Hardware & Technical Recommendations for a new Rig

It's not an ad hominem to point out someone isn't good at what they do, it's an assessment, an assessment I base on the fact that you stated empirically incorrect data. The 5900X is the fastest AMD processor without going up to the 5950X, which wasn't even in the discussion.

You miss tons of crucial details in your "Advice", first and foremost, that Clock Speed doesn't exist in a vacuum, the 5000 for example has around 20-30% more IPC uplift over Ryzen 3000, which means they could (But they aren't) clock 25% lower and still tie in gaming workloads. Also, what lower clock speed? The 5900X boosts to 4.8GHZ (4.95GHZ if you got a cooler to back it up) This is really driven home when the 5900X is beating the 10900K in many (but not all) games, despite the 10900K "Clocking Higher" on average.

Get your facts straight, and I'll retract my assessment. This isn't an attack on your character, I'm sure you're a perfectly nice, reasonable and intelligent person. You gave bad advice here.

1. An ad hominem is attacking the person rather than the argument - which by saying I am bad at what I do (pretty funny, since you don't even know what I do. It's a big industry.) rather than addressing any of the points I made, you are doing.

2. The 5900X uses two CCDs (6+6) rather than the single CCD that the 5800X (and 5600X) uses - meaning that there will be some latency introduced when cores on different CCDs communicate. I have seen this reduce performance in some applications which are not utilising many threads - which currently would include games. In regards to the lower clock speed, that was simply me mis-remembering specs as I was going off the top of my head, I'll admit to that. However I have seen a 5600X out-perform a 5900X, most likely to this CCD difference.

It's not going to be a massive difference, but it's worth taking into account if the machine is only going to be used for gaming.

3. This goes back to your first comment in this thread regarding SSDs, but I thought it was worth mentioning. Keep in mind that some manufacturers omit the DRAM cache altogether on their drives, tanking their performance quite hard. It's also worth keeping in mind the trade-offs made as you add more bits per cell.

4. I assume you'll want justification on my stance on the 3080Ti too. That one's purely economical. Performance bump vs. price bump. Makes more sense than a 3090 for gaming though.
 
1. An ad hominem is attacking the person rather than the argument - which by saying I am bad at what I do (pretty funny, since you don't even know what I do. It's a big industry.) rather than addressing any of the points I made, you are doing.

2. The 5900X uses two CCDs (6+6) rather than the single CCD that the 5800X (and 5600X) uses - meaning that there will be some latency introduced when cores on different CCDs communicate. I have seen this reduce performance in some applications which are not utilising many threads - which currently would include games. In regards to the lower clock speed, that was simply me mis-remembering specs as I was going off the top of my head, I'll admit to that. However I have seen a 5600X out-perform a 5900X, most likely to this CCD difference.

It's not going to be a massive difference, but it's worth taking into account if the machine is only going to be used for gaming.

3. This goes back to your first comment in this thread regarding SSDs, but I thought it was worth mentioning. Keep in mind that some manufacturers omit the DRAM cache altogether on their drives, tanking their performance quite hard. It's also worth keeping in mind the trade-offs made as you add more bits per cell.

4. I assume you'll want justification on my stance on the 3080Ti too. That one's purely economical. Performance bump vs. price bump. Makes more sense than a 3090 for gaming though.

Again, you get some things wrong, The number of CCDs doesn't matter, because on Zen 3, all CCDs have to go through Infinity Fabrick to the IO die. What you're saying was true on Ryzen 1 and 2, but not on Ryzen 3 and 4 when AMD started using a separate I/O die.

No PCIE SSD not using a DRAM cache is worth considering, they're rarely even competitively priced.

I wasn't attacking you as a person, I was attacking your ability. That's not an Ad Hominem. Your capability as a PC advisor was the subject of the argument. I asserted you are a bad one.

As for the 3080TI/3090, I disagree too, you undue your own argument here because you're living in a fantasy land where everything is MSRP. RIght now you can't find a 3080 for less than a 3080 TI because of LHR. And a 3090 is (Depending on model) 25-50% more than most 3080 TIs for 0-5% more performance in gaming. Now, you open up a lot of production work performance if you go for a 3090, so if that's an argument you want to make, okay. If you want a reasonably priced graphics card (3080 or 3070) there's usually a backorder waitlist even on System integrators. I'm not interested in what you could theoretically maybe possibly some day get in 2-3 months, I'm interested in what, if you absolutely need a PC today, you could get because everyone who has the option of waiting knows they need just ride out the storm for the next year, anyone with an expensive E-waste brick doesn't have that luxury.
 
Again, you get some things wrong, The number of CCDs doesn't matter, because on Zen 3, all CCDs have to go through Infinity Fabrick to the IO die. What you're saying was true on Ryzen 1 and 2, but not on Ryzen 3 and 4 when AMD started using a separate I/O die.
It is absolutely still the case with Ryzen 5000, particularly running on Windows. Communication between CCDs has to go through the IO die, which will introduce latency no matter how fast your CPU is. Normally happens when Windows decides that a running process should be moved to a different thread (which it seems to like doing).

No PCIE SSD not using a DRAM cache is worth considering, they're rarely even competitively priced.
I have been approaching this thread assuming that not everyone is aware of these facts, and thought that they should be presented so that people can make an informed decision. It is a fact that these SSDs are out there, and somebody just looking for a 1TB SSD may not be aware of this fact.

I wasn't attacking you as a person, I was attacking your ability. That's not an Ad Hominem. Your capability as a PC advisor was the subject of the argument. I asserted you are a bad one.
Then at best you came across as abrasive and quite aggressive. That is not how productive discourse works.

