Quote from: Anonymousgg on March 23, 2022, 18:05:08
5900X has a higher base clock, 5950X has a higher boost clock and better binning from what I remember. I looked at AnandTech's Nov 2020 benchmarks and they had them performing virtually the same, but 5950X should have the edge in any scenario where 8 cores from the same chiplet can be used.

AMD wants comparisons to the 5900X since more gamers bought that chip than the 5950X and they will cost around the same now.

All of which is irrelevant. It is factually incorrect to refer to 5900X as top-of-the-line. I don't think this reference comes from AMD. This isn't the first time an author here has apparently forgotten about the 5950X. Yes, the 5900X was better value.

It has lower base frequency simply because it doesn't have high enough base power budget to account for the larger number of cores. It's as simple as that. If your workload benefits from the extra four cores, 5950X will be better. And if not, it shouldn't perform any worse; it will just be poorer value.

Quote from: _MT_ on March 23, 2022, 10:56:27
Top-of-the-line is 5950X, not 5900X.

5900X has a higher base clock, 5950X has a higher boost clock and better binning from what I remember. I looked at AnandTech's Nov 2020 benchmarks and they had them performing virtually the same, but 5950X should have the edge in any scenario where 8 cores from the same chiplet can be used.

AMD wants comparisons to the 5900X since more gamers bought that chip than the 5950X and they will cost around the same now.

Quote from: vertigo on March 23, 2022, 05:31:01
Coming that close despite a 300MHz lower base and turbo and a lower voltage is pretty good, especially considering the restraints mentioned. That indicates it would likely be a fair bit faster at the same speeds, and that for equal performance it's going to require less power, which is great for laptops. And, of course, it's a new technology, so will likely improve with time. I'd say it looks promising, even if maybe not quite as much as hyped.

"Faster" at what? Cache only improves latency when fetching data from memory. You work at the same pace, you just idle less. To benefit from making a cache larger, active dataset must be larger than the old cache. Yes, there are applications that will benefit from it. The question is, however, will your applications benefit from it?

Understandably, you see a much larger effect in a multi-threaded benchmark because it's a shared cache. The more threads are running, the bigger the active dataset, the more likely it is that the smaller cache will be saturated.

When you have a code that does heavy lifting, it pays to optimize it and part of it is making it cache-friendly. Meaning, you don't want to be jumping all over the memory. When you fetch a chunk of data from memory, you want to do everything you need with it then and there before dumping it and fetching another. If at all possible. Instead of returning to it repeatedly. Or at least access data in such a way that predictor works well and can pre-fetch it in the background. Of course, the bigger the cache you're optimizing for, the more manoeuvring space you have to work with. But L3 is relatively far away. Therefore, such algorithms won't see much improvement. They're designed to minimize idling. The problem is that many developers are too lazy these days. :-) When it's a fun exercise.

Top-of-the-line is 5950X, not 5900X.

When will this Geekbench nonsense stop? Geekbench is and never was a good indicator.

Coming that close despite a 300MHz lower base and turbo and a lower voltage is pretty good, especially considering the restraints mentioned. That indicates it would likely be a fair bit faster at the same speeds, and that for equal performance it's going to require less power, which is great for laptops. And, of course, it's a new technology, so will likely improve with time. I'd say it looks promising, even if maybe not quite as much as hyped.