Silly max memory speed question

Associate
Joined
16 Jun 2014
Posts
395
The AMD Ryzen 9 7900X says it will only support 5200 memory max. But it also says it has EXPO features on it to overclock the ram.

So assuming my mainboard says it will run faster memory will putting this kind of 6000 memory in actually be worth while or should I just stick to 5200 memory?

The AMD site says Max memory speed

2x1r 5200
2x2r 5200
4x1r 3600
4x2r 3600

Which suggests it can use ram at 5200 if you use all the memory slots on the mainboard?
 
Last edited:
The AMD site says Max memory speed

2x1r 5200
2x2r 5200
4x1r 3600
4x2r 3600

Which suggests it can use ram at 5200 if you use all the memory slots on the mainboard?
No, not if there are 4 slots.

Lemme translate:

2 sticks, 1 rank each (i.e. 16GB or 24GB): 5200
2 sticks, 2 ranks each (i.e. 32GB or 48GB): 5200
4 sticks, 1 rank each: 3600
4 sticks, 2 ranks each: 3600

From what I've heard, you can do pretty well with 4 single rank sticks in some boards, but it is still not a configuration that is recommended.

4 dual rank sticks is a problem and it can achieve just 5000 or less, depending on the CPU, motherboard and memory.

So assuming my mainboard says it will run faster memory will putting this kind of 6000 memory in actually be worth while or should I just stick to 5200 memory?
For content creation the difference tends to be quite small, but it does depend on your specific workload. Puget have a DDR5 article here:

For gaming, the difference is more significant, especially with a non-X3D CPU and something best avoided (I mean, running RAM lower than 6000) if possible. HUB did a memory scaling article awhile ago that includes both X3D and non-X3D, though keep in mind that the resolution and graphics card used does change things.

 
With 2 slots populated, any AM5 CPU can do 6000, most can run 6200. When going above that gains almost nothing since memory controller runs at half speed.
It will be faster than 5200 any day

This kit seems like a no brainer. Low latency DDR5-6000 and supports XMP

If you need more capacity it is easier to go for 2x24 2x32 or 2x48 kits first rather than populating 4 slots.
 
No, not if there are 4 slots.

Lemme translate:

2 sticks, 1 rank each (i.e. 16GB or 24GB): 5200
2 sticks, 2 ranks each (i.e. 32GB or 48GB): 5200
4 sticks, 1 rank each: 3600
4 sticks, 2 ranks each: 3600

From what I've heard, you can do pretty well with 4 single rank sticks in some boards, but it is still not a configuration that is recommended.

4 dual rank sticks is a problem and it can achieve just 5000 or less, depending on the CPU, motherboard and memory.


For content creation the difference tends to be quite small, but it does depend on your specific workload. Puget have a DDR5 article here:


For gaming, the difference is more significant, especially with a non-X3D CPU and something best avoided (I mean, running RAM lower than 6000) if possible. HUB did a memory scaling article awhile ago that includes both X3D and non-X3D, though keep in mind that the resolution and graphics card used does change things.

With 2 slots populated, any AM5 CPU can do 6000, most can run 6200. When going above that gains almost nothing since memory controller runs at half speed.
It will be faster than 5200 any day

This kit seems like a no brainer. Low latency DDR5-6000 and supports XMP


If you need more capacity it is easier to go for 2x24 2x32 or 2x48 kits first rather than populating 4 slots.

So my reading here is that I need to stick to using 2 slots, but that it's best to have memory that's faster then the officially supported memory?

I.e. 6000 mhz over 5200 for the Ryzen 9 7900X?

This is me trying to advise on a friends build, like I said above he seems to have his heart set on 64gb of ram, I keep trying to find ways to save him a little bit here and there because I think he's be better served using a Ryzen 7 9800 x3d if he can stretch to afford it (or maybe even the 7 7800 x3d but then we're back in am4 land aren't we?)
 
Last edited:
So my reading here is that I need to stick to using 2 slots, but that it's best to have memory that's faster then the officially supported memory?
Correct.

I.e. 6000 mhz over 5200 for the Ryzen 9 7900X?
AMD themselves consider low latency 6000 to be the optimal speed, for 7000 CPUs. For 9000 CPUs you can potentially reach up to 6400, but 6000 is still the safest option and performs well.

This is me trying to advise on a friends build, like I said above he seems to have his heart set on 64gb of ram, I keep trying to find ways to save him a little bit here and there because I think he's be better served using a Ryzen 7 9800 x3d if he can stretch to afford it (or maybe even the 7 7800 x3d but then we're back in am4 land aren't we?)
Personally, I think 64GB is fine for a higher-end build, but if downgrading it gets you a better GPU/CPU then for gaming it is still a luxury.

AM4 is the 5700X3D, 7800X3D is just a first generation AM5 CPU. You can use the 9000 series on first gen AM5 boards (like the B650/X670) and vice versa, though a BIOS flash might be required to use 9000 series CPUs in the older boards.

I'd agree though, the 7900/9900 are generally not recommended for gaming only builds (mixed workloads are fine) due to the 6x6 configuration being suboptimal to the 8x0 config found on the 8 core CPUs (non-X3D or X3D).
 
Correct.


AMD themselves consider low latency 6000 to be the optimal speed, for 7000 CPUs. For 9000 CPUs you can potentially reach up to 6400, but 6000 is still the safest option and performs well.


Personally, I think 64GB is fine for a higher-end build, but if downgrading it gets you a better GPU/CPU then for gaming it is still a luxury.

AM4 is the 5700X3D, 7800X3D is just a first generation AM5 CPU. You can use the 9000 series on first gen AM5 boards (like the B650/X670) and vice versa, though a BIOS flash might be required to use 9000 series CPUs in the older boards.

I'd agree though, the 7900/9900 are generally not recommended for gaming only builds (mixed workloads are fine) due to the 6x6 configuration being suboptimal to the 8x0 config found on the 8 core CPUs (non-X3D or X3D).
Thanks, this mostly makes sense, but what is the 6x6 vs 8x0 configuration is it performance cores and e-cores or something?
 
Thanks, this mostly makes sense, but what is the 6x6 vs 8x0 configuration is it performance cores and e-cores or something?
The 12 and 16 core CPUs are actually two dies in one package, so with the 7900 you have 6 cores and 6 cores.

That is believed to be the reason why they perform a bit less well for gaming, due to the increased latency between cores and having only 6 cores available without a latency penalty.

If you're putting one in a PC that is doing heavy workloads and gaming, then having the 4 extra cores available on the 7900 is worth the performance loss. Otherwise, I'd just get the 8 core models which don't have this problem, because they have one fully enabled 8 core die.

The 16 core models don't suffer the same way as the 12 core models, because they can keep the game working within one 8 core die and not suffer any latency penalty, but if you don't do any CPU intensive workloads there's not much point having the other 8 cores available anyway.
 
Back
Top Bottom