ShintaiDK
Lifer
- Apr 22, 2012
- 20,378
- 145
- 106
Until you hit unsupported codec. Or anything where the CPU ends up too slow.
Your lack of humour is funny.Until you hit unsupported codec. Or anything where the CPU ends up too slow.
Duty Cycle Control only helps with reducing power use in low utilization situations- it won't help when the GPU is used in a performance intensive situation.
Example: If you have 24 EUs that have to operate at 600MHz, you could have 48 EUs at 300MHz at lower voltage.
AMD doesnt have 4K decode ability in any APU nor do any AMD product have HDMI 2.0.
Easy enough to run them at higher power if the hardware is up to it although that's not really what these processors were meant for. For instance here's my i7-4700MQ thinking it's using less than 4W while running Linpack AVX2. Less than core-m . Temperature is a bit of a giveaway though.Really, all those high-end iGPUs at 15W TDP are meaningless. Those SKUs cannot use 100% of that iGPU performance capabilities.
I thought DCC was more about having bursts of 600MHz, followed by bursts of 0MHz (i.e. power gated off), very rapidly? (I may be confusing it with something else.) But yeah, that definitely lines up with what AtenRa was saying: "Those SKUs cannot use 100% of that iGPU performance capabilities." It's about power efficiency, not flat out performance.
(Spending die area on more shaders to drop clocks is pretty common in GPUs. It's the entire basis of the mobile GPU market, for a start )
They will have the same problem. They are using 14nm with 30% less power usage than 22nm but at the same time they increased performance capabilities by more than 20-30% over Haswell.
Example: If you have 24 EUs that have to operate at 600MHz, you could have 48 EUs at 300MHz at lower voltage.
Easy enough to run them at higher power if the hardware is up to it although that's not really what these processors were meant for. For instance here's my i7-4700MQ thinking it's using less than 4W while running Linpack AVX2. Less than core-m . Temperature is a bit of a giveaway though.
The shocking reality is that it is a 47W SKU running more than 70W with less than default voltage and 2 extra bins.
HT degrades AVX, the threads you see are core x - thread 1, thread 0 is not shown. Can run it again showing those threads but it should be evident that you do not get 170GFlops with 0% usage.Core and thread usage is ~0%. The program isn't running.
You are also not running stock clocks on a locked CPU (partially unlocked). Likely the settings and the bios are having problems.
14 nm is more than 30% more efficient. That number comes from more than a year ago on a premature process.
14 nm is more than 30% more efficient. That number comes from more than a year ago on a premature process.
14 nm is more than 30% more efficient. That number comes from more than a year ago on a premature process.
Citation needed*That was the theory, in practice it s 30%, at least, less efficient than their 22nm.
Baytrail vs Broadwell.
Again, you prove how little you know about what is going on.
Pretty sure that Kabini has HDMI 2.0? But yeah, I'd forgotten about the h.264 decoder not supporting 4k until Tonga! That's definitely a no go.
HT degrades AVX, the threads you see are core x - thread 1, thread 0 is not shown. Can run it again showing those threads but it should be evident that you do not get 170GFlops with 0% usage.
No problem with BIOS settings, I modified them myself.
Here you go, I've thrown HWMon in as well, still is buggy with negative offsets I see.
That was the theory, in practice it s 30%, at least, less efficient than their 22nm.
Baytrail vs Core M. These don't share the same process. At all. One is a SOC process, one is a HP process. Different design rules, different performance, different everything. You may as well be comparing numbers from TSMC and GF and IBM and drawing conclusions about Intel process characteristics.
The other link is a bunch of screenshots with more data that can't meaningfully be correlated. Stop generating walls of text and random screenshots and then fabricating conclusions that have nothing to do with either the text or the screenshots.
You made the argument, you need to do the work to back it up.There s other datas if you want to dig the thing, 1.0V at 2.6 for core M and 3.5 for Haswell, is that enough or do you need me to make your homework.?.
You made the argument, you need to do the work to back it up.
It's really quite simple. People are complaining that the lower power CPU's are throttling too much and not able to sustain maximum performance. Read the first quote before my original attachment. By tricking the CPU into believing it is running at a much lower power than it really is then no throttling will occur providing cooling and power delivery is sufficient.And I still think your settings are wrong or something is messed up with your computer.
There s other datas if you want to dig the thing, 1.0V at 2.6 for core M and 3.5 for Haswell, is that enough or do you need me to make your homework.?.