Ryzen Owners Zen Garden

[freeagent]

My bet is on the first possibility.

Yup.

The dominant CCD will always be a couple of C apart sometimes a little more.

 

[AusWolf]

Yup.

The dominant CCD will always be a couple of C apart sometimes a little more.

What I don't understand, though, is why the two dominant cores aren't always on the dominant CCD. If core preference was like [CCD1: cores #0-7, CCD2: cores #8-15], it would improve lightly threaded and gaming performance a lot, imo, as it would negate the need for high-latency communication between the two CCDs, not to mention the clock speed difference on Zen 4 CPUs.

 

[freeagent]

What I don't understand, though, is why the two dominant cores aren't always on the dominant CCD.

I think it’s to help keep the cpu cool tbh.

 

[AusWolf]

I think it’s to help keep the cpu cool tbh.

Either that, or AMD wants both preferred cores to have access to a full 32 MB L3 cache each, even if it comes with some latency penalty during communication between the two CCDs. I still think it would be better if both of those cores were on the same CCD, but we'll never know if I'm right or wrong, I guess.

 

[Mussels]

I enabled PBO on my 7900, I noticed one of the CCDs runs considerably hotter and at slightly lower clock speeds, actually it's one core in particular #5 which seems to get pretty hot compared to all other. This is kind of odd, anyone else with a 7000 series noticed something like this ? I also noticed windows seems to really like to schedule everything on that core, which is bizarre but maybe that's also why it runs so much hotter.
View attachment 280797

That die has worse contact with your cooler, is the most obvious answer

Either that, or AMD wants both preferred cores to have access to a full 32 MB L3 cache each, even if it comes with some latency penalty during communication between the two CCDs. I still think it would be better if both of those cores were on the same CCD, but we'll never know if I'm right or wrong, I guess.

you can turn off preferred cores in the BIOS hidden away somewhere, and see if gaming performance changes?

 

[AusWolf]

you can turn off preferred cores in the BIOS hidden away somewhere, and see if gaming performance changes?

Really? Where? I haven't found anything like that so far.

 

[Vya Domus]

That die has worse contact with your cooler, is the most obvious answer

Then how come that the coolest core is also on that die ? I've watched a gamer nexus review and they found the same huge deltas between cores.

 

[freeagent]

So yesterday I dropped my X3D into my kids system thinking it would be a solid bump for him, coming from a 5600X.. he didn’t even notice lol.. so I was a bit chuffed and I gave him back his 5600X today lol

Still doesn’t notice

 

[AusWolf]

So yesterday I dropped my X3D into my kids system thinking it would be a solid bump for him, coming from a 5600X.. he didn’t even notice lol.. so I was a bit chuffed and I gave him back his 5600X today lol

Still doesn’t notice

That's what I mean when I say that most of the differences of modern hardware can only be measured in a benchmark. If you're a casual gamer, it doesn't matter if you have a 3700X or 11600K or 5800X3D.

 

[Veseleil]

That's what I mean when I say that most of the differences of modern hardware can only be measured in a benchmark. If you're a casual gamer, it doesn't matter if you have a 3700X or 11600K or 5800X3D.

I love to play benchmarks. Even though I know the lore, and the quests are mostly of a fetch type, I kinda always hope for a different ending.

 

[Braegnok]

What I don't understand, though, is why the two dominant cores aren't always on the dominant CCD. If core preference was like [CCD1: cores #0-7, CCD2: cores #8-15], it would improve lightly threaded and gaming performance a lot, imo, as it would negate the need for high-latency communication between the two CCDs, not to mention the clock speed difference on Zen 4 CPUs.

On my chip CCD 00 is the dominant CCD with core4 SP 121 and core5 SP 122 being the fastest,.. on CCD 01 all the cores are SP 113.



With Asus BIOS you can go to Advanced, AMD Overclocking, Manual CPU Overclocking, CPU Core Count Control,.. and use the Bit Map Down Core Control to disable cores on CCD 00 and CCD 01,..

Changing 1-1-1-1-0-0-0-0 disables 4-cores, you need to disable same on both CCD's, and you must disable from right towards the left as shown, exit and save.



I tested disabling 4-cores on each CCD in Core Control vs disabling CCD 01 in Ryzen Master,.. you will take a hit on L3 if you disable a CCD.

Personally running on CCD 00 the best 8-cores with CCD 01 disabled is the better option.

