Want to upgrade to a Ryzen 3000 (maybe 4000), and need to know about DDR4 and the AMD IMC

DejaWiz

Fully [H]
Joined
Apr 15, 2005
Messages
21,819
Hello [H]orde!

After almost 8 years, it's about time to put the i7-3770K out to pasture.
My plan is to go with a Ryzen 3700X (or upcoming 4000 series equivalent), a modestly-priced X570 MoBo that doesn't suffer from extremely high VRM temps (looking at you, MSI), and four DDR4 modules...I need all four slots populated, or my OCD will keep me up at night. :ROFLMAO:

After some very basic research, I've almost reached the point of going with 4x8GB single rank modules in 3200 or 3600 flavor.

Here are my questions, for those that have experience with the AMD 3000 series with 4 DIMMs:
1. Am I in the correct line of thinking to go with single rank modules in order to be able to run four DDR4-3200 at their 3200 speed?
2. If four modules would put a bit of a strain on the Ryzen 3000 IMC (versus two modules) and create the need to underclock the RAM for purposes of operational stability, then would going with four DDR4-3600 modules be a better option if I want to hit the target of 3200 speed?
3. Would going with 4x16GB instead of 4x8GB of DDR4-3200 or 3600 (and undoubtedly moving to dual rank) modules still be OK for hitting the 3200 speed target?
4. Pros and Cons of single rank vs dual rank with the Ryzen 3000 IMC?

Sorry for all the questions and if they seem a bit out there or stupid...the only experience I have with DDR4 is throwing in DIMMS in corporate computers that need zero specialized attention.
 
1.) Single ranked modules are theoretically capable of higher clock speeds when used in groups of 4 vs. dual ranked modules. That said, single, or dual rank makes only a small difference. The Ryzen 3000 series CPU's aren't good at clocking four DIMMs worth a damn. It's that simple. Hitting 3200MHz is possible, but I wouldn't bank on it. Officially, DDR4 2666MHz or 2933MHz are the highest speeds you can reliably achieve according to AMD. That said, your mileage can vary greatly on this but I certainly wouldn't expect anything beyond DDR4 3000MHz with any kind of certainty. DDR4 3200MHz is certainly possible but 3600MHz? Really isn't in the cards for most people.

Also, most of the X570 motherboards use a daisy chain setup for their memory topology. It is NOT ideal for running four DIMMs. Daisy chain's advantage lies in clocking two DIMMs higher. It's generally done this way so that user's can more reliably hit DDR4 3600MHz slightly over that as the 3000 series Ryzen's see some benefit from doing so.

2.) Going to 3600MHz modules will probably not improve your chances of reaching 3200MHz target when going with single-ranked modules. It's more up to the memory IC's used and the motherboard itself. BIOS implementation matters too. Going with a motherboard that uses a T-Toplogy configuration is what would give you the best chance. However, good luck finding one and going inexpensive on the motherboard isn't going to help you with your quest to satisfy your OCD.

3.) No.

4.) This is basically already covered by the answer to your earlier question. In fact, you've largely asked the same question pretty much four different ways.
 
I was in your shoes 6 months ago, I replaced my 3770K with the rig in my sig. I wanted 64GB of fast memory, but following the QVL, 3466 was about as fast as I could get with that much memory. I tried a bunch of settings to try and get it up to 3600 but could never get it stable, I ended up leaving it at the stock settings. In short dont expect 4 sticks to clock anywhere near as high as 2 sticks.
 
I was in your shoes 6 months ago, I replaced my 3770K with the rig in my sig. I wanted 64GB of fast memory, but following the QVL, 3466 was about as fast as I could get with that much memory. I tried a bunch of settings to try and get it up to 3600 but could never get it stable, I ended up leaving it at the stock settings. In short dont expect 4 sticks to clock anywhere near as high as 2 sticks.

That's what I've discerned from my research.

So, even with four 3600 modules, I should expect a max of around 2666-2933?
 
That's what I've discerned from my research.

So, even with four 3600 modules, I should expect a max of around 2666-2933?

I would expect that as a worst case scenario. You could and probably can do somewhat better. But I wouldn't expect to get the full DDR4 3600MHz speeds using four modules. I don't think that's likely.
 
I would expect that as a worst case scenario. You could and probably can do somewhat better. But I wouldn't expect to get the full DDR4 3600MHz speeds using four modules. I don't think that's likely.

