RAM explained: Why two modules are better than four / single vs. dual-rank / stability testing

[citay]

Please create a new thread for your issue, where we can help you specifically. This thread is more for general discussion about RAM and the memory system.

 

[The_King]

Taking into account that a single stick of RAM can achieve higher speeds. I was finally able to get my Ryzen 1700X to do 1900 even posted into windows with 1933!

 

[Mark.Smeyers]

OK, I'm a total newbie to this, but I already gathered I want 2x32Gb for my new system for photo editing, I will be getting the MSI 2690 Edge Wifi DDR4 board with an intel i7-12700K, but I'm not sure about the memory yet.
I'm debating between either Corsair 64 GB DDR4-3200 CL16 20-20-38 Vengeance RGB PRO XMP or Corsair 64 GB DDR4-3600 CL18 22-22-42 Vengeance RGB Pro RS, the latter being about 10% more expensive ... which would be the better option? I'm just not sure about the 3200vs3600 because of the higher latency figures on the latter?
The problem is when I check the 'compatibility' on the support page, neither CMG64GX4M2D3600C18 nor CMW64GX4M2E3200C16 are listed, I also checked for the Z590 Gaming Plus, but these two modules aren't listed either. I tried to locate the 32gb Corsair modules that are listed on the MSI page, but none of the 3600 (only one is listed) can be found over here, does that specifically means I can't use these DDR4 modules on this MB?

 

[citay]

I'm just not sure about the 3200vs3600 because of the higher latency figures on the latter?

You have to make sure not to confuse RAM timings and the absolute latencies that result from them. RAM timings are set or entered in the BIOS as clock cycles, but one clock cycle at DDR4-3600 is 12.5% shorter than at DDR4-3200, because the frequency is 12.5% higher. So you really have to look at the nanoseconds of the resulting absolute latencies to compare it fairly.

DDR4-3200 CL16 results in 10 ns CAS Latency
DDR4-3600 CL18 results in 10 ns CAS Latency (because 16 * 1.125 = 18)

Full primary timings to absolute latencies:
DDR4-3200 CL16 20-20-38 = 10 ns, 12.5 ns, 12.5 ns, 23.75 ns
DDR4-3600 CL18 22-22-42 = 10 ns, 12.22 ns, 12.22 ns, 23.33 ns

The absolute latencies are essentially the same, those slight differences are irrelevant and owed to the fact that the timings can only be full clock cycles. We can't see the secondary or tertiary timings, but they should also be comparable, relatively speaking.

But there is one undeniable advantage that the DDR4-3600 kit has here: 12.5% higher bandwidth. In light of this, a 10% higher price seems appropriate. The only downside of DDR4-3600 is that it's a bit more challenging for the memory system, for example the MSI BIOS will probably automatically set the memory-related voltages a bit higher for that kit, compared to what it deems necessary for DDR4-3200. This is another problem of leaving things completely in the hand of the BIOS or the RAM's XMP profile: Both err on the side of caution / stability, sometimes excessively so. For the RAM, the secondary and tertiary timings in the XMP profile are often exceptionally loose to prioritize stability, and for the RAM-related voltages, the board's BIOS (not just on MSI) will sometimes set sky-high values to ensure stability under all circumstances with all kits. This can go so far as to setting borderline dangerous auto voltages with very high-frequency RAM. Not for the DRAM voltage itself, but for things like VCCSA (System Agent voltage, a RAM-related part of the CPU).

Since this is all fairly complex subject matter and not easy to understand for the average buyer - plus manual adjustments always require stability testing - it's often easiest to make sure to get a decent enough XMP profile and stick to that, if you don't want to spend any more time on fine-tuning. In light of that, i would recommend the DDR4-3600 CL18 kit. However, if RAM performance isn't your utmost concern, and you rather save a bit of money and potentially have less trouble getting the XMP profile to work, you can also get the DDR4-3200 CL16 kit.

Please note that with Z690, as i now see i also mentioned in this thread where we talked about your possible purchases before, the BIOS is still immature. The intervals between BIOS updates are a bit longer than what i had anticipated. There might be some hiccups with XMP support still. But they should be ironed out eventually. The QVL for the RAM is often quite spotty, especially with very high-capacity kits, as they are not all too popular. But i wouldn't let this influence my buying decision too much. Almost all popular kits will still be able to run on there in the end.

 

[RemusM]

OK, I'm a total newbie to this, but I already gathered I want 2x32Gb for my new system for photo editing, I will be getting the MSI 2690 Edge Wifi DDR4 board with an intel i7-12700K, but I'm not sure about the memory yet.
I'm debating between either Corsair 64 GB DDR4-3200 CL16 20-20-38 Vengeance RGB PRO XMP or Corsair 64 GB DDR4-3600 CL18 22-22-42 Vengeance RGB Pro RS, the latter being about 10% more expensive ... which would be the better option?

64GB at 3600MHz is NOT guaranteed to run (stable) on any of the current CPUs.
That being said, I would go with DDR4-3200 CL16.
You'll get the performances of DDR4-3600 CL18 but a much more stable system.

 

[Mark.Smeyers]

Thank you for getting back to me on this, is a Z690 board a good option right now (I'm looking at early March for my purchase) or should I go back to 11th gen and a Z590 board?
I need a running, stable system as this will be my 'work horse', I don't really look forward to continuous fine-tuning for a long time to get it running stable.

 

[citay]

Like i touched on in the other thread, the Z690 DDR4 + 12th gen is far superior. The CPUs made such a big leap forward that a lower model can beat a bigger 11th gen model convincingly sometimes. Once the next BIOS version is available, the worst problems should be left behind as well. You might want to wait until they release that next stable BIOS before your purchase. As with any cutting-edge technology, there can be some problems initially. But at the core (literally), there is not a single doubt that 12th gen Alder Lake is the best thing Intel have come up since 2015, when they first unveiled the 6th gen Skylake CPU generation. Getting an 11th gen system now is a bit hard to justify, unless the price is right of course. BTW i have an 11th gen system myself, but i got it Q1/2021.