CCD 01 Disabled:

 

[freeagent]

That's what I mean when I say that most of the differences of modern hardware can only be measured in a benchmark. If you're a casual gamer, it doesn't matter if you have a 3700X or 11600K or 5800X3D.

Yup that is why I am always bouncing between my X3D and 5900X.

Honestly.. out of my 3 chips my 5600X is probably still my favorite.

 

[Space Lynx]

Yup that is why I am always bouncing between my X3D and 5900X.

Honestly.. out of my 3 chips my 5600X is probably still my favorite.

I like my 5600, its an end of life silicon since its a new chip, the manufacturing process is at its peak for this chip (thats what I hear anyway). its insane that it does 4.7 all core at 1.230 volts (LLC makes it 1.250v) original launch day chips needed 1.35v or higher to do that... kind of wild how much a few years of improvements make to a node.

 

[freeagent]

Yup 1.35 for 4700. My launch 5600X is stable at 2000fclk and can bench at 2100fclk, she’s a beast

 

[Space Lynx]

Yup 1.35 for 4700. My launch 5600X is stable at 2000fclk and can bench at 2100fclk, she’s a beast

i did a bunch of tests to see its limits on this fine tuned silicon, at 1.175v it can do 4.5 all core and pass everything across the board. absolutely nuts. i still run it at the 4.7 1.230 cause its still running cold, so meh.

 

[tabascosauz]

I enabled PBO on my 7900, I noticed one of the CCDs runs considerably hotter and at slightly lower clock speeds, actually it's one core in particular #5 which seems to get pretty hot compared to all other. This is kind of odd, anyone else with a 7000 series noticed something like this ? I also noticed windows seems to really like to schedule everything on that core, which is bizarre but maybe that's also why it runs so much hotter.

AMD just seems to have a harder time ensuring small core deltas on 2CCD CPUs. Plenty of examples from past generations, I had a CCD2 core on 5900X that was something like 15C hotter than some CCD1 cores.

The more per-core power you push, the bigger the delta gets. Kinda like the edge-hotspot difference on GPUs.

And since Tctl/Tdie is a hotspot sensor in function, that's where your temps come from.

This kinda continues to confirm my theory that Windows 11 has been treating 2CCD CPUs like big.little parts since Ryzen 5000, always one random CCD2 core that is picked for background tasks. Not sure there is any method to the madness either, mine was my worst core. I guess it works out for the upcoming 7000X3D though

 

[INSTG8R]

So yesterday I dropped my X3D into my kids system thinking it would be a solid bump for him, coming from a 5600X.. he didn’t even notice lol.. so I was a bit chuffed and I gave him back his 5600X today lol

Still doesn’t notice

I mean I went from that to that. Not gonna lie just the difference in CPU score in 3DMark…

 

[AusWolf]

On my chip CCD 00 is the dominant CCD with core4 SP 121 and core5 SP 122 being the fastest,.. on CCD 01 all the cores are SP 113.

View attachment 280989

With Asus BIOS you can go to Advanced, AMD Overclocking, Manual CPU Overclocking, CPU Core Count Control,.. and use the Bit Map Down Core Control to disable cores on CCD 00 and CCD 01,..

Changing 1-1-1-1-0-0-0-0 disables 4-cores, you need to disable same on both CCD's, and you must disable from right towards the left as shown, exit and save.

View attachment 280990

I tested disabling 4-cores on each CCD in Core Control vs disabling CCD 01 in Ryzen Master,.. you will take a hit on L3 if you disable a CCD.

Personally running on CCD 00 the best 8-cores with CCD 01 disabled is the better option.

CCD 01 Disabled:
View attachment 280992

AMD just seems to have a harder time ensuring small core deltas on 2CCD CPUs. Plenty of examples from past generations, I had a CCD2 core on 5900X that was something like 15C hotter than some CCD1 cores.

The more per-core power you push, the bigger the delta gets. Kinda like the edge-hotspot difference on GPUs.

And since Tctl/Tdie is a hotspot sensor in function, that's where your temps come from.

This kinda continues to confirm my theory that Windows 11 has been treating 2CCD CPUs like big.little parts since Ryzen 5000, always one random CCD2 core that is picked for background tasks. Not sure there is any method to the madness either, mine was my worst core. I guess it works out for the upcoming 7000X3D though

From another user in the Core i9-13900K thread who has a 7900X:

It's clearly visible that cores #1, #2 and #6 get the highest load, so two of these must be the dominant/preferred ones, confirming the CCD = bigLITTLE theory. But then:

Ryzen Master shows cores #2 and #10 as being the dominant ones, with #1 and #8 following. So um... what?

