• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

Ryzen Threadripper 3970X Overclocked to 5.75 GHz

Well i think it is correct, these are 7nm chips and heat density is a huge issue on these things. The lower you can put the voltage, the lower it's consumption will be. Add LN2 on top of that and the power requirement for the chip drops as well.
I agree on the fundamentals. But I dont care how cold it is, a chip that runs 4.3 GHz at 1.3V or greater, cant run 5.7 GHz at .2 less V even under ln2...not even close.

Here's a link from hwbot with similar clocks showing a more appropriate voltage for the cpu and clocks. You are drastically overestimating the value of the die shrink compared to how these cpus scale. They are pretty much tapped as is.
 
I cant find a proper article that goes into all that, but ive found some interesting stuff:

Furthermore, there’s a relationship between CPU temperature and power consumption; a chip clocked at 5.5GHz with liquid nitrogen uses less power than a hypothetical CPU clocked at that speed with a conventional air-cooler. The fact that a 1.65x increase in clock (from 3.4 – 5.6GHz) send power consumption flying upwards to over a kilowatt, in an ideal scenario that intrinsically lowered the CPU’s power consumption by reducing resistance within the core, is evidence for why Intel and AMD don’t pursue high clock speeds the way they used to. Past a certain point, it’s no longer possible to move heat out of the die quickly enough with conventional air or even water cooling. There have been various proposals to move heat out of CPUs more effectively, from micro-channels in thermal paste, to microfluidic cooling, to introducing cubic boron arsenide as a heat-transporting material.

Source: https://www.extremetech.com/computing/256623-liquid-nitrogen-drives-18-core-core-i9-7980xe-6ghz

I think derbauer once showed a video in where the power requirement would drop once going below zero in temps. Remember this all core OC is a working one (thats stable) and not a suicide run with highest possible clock on whatever voltage you throw at it.
 
I cant find a proper article that goes into all that, but ive found some interesting stuff:



Source: https://www.extremetech.com/computing/256623-liquid-nitrogen-drives-18-core-core-i9-7980xe-6ghz

I think derbauer once showed a video in where the power requirement would drop once going below zero in temps. Remember this all core OC is a working one (thats stable) and not a suicide run with highest possible clock on whatever voltage you throw at it.
That article says power requirements go up significantly.. it doesnt do that by lowering voltage. But yes, at the same voltage you can get a higher clock by going cold.
 
nice to see what the ic is really capable of :thumb:
 
That's my issue with the AMD design approach. Let's just increase the FPU EUs, uOP cache, L3 cache etc, no sophistication just brute force thanks to the process shrink. Look at Sunny Coves 18% uplift through more conservative but intelligent changes and the 10th gen focus on a performance shift on 4/6 cores at an entirely new power envelope. AMD is relentlessly pushing an archaic design scheme all the way to zen 4 and 5nm

I can't blame their chiplet/core focus because they need the server money. Nice of them to benefit consumers as well. The 3600 is a very good chip.
I feel like Ryzen has gone thru far more "sophisticated" changes than just about all of Intel's CPU's that came after C2Q up to today. I mean unless you count the integrated memory controller change, but AMD already had that so Intel was simply playing catch up in that area. I know one thing a 6 to 8 core CPU today thanks to Ryzen is cheaper than Intel was milking consumers for a quad core prior.
 
the clock deficit don't effect the per core performance, zen-3000 cores 4200 mhz it's equal to 9900KH at 5000 mhz, mybe becuase it have more than 10 times more cache xD and a few other thing like improve latencies
what is mega lol is the name starship xD with this thing you can realy build your own NASA :)


you have to understand it not possible what you whant :x the mosfet, caps and even resistor and even the conduits iteselfs would burn out from this :x that the problem
if you prime95 a processor like this (even old version without AVX) it will blow, on 4200 it reach 74 C with water, what you think hapen with this more than 5000 ?
well thats why u have VRMs capable of over 1000W lol
 
Back
Top