Tuesday, June 14th 2011
FX Turbo Core Technology Bumps Frequency by 1.00 GHz
With Intel's introduction of Turbo Boost technology, a new feature was introduced to the industry, where a processor overclocks itself in short bursts to handle increased CPU loads. AMD quickly followed with its own similar feature, called Turbo Core, with Phenom II X6. With the company's upcoming high-end client FX-series processors, that technology is being given an update. On the FX-series processors, the technology is designed to bump clock speeds by as much as 1.00 GHz over the processor's advertised clock speed, within the processor's TDP headroom.
A company slide reveals that FX-series processors can run all cores at nominal speeds (advertised speeds), all cores at a bumped "Turbo" state, or with half the number of cores running at max turbo speeds with up to 1.00 GHz (5.0x BClk multiplier) increase in clock speeds, with the other half the number of cores in C6 state, completely shut off. Turbo Core ensures increased performance in applications that are designed to work with lesser number of cores, most games come in this category.
Source:
DonanimHaber
A company slide reveals that FX-series processors can run all cores at nominal speeds (advertised speeds), all cores at a bumped "Turbo" state, or with half the number of cores running at max turbo speeds with up to 1.00 GHz (5.0x BClk multiplier) increase in clock speeds, with the other half the number of cores in C6 state, completely shut off. Turbo Core ensures increased performance in applications that are designed to work with lesser number of cores, most games come in this category.
55 Comments on FX Turbo Core Technology Bumps Frequency by 1.00 GHz
AMD however has completely tossed the entire design out the window. Actually given equal process I expect AMD and BD to beat intel in a number of ways but not in pure IPC per clock.
Heres what I expect
Slightly lower IPC per mhz, but a higher clock to correct for that, being able to share resources lets them stick more in a core, also their design better distributes thermals over the chip and its more segrgated in a few ways, so they can really clock up the chip and shut down uneeded core to improve IPC in workloads where high IPC matter and core count is less relevant.
They actually have done a good job, now the issue will become execution, did they get the execution right, if they did they should have a incrediable value with good thermal and power management with competitive IPC " not in a clock for clock ratio though" but in a core for core ratio.
If they did everything right.
this design really hits at where the workloads are, more INT then FPU and then a powerful FPU to cover those workloads as well. Its a fucking brilliant design. Question is do they have the quality of cache, int,fpu,decode,prefetch etc to be comparable to intel. Given the R&D time involved, I don't think they left many stones uncovered. This chip is aiming for a braod range of efficiency compromises and if done properly, should be a great product.
Let's just say that I don't plan on staying withing TDP margins when I get my hands on one of these.
2. Yeah CPUs already do this, when I'm not doing anything my cpu runs at 800mhz ( not sure if that's what you meant)
3. See 1.
I have my E8400 running @ 9*225=2.025 GHz: with speedstep, it reaches 6*225=1.35 GHz. I was wondering if the reverse of my first question was possible where, by reducing the X TDP in BIOS, thus giving me a lower speed while still giving the 1 GHz turbo increase. Underclock it to ... say ... 2.2 GHz: will you have a turbo reaching 3.2 GHz / not have a turbo / turbo reach 4 GHz due to TDP still being X (not changeable in BIOS)? This is my point in question 2.