 

[tabascosauz]

From another user in the Core i9-13900K thread who has a 7900X:

It's clearly visible that cores #1, #2 and #6 get the highest load, so two of these must be the dominant/preferred ones, confirming the CCD = bigLITTLE theory. But then:

Ryzen Master shows cores #2 and #10 as being the dominant ones, with #1 and #8 following. So um... what?

No, that disproves my theory...lol. 0, 1, and 5 are all on CCD1. The point is that W11 sometimes picks a specific core on CCD2, which isn't necessarily of particularly good quality or anything that makes it ideal to be picked. In contrast, W10 is very, very strict about keeping loads within CCD1.

Ryzen Master's dots and stars are per CCX only. Most likely Core 0 and 1 are CPPC Preferred across the entire CPU, while Core 7 and 9 are just the best cores on CCD2.

If you check HWInfo reporting of both core rankings, I'd bet that in both lists all 6 CCD1 cores are still ranked ahead of all 6 CCD2 cores, like it's always been since 2019. In that time I have never seen any 2CCD Matisse, Vermeer or Raphael CPU break that rule.

The rankings don't always correlate with outright SP. CCD2 cores are sometimes of better SP than some CCD1 cores, and some CCD2 cores can outperform CCD1 cores on pure clock especially after CO (ie. corecycler).

But even if they are better quality, they are still borderline worthless due to the restrictions of the design. CCD2 cores should never be used in games due to cross-CCX penalty, you can't really single out CCD2 cores (except with process lasso type magic where it works) due to inability to user-edit CPPC or scheduler rankings - they will never be picked because Preferred cores are always on CCD1.

Essentially, 2CCD owners had E-cores before they were cool .......but without the cache advantage of those cores

 

[freeagent]

I mean I went from that to that. Not gonna lie just the difference in CPU score in 2DMark…

Roblox doesn’t take advantage of the extra cache

 

[Vya Domus]

This kinda continues to confirm my theory that Windows 11 has been treating 2CCD CPUs like big.little parts since Ryzen 5000, always one random CCD2 core that is picked for background tasks. Not sure there is any method to the madness either, mine was my worst core. I guess it works out for the upcoming 7000X3D though

Windows schedules everything on core 5 in my case, this is on CCD1 which runs the hottest and also clocks the highest. It looks like one CCD always clocks higher than the other, actually looking at other reports from people it's only that CCD that the hits advertised highest single core clocks.

I think they're just binned differently, I can't think of any other reason why AMD would set up one CCD to always clock lower for some reason. It's most likely the case that they use 1 CCD of higher quality and 1 of lower quality. I am sure of this because with PBO enabled some cores on CCD2 now also hit the maximum boost clocks of 5.4 though not all, which suggests that they're just of worse quality.

By the way it could be the case that it's set up so that the best CCD is always "CCD1".

 

[trparky]

That's what I mean when I say that most of the differences of modern hardware can only be measured in a benchmark. If you're a casual gamer, it doesn't matter if you have a 3700X or 11600K or 5800X3D.

That's essentially why I don't upgrade every year. I upgrade every four to five years. Granted, since I have an AM5 motherboard I'm not as limited as I was in the past when it comes to upgrades, I may do more frequent upgrades especially if AMD says that their next CPU is going to feature 27% more performance. I'll just give my 7700X to my father and get the new chip as a drop-in replacement.

 

[AusWolf]

Ryzen Master's dots and stars are per CCX only. Most likely Core 0 and 1 are CPPC Preferred across the entire CPU, while Core 7 and 9 are just the best cores on CCD2.

If you check HWInfo reporting of both core rankings, I'd bet that in both lists all 6 CCD1 cores are still ranked ahead of all 6 CCD2 cores, like it's always been since 2019. In that time I have never seen any 2CCD Matisse, Vermeer or Raphael CPU break that rule.

Ah! That makes sense if it's the case.

 

[trparky]

I thought new Ryzen chips weren't vulnerable to CVE-2018-3639, Speculative Store Bypass. But according to a Powershell tool that I used, that doesn’t seems to be the case.

 

[mechtech]

A question for amd gigabyte boards. I have all old 3pin voltage fans, which used to ramp up and down all over all the time, until I flattened the curve as per Mussels recommendation.

Just curious, if PWM fans would be less susceptible to this or not?