 

[username]

So you really have to look at the nanoseconds of the resulting absolute latencies to compare it fairly.
DDR4-3200 CL16 results in 10 ns CAS Latency
DDR4-3600 CL18 results in 10 ns CAS Latency (because 16 * 1.125 = 18)
Full primary timings to absolute latencies:
DDR4-3200 CL16 20-20-38 = 10 ns, 12.5 ns, 12.5 ns, 23.75 ns
DDR4-3600 CL18 22-22-42 = 10 ns, 12.22 ns, 12.22 ns, 23.33 ns
The absolute latencies are essentially the same, those slight differences are irrelevant and owed to the fact that the timings can only be full clock cycles. We can't see the secondary or tertiary timings, but they should also be comparable, relatively speaking.

I was wondering if you could add a section about latencies in your first post to go along with all the other excellent details or a separate post to break down the differences. Along the same lines as having anything higher than 3600 being worse - when does it become cost prohibitive to get that faster CL?

 

[citay]

I'll think about it, thanks for the feedback.

 

[username]

I'll think about it, thanks for the feedback.

Your first post should really be a sticky on this forum. Very well written and easily understood. Thanks for taking the time to write it up. Cheers!

 

[The_King]

@citay
I recently purchased a Samsung B-Die Kit and am having issues with getting 3600 to run without errors. My Crucial REV. E Dual Rank Kit does 3600 CL14 runs just fine 24/7.
So I believe there is another issue with the B-Die that is causing errors in TM5. The best I can do at the moment without any errors is 3400 CL14.
What could be causing these errors at 3600?

 

[citay]

That's the Patriot Viper Steel Kit 16GB, DDR4-4400, CL19-19-19-39 (PVS416G440C9K)? Well, it's certainly not the RAM itself being the limitation here.

You must've tried a bit more voltage already, but 1.4V should be plenty for such loose timings. My guess is the 1700X's IMC. The 1000-series didn't have such a good IMC yet, so it might be that the Rev. E kit is just better suited for some reason. I bet if you used a 3000-series CPU with its better IMC, you could get DDR4-3600 or -3800 (with UCLK==MEMCLK) working perfectly with the Patriot kit.

 

[The_King]

That's the Patriot Viper Steel Kit 16GB, DDR4-4400, CL19-19-19-39 (PVS416G440C9K)? Well, it's certainly not the RAM itself being the limitation here.

You must've tried a bit more voltage already, but 1.4V should be plenty for such loose timings. My guess is the 1700X's IMC. The 1000-series didn't have such a good IMC yet, so it might be that the Rev. E kit is just better suited for some reason. I bet if you used a 3000-series CPU with its better IMC, you could get DDR4-3600 or -3800 (with UCLK==MEMCLK) working perfectly with the Patriot kit.

Do you think ProcODT may have somthing to do with it?
My B450M Mortar MAX has different settings but will not boot with anything lower than 53.3 Ohm.
Maybe that's a limitation with my 1700X not booting below 53.3 Ohms?

 

[citay]

Sorry, that i can't really answer, i have way more experience on the Intel side of RAM tuning.

Maybe you can ask in a thread like this, https://www.overclock.net/threads/official-amd-ryzen-ddr4-24-7-memory-stability-thread.1628751/

Be aware though, that forum is full of people who overclock their RAM quite severely, using active cooling and voltages that i would never use for a daily system. A lot of them are chasing some semi-arbitrary numbers, trying to out-tune each other and so on. So you have to stay level-headed there and not see their results as the norm. But there should be some people with enough tuning experience on AMD to answer such a question for you.

 

[Alan J T]

Do you think ProcODT may have somthing to do with it?
My B450M Mortar MAX has different settings but will not boot with anything lower than 53.3 Ohm.
Maybe that's a limitation with my 1700X not booting below 53.3 Ohms?

Could you screen shot that setting it not one I have seen or have just not noticed.
Or is that in ZEN timings app

 

[The_King]

Sorry, that i can't really answer, i have way more experience on the Intel side of RAM tuning.

Maybe you can ask in a thread like this, https://www.overclock.net/threads/official-amd-ryzen-ddr4-24-7-memory-stability-thread.1628751/

Be aware though, that forum is full of people who overclock their RAM quite severely, using active cooling and voltages that i would never use for a daily system. A lot of them are chasing some semi-arbitrary numbers, trying to out-tune each other and so on. So you have to stay level-headed there and not see their results as the norm. But there should be some people with enough tuning experience on AMD to answer such a question for you.

I'm one of "those" people I do post regularly in that very thread. That's were i was told to try to lower ProcODT.

 

[The_King]

Could you screen shot that setting it not one I have seen or have just not noticed.

I only managed to boot 3400 with 48 Ohms but 3600 will not post with anything lower than 53.3 Ohms

 

[Alan J T]

I would put it down to you CPU serie 1000 were not very memory OC friendly

 

[The_King]

I would put it down to you CPU serie 1000 were not very memory OC friendly

Will be time to finally upgrade this month. Most likely a 5700X maybe a 5800X3D if its in my budget.

This 1700X served me well over the years.

 

[Alan J T]

Will be time to finally upgrade this month. Most likely a 5700X maybe a 5800X3D if its in my budget.

This 1700X served me well over the years.

The 5800X3D can not be Overclocked at all was looking at that optin till I seen that, will most like grab a R9 5950X when I can grab one on Special I would not be surprised if a R7 5800X can outperform it if with a OC done