DDR5-6400 Confirmed as Sweetspot Speed of Ryzen 9000 "Zen 5" Desktop Processors

[btarunr]

AMD's upcoming Ryzen 9000 series "Granite Ridge" desktop processors based on the "Zen 5" microarchitecture will see a slight improvement in memory overclocking capabilities. A chiplet-based processor, just like the Ryzen 7000 "Raphael," "Granite Ridge" combines one or two "Zen 5" CCDs, each built on the TSMC 4 nm process, with a client I/O die (cIOD) built on the 6 nm node. The cIOD of "Granite Ridge" appears to be almost identical to that of "Raphael." This is the chiplet that contains the processor's DDR5 memory controllers.

As part of the update, Ryzen 9000 "Granite Ridge" should be able to run DDR5-6400 with a 1:1 ratio between the MCLK and FCLK domains. This is a slight increase from the DDR5-6000 sweetspot speed of Ryzen 7000 "Raphael" processors. AMD is reportedly making it possible for motherboard manufacturers and prebuilt OEMs to enable a 1:2 ratio, making it possible to run high memory speeds such as DDR5-8000, although performance returns with memory speeds would begin to diminish beyond the DDR5-6400 @ 1:1 setting. Memory manufacturers should launch a new wave of DDR5 memory kits with AMD EXPO profiles for DDR5-6400.



View at TechPowerUp Main Site | Source

 

[bug]

I guess DDR5-6400 won't break the bank, so that's good.

 

[dont whant to set it"']

* 2x32GB@6400MT go for a nice penny or two.

 

[Launcestonian]

Depends really, until benchmarks are revealed no one knows if single CCD chips will benefit from this. Zen 4 single CCD sweet spot is of course 6000/MTs, at least in gaming situations & as low a CL as possible.

 

[SL2]

As part of the update, Ryzen 9000 "Granite Ridge" should be able to run DDR5-6400 with a 1:1 ratio between the MCLK and FCLK domains.

How fast RAM would you need to theoretically beat the performance of above setting with 2:1?

 

[tabascosauz]

How fast RAM would you need to theoretically beat the performance of above setting with 2:1?

Somewhere north of 8000?

Starting to beat 6000 in certain metrics on current gen is like 8000-8400 I think.

 

[Chaitanya]

I hope to see JEDEC 6400 support as well.

 

[RogueSix]

Gamers shouldn't be worrying about this anyway since you will likely run a X3D CPU. The 3D cache CPUs are all but RAM-agnostic so DDR5-6000 vs. DDR5-6400 is going to make a difference of less than 0.5% most likely in real world benchmarks. Low latency is more important.

 

[Skylinestar]

I'm praying for 24GBx2 kit with good timing. It doesn't exist today yet.

 

[ratirt]

I'm praying for 24GBx2 kit with good timing. It doesn't exist today yet.

12GB per module? That would be a little bit odd. Why 24GB anyway? cheaper than 32GB and not much more expensive than 16GB?

 

[stimpy88]

Well, isn't this a load of cr8p. Between this and Intel beating the standard Zen 5 6 and 8, and possibly the 12 core chips in gaming perf as well as offering more PCie lanes from the CPU, AMD is starting to look slightly less attractive to me.

There is starting to get a few too many drawbacks for investing in AM5 now.

 

[dgianstefani]

* 2x32GB@6400MT go for a nice penny or two.

You likely won't be getting 64 GB of memory to run at the sweet spot.

12GB per module? That would be a little bit odd. Why 24GB anyway? cheaper than 32GB and not much more expensive than 16GB?

I'm praying for 24GBx2 kit with good timing. It doesn't exist today yet.

24 GB kits perform similarly to 16 GB kits (2x16 GB), whereas 32 GB kits perform worse, with poor timings/MT.

 

[Haile Selassie]

Somewhere north of 8000?

Starting to beat 6000 in certain metrics on current gen is like 8000-8400 I think.

My experience ~ 6400 C32 is roughly comparable to 7400 C38, albeit the latter has significantly higher memory bandwidth. When both tuned correctly the latencies are very comparable.

 

[ratirt]

24 GB kits perform similarly to 16 GB kits (2x16 GB), whereas 32 GB kits perform worse, with poor timings/MT.

I'm pretty sure if you buy, for example, 2x16 6400 CL 32 and/or 2x32 6400 CL 32 these will perform very similar.

