Graphics-Cards/Video-Battlefield-4-Running-AMD-A10-Kaveri-APU-and-Image-Decoder-HSA-Acceleratio

The person narrating the video states the frame rates are high 20's to low 30's. The action takes place in a building with no other visible characters until near the end of the demo. He also states the version of Battlefield 4 has not been optimized for Mantle yet.
 

gman97005

Honorable
Oct 31, 2012
190
0
10,710

I'm betting on the R7 260x but i will wait until Kaveri is in my hand and people like you and others have determined what the best GPU choice will be, don't want to buy one now only to discover it was poor choice..
 

theagentd

Honorable
Jun 23, 2012
14
0
10,510


I won't deny that Mantle will most likely improve performance by allowing BF4 to more efficiently utilize the GPU, for example through more effective memory management, improved pipelining and the ability to run compute jobs (e.g. lighting for BF4) at the same time as rasterization jobs, but believing that this will grant any significant changes on such a weak GPU is just wishful thinking. The most likely bottleneck this card will encounter is shader performance which is the one thing that Mantle will not affect at all. Mantle runs the exact same shaders as DirectX, most likely compiled to almost exactly the same assembler instructions for the stream processors.

If you watch the developer slides for Mantle you can see that the focus is on moving much of what has traditionally been the driver's responsibility to the developers which empowers developers to use the GPU more efficiently and closer to how the actual hardware operates, which at the same time massively simplifying the Mantle driver. The result is that the overhead is much lower since the abstraction layer is thinner and CPU utilization can be improved since you can submit render batches from multiple threads.

For these reasons the heavily CPU-optimized BF4 will see very little performance improvements for low-end GPUs which are bottlenecked by the GPU like those built into AMD APUs. On the other hand, for a game like Planetside 2 which is ridiculously CPU-bottlenecked, Mantle could essentially double or triple CPU performance for an almost equal FPS improvement while at the same time allow for an increased variety in the 3D models used thanks to the lower draw call overhead.

TL;DR: Mantle doesn't make GPUs run shaders faster. It just lets you optimize how you use the GPU for massive CPU performance gains but only a minor GPU performance boost, hence Mantle does very little for the APU mentioned in this thread.