blackkstar
Honorable
gamerk316 :
Reepca :
tourist :
Reepca :
Apologies if it's already been stated earlier in the thread (it's a lot to read through), but does anyone know if AMD plans to continue production of their FX chips (that is, desktop non-APU CPUs)? Or if it's even a possibility? Steamroller, if I understood correctly, gained a respectable IPC increase over Piledriver, so is there a chance that AMD could produce some Steamroller CPUs? Or Excavator? I understand that HSA is certainly more efficient, but the point remains that you can only fit so much into one chip - as much as I would love to have an AMD Steamroller 8-CPU-core-290X-level-GPU APU, it isn't happening anytime soon.
On another note, does anyone know how much mantle affects Dual Graphics performance? The reason Dual Graphics wouldn't normally work very well is that, by design, the APU should have a balance of CPU and GPU so that neither of them bottleneck the other. Doubling the net GPU computing power would cause a bottleneck. But where Mantle, as far as I have read, shines the most is where the CPU is the limiting factor - which sounds like Dual Graphics to me!
On another note, does anyone know how much mantle affects Dual Graphics performance? The reason Dual Graphics wouldn't normally work very well is that, by design, the APU should have a balance of CPU and GPU so that neither of them bottleneck the other. Doubling the net GPU computing power would cause a bottleneck. But where Mantle, as far as I have read, shines the most is where the CPU is the limiting factor - which sounds like Dual Graphics to me!
some games yes some no
http://www.hardcoreware.net/radeon-dual-graphics-performance-kaveri/4/
I see. But the line (last page) that I noticed the most was "Seems to be a "best case scenario" for Dual Graphics, but Mantle doesn't work yet.", describing Battlefield 4 at the time. I suppose the ideal test would be to test Dual Graphics on BF4 without Mantle and then with Mantle. Are there any other, more recent statistics I could look at (that you know of)?
Additionally, I imagine that with the newer drivers available now, perhaps we will see more gains in other games?
Dual graphics, from an HSA perspective, really doesn't make any sense. You either have the APU being starved due to the memory bandwidth bottleneck, or the discrete GPU being starved due to the latency involved in copying data across the PCI-E bus. That's why you don't see the same scaling you do with SLI or CF.
Mantle does NOTHING to address this issue. Mantle is simply a way of reducing the CPU bottleneck in games, so when in a situation where the GPU has headroom, you get an increase in FPS based on the added performance the CPU has to perform.
palladin9479 :
He needed to overclock the CPU_NB, otherwise your going to get poor performance with AIDA64. Don't ask me why but AMD has the CPU_NB running at a much lower clock speed then it's capable of, and the IMC and prefetch system run inside there. My fx8350 has it's NB clocked at 2.6Ghz and gets numbers much higher then what he posted, same with my 6800K (2.2 or 2.4Ghz can't remember). Kaveri comes with the CPU_NB clocked at something piss poor 1.8, pretty much every OC guide recommends clocking it to 2.0 ~ 2.1Ghz. This was the only weakness I could find in the 7600, it doesn't let you change the CPU_NB.
I remember running some AIDA tests on my different chips and the CPU_NB speed had a very large effect on the results.
I remember running some AIDA tests on my different chips and the CPU_NB speed had a very large effect on the results.
Personally, I have a release FX 8350 and the CPU_NB does not like to be OCed at all. It takes a lot of voltage and gets really hot really quick.
I remember back when we were all figuring out these chips, CPU_NB was a struggle to OC for a lot of people besides me. Now it seems a lot easier to crank it up. I don't think AMD has changed spec for them yet on the newer chips. Which is sort of a shame. They could have sold FX 9000 series with faster CPU_NB speeds out of the gate and really helped performance and benchmark scores.
Looking at what the new Vishera dies do compared to what I have, I think GloFo 32nm has come a long way.