I wasn't. My line of thinking was going with the higher 3600 native in hopes of having a better chance to get all four modules to 3200.
 
four DDR4 modules...I need all four slots populated, or my OCD will keep me up at night. :ROFLMAO:

The go with ITX with 2x16GB or you will has empty PCI-E slots to keep you up at night.
Ryzen 3000 should have no problems hitting 3600-3733 with 2x16GB with a decent MB\BIOS as the memory controller is a lot stronger than it was on Ryzen 2000.
 
That's what I've discerned from my research.

So, even with four 3600 modules, I should expect a max of around 2666-2933?


I recommend no matter what clock you may be able to hold stable, stick with buy 4 sticks of the fastest memory you budget for because you have the potential to be able to run that memory at lower speeds but tighter timings and the new ryzen loves it's timings as much as it loves ram speed

I have a 1700x / crosshair 6 hero / 4x8gb trident z 3200 c16... When I was building this computer c14 sticks were available but they were easily 2x the price and totally not worth the price increase to me... when it was first running I had zero issues to my surprise hitti g the xmp 3200 c16 speeds, but there was a variation of bios updates that completely fu***d with my memory controller and for a while I couldn't even get to the bios with speed set to anything over 2900 even reverting back to a known working bios gave me issues that I never had before... Now with the latest I can run my 3200 at it's full xmp speed with not even a pickup.... All I can say is first gen ryzen has been one hell of a rollercoaster for me...

I definitely recommend also that no matter what board you go for, make sure it has the debug LED's, that really is a nice thing to have especially when you go to hit the power button on your PC and it screems at you for no apparent reason and won't come on...
 
No problem using 4 x 8gb dims rated at 3200 14,14,14,34 (B-die) running at 3800mhz 16,16,16,32 on my Ryzen 3900x, x370 motherboard, ASUS CrossHair 6 Hero. The motherboard is a T-Topology configured memory slots. Many also achieved these results as well with the same motherboard as a note.

Tested 2 16gb (Hynix) modules in the same board but only two slots would work at rated speeds of 3466mhz and timings. Didn't try all four of the 16gb dims at the same time (on the shelf now).
 
As noko mentioned, it has more to do with the Mobo topology than the CPU. I'm currently running 4x8 @ 3200MHz 14-14-14-34 on my Gigabyte X370 board.
 
No problem using 4 x 8gb dims rated at 3200 14,14,14,34 (B-die) running at 3800mhz 16,16,16,32 on my Ryzen 3900x, x370 motherboard, ASUS CrossHair 6 Hero. The motherboard is a T-Topology configured memory slots. Many also achieved these results as well with the same motherboard as a note.

Tested 2 16gb (Hynix) modules in the same board but only two slots would work at rated speeds of 3466mhz and timings. Didn't try all four of the 16gb dims at the same time (on the shelf now).

Thanks! I have this exact same combo and am currently using 3000 cas 16 (2x16GB) and was thinking of going faster....
OP: If you have a MC nearby, this board is $120 on closeout, and $100 after cpu combo. Might fit your needs even if not up-gradable to Zen3.

Edit: Also: A mem utility used (thaiphoon) said my 16G sticks are single rank.
 
Last edited:
Thanks! I have this exact same combo and am currently using 3000 cas 16 (2x16GB) and was thinking of going faster....
OP: If you have a MC nearby, this board is $120 on closeout, and $100 after cpu combo. Might fit your needs even if not up-gradable to Zen3.

Edit: Also: A mem utility used (thaiphoon) said my 16G sticks are single rank.
That is a pretty good deal for this feature rich motherboard. This board had a torturous path, awesome enthusiast support as well as ASUS continued improvements. I would only recommend this route if one is on a very tight budget and can get a very good deal on a CPU be it new or used. When quality X570 boards with more M.2 slots, support for PCIe 4.0 is around $100 more (`$200) for a decent one that will work with Ryzen 4 or Zen 3 CPUs -> That extra $100 maybe well worth it if one can afford it.
 
That is a pretty good deal for this feature rich motherboard. This board had a torturous path, awesome enthusiast support as well as ASUS continued improvements. I would only recommend this route if one is on a very tight budget and can get a very good deal on a CPU be it new or used. When quality X570 boards with more M.2 slots, support for PCIe 4.0 is around $100 more (`$200) for a decent one that will work with Ryzen 4 or Zen 3 CPUs -> That extra $100 maybe well worth it if one can afford it.
I think that is true if you plan to upgrade within the same socket.

