Yeah, marketing is tricky because there's so many model numbers and configurations.
The most important tidbit about "TDP" is that its "THERMAL design power", not actual power draw. So strictly speaking, TDP measures the size of the heatsink needed to keep the CPU functional.
This means that different chips, even with the same TDP, offer grossly different power consumption rates. Besides, every chip's actual power draw is slightly different, even chips of the same design (see "binning" and "silicon lottery").
-----------
When you understand how chips are made, then you see why things get complicated. Its a well known fact that the entire AMD Zen series uses only two die designs: Zeppelin, and a 2nd design with the iGPU.
That's right, the singular "Zeppelin" design covers AMD EPYC, AMD Ryzen3, 5, 7, and Threadripper (1900X, 1920X, 1950X).
How?? The difference between the dies is in "binning". If one or two cores are broken because of manufacturing defects, AMD sells it as a Ryzen 5 (6-core model) instead of the Ryzen7 (8-core model).
Very broken designs (3 or 4 broken cores) are sold as a Ryzen3 (4-core model). But when manufactured, these are all 8-core chips.
Between chips with full functionality between all eight-cores, some will be able to reach higher clocks than others. The 8-core designs are tested, and the ones that reach the highest clocks are 1800X, and the ones that reach the lowest clocks are sold as a 1700.
And that's how ONE design gets turned into 10 different SKUs sold to the customer. Because the manufacturing process is innately variable.
-------
Intel is known to bin for power-efficiency, high clocks and so forth. The marketing is a way to sell "broken chips" to people who don't care as much about high speeds or low-power draw.
> The most important tidbit about "TDP" is that its "THERMAL design power", not actual power draw. So strictly speaking, TDP measures the size of the heatsink needed to keep the CPU functional.
Your caps are misplaced here. The key word here is DESIGN power - inherently an approximation. It's more of a "power category" than an attempt at a rigorous measurement - really they're saying that CPU X needs to be paired with cooler Y. For example, a 6800K and a 6950X are both listed as a 140W TDP, but a 10-core processor is obviously going to pull more power than a 6-core processor under a full load.
Also, what you're measuring can significantly impact the number you get as a result. Measuring at all-core AVX, all-core base, single-core turbo, etc will all give you different numbers. There is frankly more marketing that goes into this than actual technical backing.
If you're implying that heat != power then no, that's incorrect. A CPU is essentially a nearly-perfect resistive load and all power that goes in is converted to heat.
AMD has attempted to spin this one in the past (I believe it was AMD_Robert or AMD_James who made a handwavey post about how TDP was not actually about electrical input but about heat output, as if they are somehow different), simply put they are being misleading. Turboing above average power consumption means that you need to reduce power later to average things out. Otherwise, you need to dissipate a greater amount of heat. That's why it's an average power measurement, and not a hard cap. And it certainly does not imply that "power in != heat out".
Yeah, simply put, thermal watts and electrical watts are the same thing. There is no power that goes into a CPU that is not converted into heat, and there is no heat that is not generated from the electrical input (or the ambient temperature of the room). AMD did not disprove the laws of thermodynamics.
The rest is some handwaving over a formula that he claims doesn't include power, but it's right there in the ϴca term (°C/W), the W is power. It's true that you can sprint above the cooling capacity of your cooler for a short time, but then the heatsink will heat up, and at some point you'll have to reduce power to compensate. Again, that's why it's an average and not a hard cap.
Pretty cringey stuff for an engineer to be spouting and it should not be repeated. Again, there is a grain of truth that TDP is typically not reported as an accurate, measured number but rather more of a general category, but the idea that electrical watts and thermal watts are different things is horseshit.
> How?? The difference between the dies is in "binning". If one or two cores are broken because of manufacturing defects, AMD sells it as a Ryzen 5 (6-core model) instead of the Ryzen7 (8-core model).
This is how virtually all CPUs do it, not unique to Ryzen. What is unique about Ryzen is that all products are constructed from a single die, from laptop to server products, whereas Intel has five: laptop, client, LCC server, HCC server, and XCC server. But within each die, everyone employs die-harvesting to increase yields.
> Very broken designs (3 or 4 broken cores) are sold as a Ryzen3 (4-core model). But when manufactured, these are all 8-core chips.
Side note, but there are typically not enough broken dies to fulfill all the demand for the very cheap SKUs, so many of these are actually fully-functional dies that are locked at the factory.
Back in the day, you used to get Phenom X3s that could be unlocked to the full 4 cores, and you also get GPUs that could have additional shaders unlocked (most recently on Fury, I believe).
There's some Ryzens that are sold with more cores enabled than they're supposed to have, but that is different because they come from the factory like that. Pretty sure it falls into the category of manufacturing errors - someone screwed up and didn't blow all the fuses they were supposed to.
The most important tidbit about "TDP" is that its "THERMAL design power", not actual power draw. So strictly speaking, TDP measures the size of the heatsink needed to keep the CPU functional.
This means that different chips, even with the same TDP, offer grossly different power consumption rates. Besides, every chip's actual power draw is slightly different, even chips of the same design (see "binning" and "silicon lottery").
-----------
When you understand how chips are made, then you see why things get complicated. Its a well known fact that the entire AMD Zen series uses only two die designs: Zeppelin, and a 2nd design with the iGPU.
That's right, the singular "Zeppelin" design covers AMD EPYC, AMD Ryzen3, 5, 7, and Threadripper (1900X, 1920X, 1950X).
How?? The difference between the dies is in "binning". If one or two cores are broken because of manufacturing defects, AMD sells it as a Ryzen 5 (6-core model) instead of the Ryzen7 (8-core model).
Very broken designs (3 or 4 broken cores) are sold as a Ryzen3 (4-core model). But when manufactured, these are all 8-core chips.
Between chips with full functionality between all eight-cores, some will be able to reach higher clocks than others. The 8-core designs are tested, and the ones that reach the highest clocks are 1800X, and the ones that reach the lowest clocks are sold as a 1700.
And that's how ONE design gets turned into 10 different SKUs sold to the customer. Because the manufacturing process is innately variable.
-------
Intel is known to bin for power-efficiency, high clocks and so forth. The marketing is a way to sell "broken chips" to people who don't care as much about high speeds or low-power draw.