| Electronics > Projects, Designs, and Technical Stuff |
| CPU Clock Speed based on MOSFET Physics Gate Length, Electic Field... |
| << < (6/6) |
| Agent:
--- Quote from: viperidae on December 17, 2018, 10:27:41 am ---Why would you ignore turbo frequencies? The silicon is perfectly capable of running at the higher speeds, the only issue is thermal. --- End quote --- I ignored turbo because it's still kind of new when comparing to the entire history. That gets into OC'ing and I need to draw the line somewhere. Plus it's tops out at 5.0 GHz, so not much of a huge difference from AMD's base clock of 4.7 GHz. --- Quote from: T3sl4co1l on December 15, 2018, 06:21:24 pm ---I wonder if there is any interest in making highly serialized, simplified cores. Namely: strip out some of the complicated logic (floating point? SIMD? simpler memory addressing models?) to push the clock rate higher. The assumption being higher clock speeds give higher performance in frequently-branching (difficult-to-predict) code. Or, higher order prediction and speculative execution units. *shrug*, it may be there isn't any way to optimize such problems (namely, those that are strongly locked by Amdahl's law), and the best approach can only be a special-ish-purpose ASIC. (Some crypto algorithms do this intentionally.) The hope would be that something a bit more general (a CPU) would be capable of doing other Amdahl-locked problems faster. Programs that are heavily serial in logic and light in arithmetic, even if they could be parallelized properly. Idunno. They're probably already doing as well as they can, considering how much of today's programs fit this bill. Tim --- End quote --- http://www.cpushack.com/2013/02/21/charles-moore-forth-stack-processors/ I was told that the same guy that developed this made a RISC processor in the 90's that would have blown everything else out of the water. He sold two exclusive licenses to a company called Corda and the other one to Intel. There was no way Corda could afford to battle Intel in court. Intel ended up burying the technology. I wanted to stick with the x86-64 instruction set. |
| T3sl4co1l:
Stack-based reminds me, there are a lot of chips offering Java bytecode support to some extent or another. The world came around eventually, just with a more accessible (and whether you want to argue if it's better or worse, that's a separate issue :P ) language. Tim |
| Kilrah:
--- Quote from: Agent on December 17, 2018, 04:35:06 pm ---I ignored turbo because it's still kind of new when comparing to the entire history. --- End quote --- But you should see it the opposite way. Turbo is misleading, it suggests a core is "boosted" past its native clock, but it's actually the opposite. Turbo is as close to what the specced "native" clock of that CPU is (which best compares to older CPUs), and "base clock" is an arbitrary crippling of it for thermal reasons when multiple cores are packed together. Base clock really means nothing at all. I have a mini PC running an m3-6Y30 CPU (GPD Win 2) with lots of BIOS customization options. That CPU has a "base" of... 900MHz, Turbo 2.2GHz. In the computer it's by default configured as 7W TDP up from Intel's 4.5W recommended setting for that CPU. So it'll limit itself to a long term power dissipation of 7W. That leads to both cores running at about 1.6GHz under load aka nothing close to the 900MHz "base clock", if only one core is active it will go to the max turbo of 2.2. Now adjust the platform TDP to 14W in the BIOS... Both cores will happily run at 2.2 continuously (cooling obviously just needs to handle 14W, which it does). They really are 2.2GHz cores, NOT 900MHz ones. |
| NiHaoMike:
--- Quote from: Kilrah on December 17, 2018, 09:31:24 pm ---Base clock really means nothing at all. I have a mini PC running an m3-6Y30 CPU (GPD Win 2) with lots of BIOS customization options. That CPU has a "base" of... 900MHz, Turbo 2.2GHz. In the computer it's by default configured as 7W TDP up from Intel's 4.5W recommended setting for that CPU. So it'll limit itself to a long term power dissipation of 7W. That leads to both cores running at about 1.6GHz under load aka nothing close to the 900MHz "base clock", if only one core is active it will go to the max turbo of 2.2. Now adjust the platform TDP to 14W in the BIOS... Both cores will happily run at 2.2 continuously (cooling obviously just needs to handle 14W, which it does). They really are 2.2GHz cores, NOT 900MHz ones. --- End quote --- When I first got my i7-3930k, I was able to tweak the BIOS settings to allow it to run at 4.1GHz on all 6 cores at once for an indefinite amount of time. I don't run it that high very often nowadays since a lot of the compute intensive applications I use care more about the GPU. It turns out the default 3.2GHz ends up running it pretty much at its peak efficiency point, so I normally disable turbo in sysfs, but have the option of reenabling it when the need arises. Too bad mining earnhoney is no longer profitable on that old Sandy Bridge-E, since it otherwise would be a good time to rev up those cores in light of a recent event... |
| Berni:
Yes turbo speeds are only ever reached with 1 or 2 cores loaded and other sitting idle. With all cores loaded you usually only get a bit more than the base clock (Or even less than the base clock if its a Apple Mac where they rather have the CPU cook itself than ramp up the fans) I do recommend going into the BIOS and manually increasing the TDP limits. With the right settings you can get all cores simultaneously running at full turbo speed when needed. But you should also have a big beefy cooler mounted on it as the higher end chips put out a lot of heat when running like this, the wimpy stock cooler has no chance. |
| Navigation |
| Message Index |
| Previous page |