For me, I'm coming from a 4770k...by the time I upgrade there will be newer boards than X570 available. If you had a choice of 3700x + x570 or 3900x + x370...what would you choose? (roughly the same price)

Crosshair VII $200 + 3700x $270 = $470
Crosshair VI $100 + 3900x $390 = $490

Price differential is 4% to have +50% cores.
 
Edit: Also: A mem utility used (thaiphoon) said my 16G sticks are single rank.
What chips does it say are on your sticks? sounds like we may finally be starting to see some of the new high density chips making it into the market.
 
Whatever motherboard you get, I would make sure you get one that has usb-c (some x370s do not)
 
I think that is true if you plan to upgrade within the same socket.

For me, I'm coming from a 4770k...by the time I upgrade there will be newer boards than X570 available. If you had a choice of 3700x + x570 or 3900x + x370...what would you choose? (roughly the same price)

Crosshair VII $200 + 3700x $270 = $470
Crosshair VI $100 + 3900x $390 = $490

Price differential is 4% to have +50% cores.
That depends on how much money I could afford to spend now and immediate future. If this is only a one time deal for a couple of years then of course the 3900x + x370 which works virtually perfect with the CH6 Hero. Right now I am looking at Threadripper as a workhorse setup but that is just me and my situation. I have a 1TB NVME in the M.2 slot and a 2TB NVME M.2 in an adapter using the third pcie 16x slot, works perfect. Plus some SATA drives. The board has plenty of USB slots for keyboards, multiple mice, camera, SD cards, VR, etc. It has a USB C slot. Can flash the bios without the CPU, clear bios and flashback switch on the back panel, Start, Safe Boot (very convenient), Retry buttons on the motherboard. The board is packed with features, had SLI 1070's and 1080Ti's and Crossfire Vega FE's in it prior, currently 5700 XT AE. Running B die, 4 sticks at 3800mhz 1:1 ratio. Bios is rich with features and settings. Board has been overall very stable after about the first year which is more attributed to AMD AGESA and Bios updates with the original Ryzen. On the fourth CPU in it, 3900x. $100 is a good price if you are not really going to update to a Zen 3 or Ryzen 4, for me if Ryzen 4 is a 15% performance increase I would not see the point, won't know until then. Only real negative if even a negative, does not bother me, it takes a little longer to boot.
 
Last edited:
Dropped this in another thread, but:

GSkill Ripjaws DDR4 3200CL14 - Just got this for my 3950x / x570 Aorus Ultra setup and it does CL14@3666 at 1.44v right off the bat. It can go higher, but my chip IF doesn't like the 1866 1:1 clocking.

350200_ram.png
 
Dropped this in another thread, but:

GSkill Ripjaws DDR4 3200CL14 - Just got this for my 3950x / x570 Aorus Ultra setup and it does CL14@3666 at 1.44v right off the bat. It can go higher, but my chip IF doesn't like the 1866 1:1 clocking.

View attachment 248749
That is pretty good, have you tried:
Enable Gear Down Mode? Dual rank modules can do better when this enabled, looks like you have 16gb dims
What is your proODT set at? 40? 36?
Have you tried variations with your CAD bus: I use 24, 24, 30, 30 for higher clock speeds
SOC voltage?
Memory voltage?

Anyways you have some nice tight timings which at that speed should give you great performance. I use Aida64 memory tester to verify results, MemTest, Prime95 and at times Games/3DMark to see if I am actually improving performance and have stability.
 
...I need all four slots populated, or my OCD will keep me up at night. :ROFLMAO:
Alternatively ... you can get a huge heatsink like a Noctua NH-D15 or the like that will cover up your RAM slots so you can't see them to trigger your OCD
I run the Noctua on my 3700x/ASRock x570 Steel Legend/2 x 16GB HyperX Predator 3200mhz CAS16 @ 3800mhz CAS18 w/ 1:1 IF 1.41v
 
Last edited:
I had been running Corsair Vengence Pro RGB 3600 4x8GB at 3800Mhz with the IF at 1900Mhz for 7 months on my X570 board. That was C18 Ram. I just upgraded to a new kit and it is 3600C14 and I am running that at 3600 14-14-14-28, I could get it at 3733 but was not happy with the voltage needed to make it stable with no memory errors so I tweaked the timings more and dropped it back to 3600.
 

