Thursday, April 23rd 2020
AMD FX-8350 Pushed to 8.1 GHz via Extreme Overclocking by Der8auer
AMD's Bulldozer architecture is a well-known quantity by now, and seemingly straddles a line between loathing and love between tech enthusiasts. Slow and power hungry compared to Intel's options, it harkens back to a time where the roles were reversed, and AMD were looking to compensate for architectural deficiencies (and architectural design decisions that can either be claimed as erroneous or ahead of their time) via increased clockspeeds. However you look at these Bulldozer CPUs, the fact is that they remain some of the best overclockers of all time - at least when it comes to maximum operating frequencies, especially at absolutely scorching vCore values.
To achieve that operating frequency, Der8auer used an Elmor EVC2 controller and diagnostics chip, which, connected to a usually unpopulated pin area in the ASUS 970 PRO GAMING/AURA motherboard, allowed him to read-out everything that was running through the motherboard's VRM circuitry, and perform manual adjustments. Corsair Vengeance 2,666 MHz DDR3 memory was also used in the system. An accident happened along the way, though: when pulling AMD's stock cooler from the motherboard, the CPU remained attached to the cooler, which resulted in some bent pins (screams in horror). Luckily, things were fixed with a screwdriver - let that serve as a warning, alert, and tip, should this happen to you.Anyway, the AMD FX-8350 achieved an 8,127 MHz speed with a 1,920 vCore, which is an absolutely incredible voltage for a 32 nm CPU. Running at 7,500 MHz for a single-core performance benchmark, the CPU was pulling 100 W of power - for a single core to operate at that speed, mind you. Even so, the AMD FX-8350 only achieved a single-core score of 172 points - for comparison sake, AMD's six-core Ryzen 5 2600X, running at stock clocks of 3.6 GHz with all cores enabled, achieves 176 points in the same benchmark. Watch the video below for the full rundown on this experiment.
To achieve that operating frequency, Der8auer used an Elmor EVC2 controller and diagnostics chip, which, connected to a usually unpopulated pin area in the ASUS 970 PRO GAMING/AURA motherboard, allowed him to read-out everything that was running through the motherboard's VRM circuitry, and perform manual adjustments. Corsair Vengeance 2,666 MHz DDR3 memory was also used in the system. An accident happened along the way, though: when pulling AMD's stock cooler from the motherboard, the CPU remained attached to the cooler, which resulted in some bent pins (screams in horror). Luckily, things were fixed with a screwdriver - let that serve as a warning, alert, and tip, should this happen to you.Anyway, the AMD FX-8350 achieved an 8,127 MHz speed with a 1,920 vCore, which is an absolutely incredible voltage for a 32 nm CPU. Running at 7,500 MHz for a single-core performance benchmark, the CPU was pulling 100 W of power - for a single core to operate at that speed, mind you. Even so, the AMD FX-8350 only achieved a single-core score of 172 points - for comparison sake, AMD's six-core Ryzen 5 2600X, running at stock clocks of 3.6 GHz with all cores enabled, achieves 176 points in the same benchmark. Watch the video below for the full rundown on this experiment.
64 Comments on AMD FX-8350 Pushed to 8.1 GHz via Extreme Overclocking by Der8auer
I was certain this wasn't the first or fastest, it isn't 8.429Mhz is.
Thankfully I've not had this issue for many, many years. :)
Wants his channel to be just like Jayz and Linus'. $$$$
"But I'm NOT an entertainer, I'm only an engineer" :)
I wonder how much Voltage can 14 nm take
AMD has come a long way lol.
This was done shortly after release and isn't even the fastest! Why is this news? FTFY
LVT => Not a lot of voltage scaling potential.
RVT => A lot of voltage scaling potential.
HVT => Not meant for voltage scaling but leakage control and etc.
Bulldozer/Steamroller should have lower frequency per voltage step but more voltage steps that are worthwhile. While, Zen should have intrinsically higher frequency per voltage step but less voltage steps that are worthwhile.
12FDX retains voltage range from PDSOI/Bulk. Improves LVT scaling at high frequency and high voltage from 28nm. It adds and retains worthwhile modulation knobs; channel length(long length vs wimpy gates like Zen which add mask count and process steps), as well as body-biasing with forward(reduces Vt)/reverse(increases Vt) diodes, etc.
Do to Family 15h physical design, if it is a ported design, then it will probably want to target 12FDX for the 14nm/12nm node.
Family 17h(Zen) is physically built more like Family 16h(Jaguar);
Which means it can scale with higher performance transistors. While Bulldozer/Steamroller-like physical designs won't scale well on such nodes.
They all say LN2... but may not be (LHe is an option at hwbot). I know many of those were LN2 though.
For those few who know and care... I agree. While they were shit for daily use compared to Intel (and like 3 preior generations, lol), they hit the track like no other!!!
A lot of stuff in Bulldozer lost what it was initially based on. It was a beefy big core when it was only a 32-bit processor w/ units that would only pop eventually in Zen.
32-bit core => patents.google.com/patent/US6240503B1
L1i => 128 KB 2-way associative cache
L0i => 512 byte fully associative cache
L1d => 128 KB 2-way associative cache
{ 2x 32-bit ALUs
2x 32-bit FPUs
2x 32-bit MMXs
1x LD AGU
1x ST AGU ] x2
Shared Scheduler/Retirement => Independent Renamers
Similar core => patents.google.com/patent/US6553482B1
L1i => 256 KB 4-way associative cache
Retirement queue => 128-entries
Shared Retirement/Renamer => Independent Schedulers
Both shared L1d/LSU.