That ASRock x570 Steel Legend sensor must be in a really terrible place to be so far off the mark.
That is the big problem. Been discussing this with
@ir_cow and motherboard reviewers on how they test VRMs and the answer came back that it's bascially a nightmare
Some VRM's can safely run at 120C, others thermal throttle at 80C.
Some can handle 50A at any temp, others are 80A at 25C and 20A at 80C.
On many boards, the hotspots are where you expect - others it must seep away through the traces in the PCB and heat up seemingly unrelated areas, especially in a blank spot between those heat sources where that heat may simply accumulate due to lack of airflow
@ir_cow made the silly mistake of getting me thinking on how to judge motherboards completely and fairly, so here's an in-depth teardown on that board above and why so many AM4 users had issues they couldn't resolve without a board swap.
This is that board with it's high VRM temps and shoddy marketing, but tended to be actually liked by it's owners. How can a high VRM temp board with weird claims not be trash?
I mean honestly, look at the marketing for this - DDR4 4666+ on AM4?
10 phase power design (later called 8+2) that's downright false
We've got a dedicated thread to trying to get people to 4000 stable, let alone that huge number.
Quad crossfire?? I'm sensing a basic math problem with that
VRM's look beefy, but its a basic 4+1 setup (general rule is inductors divided by 2)
They say 8+2, but this is when people started to get annoyed if they failed to mention doublers in marketing
Next slide is about "premium" 60A chokes
Cheating a little here since someone else identified the parts for me but throwing in my interpretations on top.
A quick google on this board shows it has a PWM controller that can only do 7 phases
That article covers that it's been used in a 4+2 (doubled) setup
The x570 Taichi that people loved had the same VRM's but used
10 of them (and four weaker ones for the SoC)in a 6+2 setup covered here on TPU.
Problem is that most VRM's are a rated a little dishonestly as Asrock says they're 60A,
but the spec sheet says 50A -
with 55A burst capabilities for only 10 milliseconds while the spec sheets show a realistic limit of 45A before they start pumping out heat
These are at least safe to high temperatures, showing they can handle upto 110C problem free - as long as you're under 50A. Full power, always? Nice.
The efficiency chart shows why that limit exists - because passing 20A, the sweet spot is gone - and past 30A you're at 15% of your power thrown to heat.
AM4's default PBO *on* limits are
TDC: 95A and EDC: 142A with PPT: 142W
(At some point an "EDC" bug was introduced in AGESA that stopped boards passing these amperage limits, my guess as a failed safety limit - that's been corrected)
8 of those VRM's at AM4's stock 142A with a 5800x would be 17.75A each, quite happily in that sweet spot... before board makers bad decisions get involved with BIOS settings.
5800x uses the same power limit as 5950x with double the cores - so in theory if nothing like thermals or BIOS settings limited you a 5950x could use double the values of a 5800x.
(Same voltage, double amperage and wattage to twice as many cores)
My 5800x on an Aourus B550-i ITX ran at 130A in R20 and R23, while kitgurus review of a 5950x with unlimited PBO went upto 215A - close to that theory
From an engineering perspective, that board gets a pass - that'd be very hard to overheat to 120C, and the VRM's should in theory handle even a power unlocked 5950x.
215A / 8 = 26.875W, within the VRM's happy place let alone their 50A maximum.
But you see 90C and you think oh god it's bad, because OTHER vrm's overheat at 80C or less
Ironically this chart shows an x570 steel legend with another 32c left on it's VRMs going by it's sensor readings while some of these others may be toeing the line to overheating or already throttling the CPU
As another metric that's ignored, the inefficiency of designs can lead to extra power wasted as heat - a 30W difference is huge for a small CPU like a Ryzen 2600
I get the feeling the checklist to actually test boards is a relatively large one with everything needing to be tested simultaneously to be useful
1. VRM specs, especially the temperature they start to lose efficiency
2. All testing done with the highest power consumption CPU you can get, to find a boards actual limits
3. Also testing within recommended limits - AMD's stock PBO and Intels PL1/PL2 targets.
R23 test runs would need all the relevant data slapped together in one graph somehow:
- CPU's effective sustained clock speed (indicator of throttling)
R23 score (performance loss vs competitors)
- Maximum amps drawn by CPU (HWinfo)
- Wattage used by system (Wall reading)
- VRM temps (and their rated maximums)
I also just ran into an issue on my ITX system where "PBO disabled" boosts harder than PBO with maxed out settings because ?!?!?
Crap like that is why people have issues with 90c at stock