Attachments

  • cachemem06232020.png
    cachemem06232020.png
    92.3 KB · Views: 0
  • ZenTimings_Screenshot.png
    ZenTimings_Screenshot.png
    15.6 KB · Views: 0
I had been running Corsair Vengence Pro RGB 3600 4x8GB at 3800Mhz with the IF at 1900Mhz for 7 months on my X570 board. That was C18 Ram. I just upgraded to a new kit and it is 3600C14 and I am running that at 3600 14-14-14-28, I could get it at 3733 but was not happy with the voltage needed to make it stable with no memory errors so I tweaked the timings more and dropped it back to 3600.

What ram volts now for that 3600CL14 tune?
 
nice timings .. now I don't feel as concerned running my stuff at 1.41v's to run way looser timings @ 3800 with 1:1 IF .. something like 18-22-22-38 ..
I had been running Corsair Vengence Pro RGB 3600 4x8GB at 3800Mhz with the IF at 1900Mhz for 7 months on my X570 board. That was C18 Ram. I just upgraded to a new kit and it is 3600C14 and I am running that at 3600 14-14-14-28, I could get it at 3733 but was not happy with the voltage needed to make it stable with no memory errors so I tweaked the timings more and dropped it back to 3600.
getting similar Read and Latency numbers .. Read is a little higher with mine, but latency is 68. something (at work so running off of memory)
 
nice timings .. now I don't feel as concerned running my stuff at 1.41v's to run way looser timings @ 3800 with 1:1 IF .. something like 18-22-22-38 ..

getting similar Read and Latency numbers .. Read is a little higher with mine, but latency is 68. something (at work so running off of memory)

Upped it to 3800 this morning to see if it would be stable.
 

Attachments

  • cachememGS3800.png
    cachememGS3800.png
    92.4 KB · Views: 0
  • ZenTimings_Screenshot3800.png
    ZenTimings_Screenshot3800.png
    14.1 KB · Views: 0
Wow, those latencies seems really high considering the speed and timings of the RAM. I am not terribly familiar with the new AMD processors (although I am targeting protentional upgrades to Ryzen 4000 or whatever is out from Intel by then).

Is that typical, good or bad for AMD?

I get about 55.0ns latency on my current setup and I know some newer intel setups that are Dual Channel only (not Quad like mine) get latencies in the 40ns range. I did tune my second and third level timings too (yeah, I guess I like the madness).

In the end, does it make a terribly large difference in gaming or everyday applications vs. the speed of the RAM? I know there are a TON of articles on this, and I have read them a ton of times over the years, but curious how real world people using their systems can relate. I was always under the impression lower latency was better, and if you can achieve more speed while keeping it low, all the better, but at some point it could cause more harm than good.
 
Wow, those latencies seems really high considering the speed and timings of the RAM. I am not terribly familiar with the new AMD processors (although I am targeting protentional upgrades to Ryzen 4000 or whatever is out from Intel by then).

Is that typical, good or bad for AMD?
It's normal, AMD removed the memory controller from the CPU and put it on a separate on-package IC with a fuckton of cache to hide the resulting latency. Works fairly well for real workloads, but synthetics that focus on subsystems will reveal it.
 
The high latency with Ryzen is the main reason it falls behind Intel in programs like games but is competitive in single threaded Cinebench which stays mostly within the cache.
On the other hand AMD has a advantage with there multithreading being more efficient than Intel HT.

The large L3 cache on new Ryzen CPU helps negate the performance penalty in some games that dont greatly overflow the cache like Counter Strike but open world games that really hammer the RAM see next to no benefit from the cache size increase as it is simply no where near big enough.

But it is not just the latency between RAM and cache that hurts Ryzen but the latency between cores on separate CCD which you can see the impact of if you look at 3100 vs 3300X review
 
The high latency with Ryzen is the main reason it falls behind Intel in programs like games but is competitive in single threaded Cinebench which stays mostly within the cache.
On the other hand AMD has a advantage with there multithreading being more efficient than Intel HT.

The large L3 cache on new Ryzen CPU helps negate the performance penalty in some games that dont greatly overflow the cache like Counter Strike but open world games that really hammer the RAM see next to no benefit from the cache size increase as it is simply no where near big enough.