As for the 3080TI/3090, I disagree too, you undue your own argument here because you're living in a fantasy land where everything is MSRP. RIght now you can't find a 3080 for less than a 3080 TI because of LHR. And a 3090 is (Depending on model) 25-50% more than most 3080 TIs for 0-5% more performance in gaming. Now, you open up a lot of production work performance if you go for a 3090, so if that's an argument you want to make, okay.
I think you misunderstood my point here. Firstly, I am not talking about the price/perf difference between the 3080Ti and the 3090. I am talking about the 3080Ti and the 3080. I brought up the 3090 because in my opinion that card makes no sense for gaming. Production/compute loads (or anything that can take advantage of that VRAM really) are another matter altogether.

I also never suggested that you can find these cards at MSRP, hence why I suggested going through an SI rather than OP building the system themselves - at the moment, this has a higher chance of getting you a decent price, and the one I mentioned specifically shipped a system with a 5600X and 3070 at roughly MSRP (a bit added for labour etc obviously) within a few days.

In any case, we're derailing this thread a bit and it looks like OP's got a decent response by now. Feel free to DM me if you really feel like continuing this discussion, but for now I'll leave it at that.
 
Ok - my PC blew up and my Mac Bootcamp cant handle it any more. I know this might be a big thread - but…

Recommendations for a new pre-built rig and UK company?
I recommend 'OverclockersUK' whether for a complete pre-built or components.

 
The 5900X uses two CCDs (6+6) rather than the single CCD that the 5800X (and 5600X) uses - meaning that there will be some latency introduced when cores on different CCDs communicate. I have seen this reduce performance in some applications which are not utilising many threads - which currently would include games. In regards to the lower clock speed, that was simply me mis-remembering specs as I was going off the top of my head, I'll admit to that. However I have seen a 5600X out-perform a 5900X, most likely to this CCD difference.

It's not going to be a massive difference, but it's worth taking into account if the machine is only going to be used for gaming.

The dual-CCD Zen 3 parts tend to be slightly faster, even at gaming, than the single CCD parts. The reason for this is that there are more good cores to choose from and the boost clocks are higher. Scheduling a single application across different CCDs is uncommon for a properly configured system, unless that application is very well threaded, in which case the greater number of cores, larger caches, and higher boost clocks are likely to help more than occasional inter-core latency issues hurt.

You might find a few outliers, even after account for bad settings or broken schedulers, but by and large, the 5900X and 5950X are going to slightly outperform the 5600X or even 5800X, even in dedicated gaming systems.

Personally, I think the 5900X is the best value for a general purpose higher-end setup, but I'd probably recommend a 5600X or a 10th gen Intel for a more budget focused, gaming oriented build. I have a 5800X in my gaming setup, but only because the 5900X was hard to find when I built it.

Again, you get some things wrong, The number of CCDs doesn't matter, because on Zen 3, all CCDs have to go through Infinity Fabrick to the IO die. What you're saying was true on Ryzen 1 and 2, but not on Ryzen 3 and 4 when AMD started using a separate I/O die.

Number of CCDs still matters on Zen 3. Cores in the same CCX (which is the same as a CCD in Zen 3) don't need to talk to each other over Fabric, they use the L3. This is why you see very low inter-core latency within the same CCX.

Zen 2 (Ryzen 3000) was even more dependent on fabric clocks because the maximum number of cores per CCX was 4 (rather than 8 in Zen 3). Any communication outside the CCX requires going through the I/O die. A single CCD Zen 3 part only has one CCX so the cores never need to reach out to the I/O-die to talk to each other, only to access memory or I/O.

Examples: https://www.anandtech.com/show/1621...e-review-5950x-5900x-5800x-and-5700x-tested/5
 
The dual-CCD Zen 3 parts tend to be slightly faster, even at gaming, than the single CCD parts. The reason for this is that there are more good cores to choose from and the boost clocks are higher. Scheduling a single application across different CCDs is uncommon for a properly configured system, unless that application is very well threaded, in which case the greater number of cores, larger caches, and higher boost clocks are likely to help more than occasional inter-core latency issues hurt.

You might find a few outliers, even after account for bad settings or broken schedulers, but by and large, the 5900X and 5950X are going to slightly outperform the 5600X or even 5800X, even in dedicated gaming systems.

Personally, I think the 5900X is the best value for a general purpose higher-end setup, but I'd probably recommend a 5600X or a 10th gen Intel for a more budget focused, gaming oriented build. I have a 5800X in my gaming setup, but only because the 5900X was hard to find when I built it.



Number of CCDs still matters on Zen 3. Cores in the same CCX (which is the same as a CCD in Zen 3) don't need to talk to each other over Fabric, they use the L3. This is why you see very low inter-core latency within the same CCX.

Zen 2 (Ryzen 3000) was even more dependent on fabric clocks because the maximum number of cores per CCX was 4 (rather than 8 in Zen 3). Any communication outside the CCX requires going through the I/O die. A single CCD Zen 3 part only has one CCX so the cores never need to reach out to the I/O-die to talk to each other, only to access memory or I/O.

Examples: https://www.anandtech.com/show/1621...e-review-5950x-5900x-5800x-and-5700x-tested/5
Cheers for clarifying the CCD/CCX stuff, that article's pretty interesting. I agree overall with your recommendation; the 5900X is a damn powerful chip, but unnecessary for just gaming really.

I wonder if my odd results with the 5900X/5600X could be due to the fact that I was running tests in a KVM, with the least performant core in each case left attached to the host OS. Windows does do some odd scheduling stuff for me, but I haven't installed it on bare metal in years.
 
Been reading with interest about CCD CCX stuff. Gosh been a long time since I looked properly at chips. Last time I looked properly at processors was the Z80 and Arm - been a long long time.
 
Back
Top Bottom