Wanna join the discussion?! Login to your HotHardware Forums forum account, or Register a new forum account.
Personally I don't think they should. Steamroller isn't good enough to deserve it, and they'd have to crank clock down harder for eight cores.
An eight-core Steamroller-based FX chip might be a 3.7 - 4Ghz part. It'll be the same Cpu performance story all over again.
Uhm... AMD definitely SHOULD use GDDR5 because IGP's are extremely bandwidth starved for one thing, and AMD's APU's actually perform a lot better with higher bandwidth - which is much higher in GDDR5 compared to DDR3 and DDR4.
Also, I think people latch on an outdated way of thinking too much, separating CPU and GPU.
AMD has gone a different route, towards unifying both and using both to accomplish the same work much faster and with far greater efficiency.
Steamroller DID deliver a 20% IPC increase in multithreaded tasks.
Also, people fail to take into account that a lot of the software uses 'Intel genuine' code which utterly misrepresents AMD's overall capabilities and runs code slower on AMD's hardware.
Software should be written to take advantage of ALL cpu architectures efficiently, not just the one that's more 'dominant' (because if you don't, you end up with highly skewed results).
To that effect, I do think Intel would still have stronger single-threaded performance, but this in particular is falling out of relevance fast as more and more programs use multithreading (and AMD to my knowledge DID score all console wins, as well as has the HSA foundation).
And now, AMD has gone a step further, but the software has been left playing catch up.
What AMD may have sacrificed a bit in FPU performance (which is reflected in specific situation), it was supposed to have been replaced by HSA, but as I already said, the software is essentially outdated and cannot fully test this Kaveri properly.
What we did see of HSA potential was a tidbit at best... and it easily blasted through Intel 'brute force'.
Also, any gaming comparisons at this stage is utterly irrelevant.
The drivers for one thing have not been optimized for Kaveri yet, and there's still a question of Mantle - so I think people should be much less dismissive and actually ask why the heck are we still using horribly outdated software?
1). Agreed that GDDR5 would improve GPU performance. Doesn't make it make sense for CPU perf, and that doesn't mean mobo manufacturers want to adopt it.
2). GenuineIntel hasn't been the same issue for years. The FTC went after Intel for it, as did AMD's own lawsuit. It's not crippling AMD's performance these days and the tests we use aren't compiled with Intel compilers.
3). Steamroller did deliver a significant IPC gain (the amount varies) alongside a 10% frequency loss that caused the high-end chip to give back most of what it offered. The 45W part is better.
4). AMD's HSA SDK won't be ready for months. The OpenCL 2.0 driver won't be ready until 2015. No one can write code until AMD makes drivers and SDK available.
5). Seriously replacing the FPU altogether was more an idea than reality. Until the GPU can run native x86 code (never), this can't happen.
6). Gaming comparisons are absolutely relevant. Your non-Mantle games will not run faster because AMD releases Mantle.