 

[Kaleid]

Likely no reason to upgrade from 6000 in other words

 

[dgianstefani]

I'm pretty sure if you buy, for example, 2x16 6400 CL 32 and/or 2x32 6400 CL 32 these will perform very similar.

...

Well, for starters, the only timings that the XMP/EXPO profile store are the main timings, the subtimings (which on DDR5 have more impact on performance) are worse on high capacity kits because the motherboard trains the memory and sets those timings, which are worse because more memory is harder to run.

Easy to forget how low AMD memory support is relatively, and associate what the platform can do with what the memory can do, which aren't the same thing. Available 2x16 kits go up to 8200 MT, 2x32 go up to 6800 MT max, but good luck running at more than 6800 on Intel, or 6200 MT on AMD, on a 1DPC motherboard, worse on 2DPC.

Now look up how likely you are to run 64 GB of 6400 MT CL32 vs how likely you are to run 32 GB (generally plug and play on XMP, even on Zen 4).

Lower capacity DIMMs run both faster (higher MT/lower latency) and easier than higher capacity DIMMS.

 

[ratirt]

...

Well, for starters, the only timings that the XMP/EXPO profile store are the main timings, the subtimings (which on DDR5 have more impact on performance) are worse on high capacity kits because the motherboard trains the memory and sets those timings, which are worse because more memory is harder to run.

Easy to forget how low AMD memory support is relatively, and associate what the platform can do with what the memory can do, which aren't the same thing. Available 2x16 kits go up to 8200 MT, 2x32 go up to 6800 MT max, but good luck running at more than 6800 on Intel, or 6200 MT on AMD, on a 1DPC motherboard, worse on 2DPC.

Now look up how likely you are to run 64 GB of 6400 MT CL32 vs how likely you are to run 32 GB (generally plug and play on XMP, even on Zen 4).

Lower capacity DIMMs run both faster (higher MT/lower latency) and easier than higher capacity DIMMS.

I'm not associating anything, I'm saying if you are aiming for 6200Mhz mem and 32CL for instance, it does not matter if you get 2x16 or 2x32. These will perform the same. As an example.
Obviously if you want higher frequency you may need to get higher latency modules.
7200Mhz with CL34. Yes it has higher CL but also higher frequency. Both are available with the 2x16 and 2x32 configuration.
8000Mhz frequency modules don't have kits 2x32 (my reseller does not have these) due to the reason you mentioned but that is an extreme. You can still get 2x24 and 2x16
Besides, there's a lot of videos on YT done about the memory and the crazy frequencies do not benefit either Intel nor AMD after you go over certain threshold. You wont see much difference with Intel if you get 8000Mhz kit vs 7200Mhz. maybe certain scenarios will benefit but I'm sure it is not worth it.

 

[Carlyle2020hs]

Let´s quote the source correctly please: "..., the Ryzen 9000 series still targets DDR-5 6000 memory as a sweet spot, but he upper limit to 6400 MT/s while maintaining a 1:1 FCLK to MCLK ratio.

Was that a humans error during summary or did something else make that mistake?

 

[phints]

Sweet spot of DDR5 6400 is not a problem, that will help keep the prices down a bit. I'd like to know what CL speeds they recommend, with the 6000 it's CL30. Will have to wait for reviews I guess.

The bigger concern is boot times, they were attrocious when Ryzen 7000 launched mostly due to RAM training, will this be 'fixed' with 9000?

 

[bug]

Likely no reason to upgrade from 6000 in other words

Yeah, if you're already rocking DDR5-6000, it would be a rather minute improvement.

 

[ratirt]

Yeah, if you're already rocking DDR5-6000, it would be a rather minute improvement.

Or you can just OC the kit.

 

[rv8000]

...

Well, for starters, the only timings that the XMP/EXPO profile store are the main timings, the subtimings (which on DDR5 have more impact on performance) are worse on high capacity kits because the motherboard trains the memory and sets those timings, which are worse because more memory is harder to run.

Easy to forget how low AMD memory support is relatively, and associate what the platform can do with what the memory can do, which aren't the same thing. Available 2x16 kits go up to 8200 MT, 2x32 go up to 6800 MT max, but good luck running at more than 6800 on Intel, or 6200 MT on AMD, on a 1DPC motherboard, worse on 2DPC.

