So anyway, what you are saying is Zen will give the performance of a hex core intel for less than 160.00. OK........ Anything is possible I guess. But if Zen manages to reach that level of performance, which is highly debatable, AMD would be fools to sell it for 160.00.
I think Zen will make us consumers miss Bulldozer.
Nvidia in terms of hardware has basically one GPU architecture focused in graphics, another focused in GPGPU processing and a SoC line that has both a custom and vanilla ARM cores but uses the graphics part of their GPU-focused chips. The custom core part is being scaled back. That's much more focused in my books.
In any case bulldozer is an interesting processor, historically speaking.
They are indeed more focused on the core graphics business, and yet they are constantly trying to break out into new markets. Why do you think they are willing to lose millions on Tegra? Why do you think they bought a modem vendor? Why do you think they are pushing into cars? They are trying to spread their bets more, so they are not so totally reliant on the core GPU business. They are actively trying to move away from the very business model that you say AMD should emulate.
In any case bulldozer is an interesting processor, historically speaking.
They have single-precision oriented GPUs and double-precision oriented GPUs, but they are reused across many markets. GK104 ended up in Geforce, Quadro, Tesla, GRID, mobile Geforce and mobile Quadro. They are believers in making their parts multipurpose.
They are indeed more focused on the core graphics business, and yet they are constantly trying to break out into new markets. Why do you think they are willing to lose millions on Tegra? Why do you think they bought a modem vendor? Why do you think they are pushing into cars? They are trying to spread their bets more, so they are not so totally reliant on the core GPU business. They are actively trying to move away from the very business model that you say AMD should emulate.
But yes, AMD is very thinly spread at moment. If Zen doesn't break into servers, I would expect them to give up on high end desktop and focus on making SoCs, APUs and discrete GPUs. And trying to break into the ARM Android market is another distraction they really don't need.
Lol, they don't even know what a block diagram is. wccftech is a great source of entertainment.
I'm not saying that Nvidia is not or should not diversify its business, or that it does not reuse IP from one product into another. What I'm saying is that Nvidia does this much more carefuly than AMD does, and within a smaller scope too. Nvidia isn't using its GPGPU architecture top to bottom as AMD does with their, and Nvidia isn't building a CPU core that goes from tablets to servers without the billions Intel has in its bank account, like AMD does.
Anyone else surprised about the tiny 5% IPC improvement? For years we've been hearing things that would suggest that Excavator cores would bring major performance improvements. The change to 32KB L1 dcache, while not exactly trivial, could realistically account for the IPC difference all by itself and doesn't represent any sort of innovation in CPU design. I'm left wondering if AMD had other changes in the pipeline that were scrapped at some point. I do also wonder if this number includes the AVFS overhead or not - technically that would not constitute a reduction in IPC but a dynamic reduction in frequency. I wouldn't expect it to be included.
I'm curious as to what the top turbo clocks end up being; I'm wondering how much the move to HDL has impacted that (if any). A big trump card of Intel's low power chips (at least in the most expensive SKUs) is their huge dynamic range in CPU clock. If Carrizo cuts back dynamic range AMD is going to be at a big loss even if they've made huge perf/W improvements for the ranges they can reach.
It is 5% IPC gain at 40% LESS POWER not clock to clock.
It is 5% IPC gain at 40% LESS POWERnot clock to clock.
edit: my bad, not what i wanted to say
Instructions Per Clock, so wouldn't the conventional interpretation be it is 5% faster (with whatever test suite they use) per clock but uses (up to) 40% less power at certain clock speeds?
http://wccftech.com/amd-carrizo-apu...amroller-die-consists-31-billion-transistors/
Carrizo slides are out.
This is must be why there is no desktop Carrizo, above 25W the Kaveri part is clocking higher.
Instructions Per Clock, so wouldn't the conventional interpretation be it is 5% faster (with whatever test suite they use) per clock but uses (up to) 40% less power at certain clock speeds?
Lol, they don't even know what a block diagram is. wccftech is a great source of entertainment.
At up to 40% less power. Obviously as you increase the frequency the efficiency drops until it starts getting worse.
The slide says 5% IPC gain at 40% less power, not up to.
I have no idea at what frequency Carrizo starts to have worst efficiency than Kaveri, but i believe that will be above 3.5-4GHz.
The graph shows Kaveri start clocking higher a little above 20w.
A 19w kaveri turbos to 3.3. So I'd say 3.5 is the max we'll see in Carrizo 25w.
The slide says double digit performance gains. That could be 10 or 99%. Given the history, I expect 10-15%.
X axis is "power per Core pair" or power per Module not TDP or anything else.
The idea of the APU is a great one when executed well. The problem is AMD's engineers overestimated the speed with which software became multi-threaded.