But it is not just the latency between RAM and cache that hurts Ryzen but the latency between cores on separate CCD which you can see the impact of if you look at 3100 vs 3300X review
Overall, it's a wash; if one is in the minority where the extra per-core performance of Intel platforms matters, well, Intel has a platform to sell you.

Otherwise, AMD will sell you more cores for less, or just charge you less for the same cores, and that's a hard deal to pass up.
 
No denying AMD has the best value option but RAM latency is something they have to work on if they want to steal the crown from Intel for gaming performance with plenty of 2080ti owners out there willing to pay the premium Intel asks and even the 10600K overclocked with a high speed kit of RAM is one hell of a CPU for gaming.

With 3600 14-15 on 7700K my system runs ~36ns vs ~64ns above.
I am hoping Ryzen 4000 will improve enough to give me a upgrade path for the games I play with pore multithreading that drop under 60FPS.
 
Last edited:
No denying AMD has the best value option but RAM latency is something they have to work on if they want to steal the crown from Intel for gaming performance with plenty of 2080ti owners out there willing to pay the premium Intel asks and even the 10600K overclocked with a high speed kit of RAM is one hell of a CPU for gaming.
The issue of memory latency is one that appears to have been a bigger theoretical problem than a real problem with Zen 2. Realistically, if they could increase clockspeeds by about 500MHz, they'd be fully up to par with their current architecture.

Just keep in mind that this is just one element of a system, not the whole system, and it may not affect performance the way you imagine.
 
True they could just increase clock speed or further increase there IPC lead over Intel to get around the performance hit the CPU is taking from RAM latency.
But 500MHz is probably only enough to catch them up to a stock 10900K with both systems running the same speed RAM they would probably need about an extra 800MHz to take the lead vs a overclocked 10900K with faster RAM give or take a bit depending on the title.

Here Ryzen 3900X is seeing ~15-20% increase in game performance with overclocked RAM vs 3200c14.
https://www.gamersnexus.net/guides/3508-ryzen-3000-memory-benchmark-best-ram-fclk-uclock-mclock

Here 10600K is seeing ~6% increase with 4000c15 over 3200c14 but I am sure they could get another 5-10% easy by tweaking sub timings assuming they are not already running into a GPU bottleneck.


3300X seems to be a interesting exception not seeing a significant performance gain and I cant help wonder why that is.
Sure there is less cores fighting over the memory so you can expect it to matter less but still it is seeing next to no change.
Or maybe it is just something with there test setup since they are seeing larger gains here just going from 3000 to 3600 so it would have been interesting to see Aida64 results from both review sites to see what is going on.
https://www.eurogamer.net/articles/digitalfoundry-2020-amd-ryzen-3100-3300x-cpu-review?page=4

Crysis 3 seems to be a exception to most games since it doesn't care anywhere near as much about memory speed so it would be interesting to see sombody do a IPC comparison with 10900K vs 3900X in this game.

Ryzen has ~10% IPC advantage over Intel in a program like Cinebench single threaded which doesn't really care about latency.
But that swings back to Intel having up to ~10% performance advantage over AMD in latency sensitive games.
https://www.techspot.com/article/1876-4ghz-ryzen-3rd-gen-vs-core-i9/
 
Last edited:
Ryzen has ~10% IPC advantage over Intel in a program like Cinebench single threaded which doesn't really care about latency.
But that swings back to Intel having up to ~10% performance advantage over AMD in latency sensitive games.
This is a 'results' based perspective, but from a broader standpoint, it's kind of like 'does it fit in cache?' and 'does it benefit from higher clockspeeds?'.
 
Yes the more time the CPU spends waiting on RAM due to something not being in cache the less difference increases in clock speed makes.
So in some games even a 1000MHz increase to core clock may not be enough to catch Intel as it is spending too much time waiting on RAM making the % gain from increased clock speed significantly less.
While in something like Counterstrike AMD may already have the lead thanks to the L3 cache being large enough for this simple game although it hardly matters since games like this are running over 400FPS on both platforms.

At the end of the day for the games I play the 7700K is still faster than anything AMD has to offer and anything new Intel has to offer is just a very marginal improvement with maybe 200MHz higher OC and higher potential RAM speed for a little more bandwidth but not much better latency.
Ether platform will be a decent improvement over the 3770K.
 
Last edited:
Back
Top