Now look up how likely you are to run 64 GB of 6400 MT CL32 vs how likely you are to run 32 GB (generally plug and play on XMP, even on Zen 4).

Lower capacity DIMMs run both faster (higher MT/lower latency) and easier than higher capacity DIMMS.

With respect to latency and overall performance on AMD, the majority of gains are from primary timings, trefi and trfc. The remainder of secondary and tertiary timings largely effect bandwidth and provide diminishing returns latency wise. Secondary and tertiary timings aren’t overly important unless you’re benching for a quarter ms off pi calcs.

It’s potentially disappointing to hear the potential “sweet spot” is 6400, might dispel the previous rumor of zen 5 having 2400mhz IF out of the box, unless theres some weird ratio between mclk:fclk:uclk now. 6400 would more than likely mean a 2133mhz fclk.

May be fun to pick up a 9700X and see if 6600 and 6800 will be more consistently possible in 1:1 now.

 

[dgianstefani]

With respect to latency and overall performance on AMD, the majority of gains are from primary timings, trefi and trfc. The remainder of secondary and tertiary timings largely effect bandwidth and provide diminishing returns latency wise. Secondary and tertiary timings aren’t overly important unless you’re benching for a quarter ms off pi calcs.

It’s potentially disappointing to hear the potential “sweet spot” is 6400, might dispel the previous rumor of zen 5 having 2400mhz IF out of the box, unless theres some weird ratio between mclk:fclk:uclk now. 6400 would more than likely mean a 2133mhz fclk.

May be fun to pick up a 9700X and see if 6600 and 6800 will be more consistently possible in 1:1 now.

I doubt they can push IF before Zen 6 moves to an active interposer, max another 1-200 MHz, like Zen 3 to Zen 4 (1900 to 2000).

The catch with AM5/Zen 4 was that the IF wasn't actually running faster than AM4 could, there was just a 3:2 ratio in place, 3000 MHz memory, 2000 MHz IF. Kinda sucky tbh.

Even my 5950X/5800X3D could do 2000/2000 MHz for memory/IF.

 

[rv8000]

I doubt they can push IF before Zen 6 moves to an active interposer, max another 1-200 MHz, like Zen 3 to Zen 4 (1900 to 2000).

The catch with AM5/Zen 4 was that the IF wasn't actually running faster than AM4 could, there was just a 3:2 ratio in place, 3000 MHz memory, 2000 MHz IF. Kinda sucky tbh.

Even my 5950X/5800X3D could do 2000/2000 MHz for memory/IF.

I personally don’t think 2400 IF is out of the realm of possibilities on zen 5. Based on what I've seen in memory oc threads and personal experience, a good portion of zen 4 chips are actually capable of running 2200mhz IF, it’s just very few IMC/motherboard combos can do 6600mhz 1:1 to take proper advantage of it.

We’ll see soon enough, I just hope motherboard manf. don’t miss the opportunity to release some actual memory oc boards for the 8xx gen board and don’t leave us with another castrated/overprice m-itx board as the only option.

 

[AnarchoPrimitiv]

Well, isn't this a load of cr8p. Between this and Intel beating the standard Zen 5 6 and 8, and possibly the 12 core chips in gaming perf as well as offering more PCie lanes from the CPU, AMD is starting to look slightly less attractive to me.

There is starting to get a few too many drawbacks for investing in AM5 now.

Based on the character of your past comments, I would be surprised if you ever owned AMD hardware or ever intended to....

...drawbacks to AM5? Like what? The extremely long support life that Intel has literally never even approached in the past 15 years? The high probability that AMD is going to be releasing new CPUs for AM5 for years to come if AM4 is anything to go by? What exact "drawbacks" are you referring to that AM5 has and Intel's current platform doesn't have? And please only refer to DOCUMENTED issues, not anecdotal ones or generalized, ambiguous statements with no evidence such as "stability" (though if I remember correctly, intel's latest 14th gen chips have had serious and widespread stability issues without a solution that addresses the root cause...)

"...with Intel beating the standard Zen 5 6 and 8..."

I might be misreading here, but are you actually making the claim that Intel has already "beat" the Zen 5 six and 8 core CPUs that literally haven't been released yet and nobody has reviewed, and that zintel has done this with CPUs it hasn't released and nobody has reviewed yet? Any clarification would be much appreciated.