Question Difference of 5950X and 5900X, architecturally speaking?

[Iver Hicarte]

We all know that on paper the 5950X has four more cores and twice the threads compared to the 5900x, but when it comes to how each core was architectured, engineered and made, is there a difference there? Because if the only difference is that AMD decided to just add four more physical cores that pretty much resembles the cores that the 5900X has, then it's just a waste of money . So maybe the difference lies on how each core was made, maybe the 5950X not only packs four more cores and twice the threads, but maybe the 5950X's cores are made differently compared to the 5900X's cores? Can someone shed some light into this, it's been a futile hunt with Google concerning this topic, so might as well ask around this forum with knowledgeable people.

 

[hotaru.hino]

The cores on a 5950X are no different than a 5900X.

Non-APU Ryzen since the 3000 series have two main types of die: the I/O die (IOD) and the Core Complex Die (CCD). Each CCD was built to have 8-cores, whether or not the file product will use them. So with that said, a 5600X uses 1 CCD, just that two cores are disabled. A 5700/5800(X) is 1 CCD with all 8 cores. A 5900X is 2 CCDs with 2 cores disabled in each. A 5950X is 2 CCDs with all 8 cores. AMD did it this way to simplify manufacturing costs.

 

[Math Geek]

it's not a waste of money if you NEED the extra cores in your usage of the PC.

it is a waste if you are only doing stuff that does not need all the cores. the 5900x is also a waste if you aren't using all the cores. you got a 5600x option which is more than ample for most users.

every cpu is made the way described above and has been for a long time. make a 4 core cpu with hyperthreading and it's a 4790k. test it and see that the hyperthreading is not working, so disable it and it's a 4690k with 4 cores.

test them some more and find some won't clock as high so it's now a 4690 (non k) or other lesser model with slower clock speeds. and so on and so on. disable 2 cores and it's a 2 core celeron or pentium model

we just got more cores to work with now so they can disable a lot more and make a wider variety of products from the same samples depending on how they test and needs at the moment.

it's the most cost effective way to utilize all the chips being made. no need to throw away the whole lot if one ccd is sketchy. disable it and it's now up to an 8 core chip. and so on and so on.

no conspiracy here, just how it's always done. it's not their fault people keep insisting that somehow a 5900x is better than a 5600x simply due to the bigger number. still the same cores on each with the same performance. you just get a few more if you need them.

same with the 13900k. it's not needed either for most people but you won't find anyone shying away from it when their e-pride is talking either

 

[drea.drechsler]

We all know that on paper the 5950X has four more cores and twice the threads compared to the 5900x, but when it comes to how each core was architectured, engineered and made, is there a difference there? Because if the only difference is that AMD decided to just add four more physical cores that pretty much resembles the cores that the 5900X has, then it's just a waste of money . So maybe the difference lies on how each core was made, maybe the 5950X not only packs four more cores and twice the threads, but maybe the 5950X's cores are made differently compared to the 5900X's cores? Can someone shed some light into this, it's been a futile hunt with Google concerning this topic, so might as well ask around this forum with knowledgeable people.

All of the Ryzen CPU's (as well as Threadripper and Epyc server CPU's) since Zen 2 (3000 series) are sourced from the exact same CPU dies made at TSMC. The only differences is the bin sort where they look for performance characteristics that inevitably result from variations during the foundry processes...and the boost algorithm used.

Too, they only make 8 core/16 thread CPU dies so 16 core and 12 core chips have two dies. In the case of dies with 6 CPU cores (Ryzen 5 part, for instance) they fuse off two complete cores/four threads so 6 remain. In the case of 5900X, two 6 core/12 thread dies make up the 12 core/24 thread CPU. I'm sure you picked up on that in your research though. Ideally, they're only fusing off cores that would be unsuitable for any other CPU, but you never know.

Aside from that, each part gets a boost algorithm that leverages the unique characteristics of the silicon that the bin sort identified when a chip is directed to it's product family. And, as well, the algorithm considers the makeup of the CPU: 8 cores on one die has a higher thermal density than 6 cores, for instance. But the OS gets in the game too with an architecture-aware scheduler that can spread active processes across multiple dies to level out the heat load. But it seems to me the combination of bin sort and boost algorithm is AMD's way of exploiting the manufacturing margins in the desktop CPU's that overclockers customarily would.

Depending on what they get out of TSMC's foundry it may very well be a complete waste of money (or profit, more correctly) for AMD to make any desktop CPU's at all. That's because the same dies that go into them also go into vastly higher priced Epyc server chips and Threadripper HEDT chips. So why not just serve those markets at higher margins? Server chips in particular are hard to get hold of as I understand. It probably depends on how robust the process is, and therefore how many die meet the very critical requirements to be a super dense core count CPU, running at low voltage, low power demand on all 8 cores.

 

[DSzymborski]

We all know that on paper the 5950X has four more cores and twice the threads compared to the 5900x, but when it comes to how each core was architectured, engineered and made, is there a difference there? Because if the only difference is that AMD decided to just add four more physical cores that pretty much resembles the cores that the 5900X has, then it's just a waste of money . So maybe the difference lies on how each core was made, maybe the 5950X not only packs four more cores and twice the threads, but maybe the 5950X's cores are made differently compared to the 5900X's cores? Can someone shed some light into this, it's been a futile hunt with Google concerning this topic, so might as well ask around this forum with knowledgeable people.

They don't add cores to anything in this way. The creation of differentiated CPUs (and RAM) is a binned process.

To put it in the most simple terms possible, a 5900X is simply a 5950X that was too defective to sell as a 5950X.

A stick of 2666 MHz memory is just memory that was too defective to sell at 3000 MHz or 3400 Mhz (etc.)

Now, sometimes, they'll fuse off these areas for market segmentation issues. If you need a bunch of 5600Xs and you don't have enough defective 5950Xs, then you have to make some.

You saw a good example of this in the old days. AMD didn't have enough defective chips to make the tri-core Athlon II X3s that the market demanded so they repurposed some good CPUs in addition to the defective ones to serve that need, but at the time didn't effectively block a non-defective core. So it was common practice then to try and unlock that fourth core and see if you had one that had an actually defective fourth core, and a lot of people ended up with Athlon II X4s for the Athlon II X3 price.

 

[Viking2121]

Same cores, same every thing as far as I can dig into it just the 5900x has less of it and less L2, 5950x has a lower base I would imagine to keep TDP at 105watts, but thats not really going to matter for performance with PBO. But thats it, Better off trying to figure out the the difference between how different generations of Ryzen works than comparing 2 CPU's that are the same generation and with practically the same CCD layout.

There will be slight difference worth noting on a CPU with multiple CCD's like the 5950x vs a 5800x with 1CCD, though I do think I read that some 5800x's have 2 CCD's, that might be looking up and reading about it, I would imagine it would be slightly slower in some latency sensitive things vs a 5800x with CCD.