9800x3d max ram speeds?

[76561199793899658]

what would be the max mhz ram speeds on the best motherboard for it atm.
-
with 4 dimms and with 2 dimms?

and how much performance you loose by going 4 dimms?

I currently have an x99 system with 8x16gb ddr4.
I kind of like to double the specs every new pc..
-
so 256gb ram.

as the old intel enthousiast x-platforn is kind of dead.. and treadripper is more of a modern xeon than a replacement for that old enthousiast series that used to sit in between servers and mainstream..

the days of 4 lanes for ram (and having 8 ramslots) in a gaming pc are kind of over.

was the 6950x at 1600 euro at least a 10% faster than the 500 euro mainsteam cpu
plus you did get stuff like usb 3.2, m.2 and ddr4 one or two generations before maibstream got it.
and you got more than double the lanes in the days of sli essential.

thats not today.

so the only way to get 256gb ram is by having 4x64gb now. (or at least 4x48gb as it remains to be seen if any 64gb modules are to be released at all as non ecc)

but well that does need all 4 ramslots as just using 2 would give me 96gb ram which is less than my old pc had and not acceptable..

so.. what is the neth fps loss by going 192gb (4x48gb) vs 96gb (2x48gb) and at what mhz could each of these kits be expected to run?

[Illusion of Progress]

Ursprünglich geschrieben von Outcast82:

well this ♥♥♥♥♥♥♥♥ just did not use to excist before.. not even with xmp.

Times change.

More DIMMs and ranks would have always been harder to stabilize. The difference back then was that other things were also often the limiting factor before those things were, so the difference was masked.

For example, motherboards used to be a mix of daisy chain and t-topology. The latter did better with four DIMMs, and the former pushes two DIMMs higher. In recent years, almost all consumer motherboards have moved towards the former type. This helps drive speeds for two DIMMs at the cost of four DIMMs.

And then DDR5 made things harder for a number of reasons, some of them changes it made, others no fault of its own and just a fact of "reaching higher speeds is getting harder".

You seem to often expect that things will continually scale linearly, and despite running into situation after situation where it's clearly not scaling that way anymore, you seem to refuse to want to accept it. I don't know what else to tell you. Those advancements haven't been happening at the same pace so what you want doesn't exist. In the early years of any new things, advancements will often be fast since the thing is young and still has a lot of room to grow. Once the low hanging fruit has been picked, so to speak, further advancements get harder as a given thing matures. That's what happening with tech (until some massive breakthroughs happen). Scaling peaked in the 1990s and 2000s and it's been gradually slowing in a lot of ways since.

Ursprünglich geschrieben von Outcast82:

still does not answer how large the gap is.. on general or what it translates to in actual fps..

Because there is no singular frame rate number. It varies.

You need to look for results with RAM speed at X speed versus Y speed to get that answer, but it's not always to find, and even then results will only apply to the listed games since all are different. Most testing on AM5 is done at 5,600 MHz or 6,000 MHz more commonly, and most current RAM should reach that unless you use four DIMMs, and few gaming (or even consumer) PCs really need more capacity than what two DIMMs can offer. Now yes, there's certainly people who need (or simply want) more than that, and I'd even be one of them, but you're going to have to make a performance tradeoff now in that case.

[Ontrix_Kitsune]

Ursprünglich geschrieben von Outcast82:

but what you say is :

*use the specific motherboard list.
(usually listed as
1 dimm 1 lanes
2 dimms 2 lanes
4 dimms 2 lanes )
-
so no longer it lists one speed it supports but 3.. one for each config.
**and you say this WILL always 100% work.

All modern motherboards for both Intel and AMD since at least the past 5-6 years have all been this way with modern computers.

Ursprünglich geschrieben von Outcast82:

still does not answer how large the gap is.. on general or what it translates to in actual fps..

No one can "generalize" this for you and tell you what your system can use ram at what speeds or configuration because it's completely different for each motherboard. Some motherboards may be able to support 4 x 24GB @ 7000 Mhz ram while some other motherboard may not be able to do that. This is why you have to refer to the tested QVL list. You can't just grab a random memory kit and put it in a board and expect it to work. The computer probably won't even turn on half the time if you do that.

Ursprünglich geschrieben von Outcast82:

just go look it up on motherboard manufacturer? that ♥♥♥♥ aint on the webstores search options.. so you will one by one have to look that up for 100dres of motherboards.. an awfull amount of work...

Yep if you want to build a computer yourself then system compatibility is a lot of work if you want the computer to be perfectly stable and never crash. If you don't want to deal with any of that then buy a computer from a OEM like iBuyPower. They do all of that for you and warranty their configurations for a few years.

Ursprünglich geschrieben von Outcast82:

as for people claiming 16gb dimms work faster than 48gb dimms..
sound bogus.
sure you can buy 16gb dimms at higher speeds than 48gb dimms.
but if your motherboard/cpu aint runninh it that be mute.

and everybody keeps saying amd never ever will run over 6000mhz.. 6400mhz if very very lucky..
-
8000mhz + amd cpu thus : never ever??

AMD Ryzen CPU's can run ram faster than 6000 Mhz but if you just go into the system bios and select the XMP/EXPO profile and F10 to save and restart then it will run the ram on a divider (slower) for anything faster than 6000 Mhz. The fast and speedy way to get faster ram with AMD is to match AMD's Infinity fabric clocks to the ram clock speeds when increasing ram speeds (no divider) but that's a lot of manual ram clocking and not guaranteed to work with all processors and ram.

[Ontrix_Kitsune]

Ursprünglich geschrieben von r.linder:

You have no clue what you're talking about, you can't guarantee that and you're setting people up for failure. QVL does not guarantee stability

Yes it does. That is literally the entire reason that QVL lists exist and are published at all. As long as someone has the exact same ram kit that is in the QVL list for their motherboard, and is using the correct processor for that QVL list then it will always work in their system at the rated speed listed in the QVL list. 100% perfect stability always forever. That's why we have QVL lists: To ensure system stability by using tested ram configurations guaranteed to work with our motherboard and computers.

Ursprünglich geschrieben von r.linder:

it is merely what the motherboard manufacturer has tested and validated as bootable in the factory, doesn't mean it's going to work with every single configuration that uses the same motherboard and RAM kit.

Yes it does. As long as people follow the QVL list and configure their system correctly it will always work 100% perfectly with perfect stability.

Ursprünglich geschrieben von r.linder:

I've used kits on QVL before with AMD and Intel configurations alike and there was still instability issues, so I know for a fact that's a bunch of BS.

That's entirely your fault. You didn't configure the system correctly according to the QVL list, and/or you used the wrong processor that didn't match the QVL list.

Ursprünglich geschrieben von r.linder:

Still waiting for the actual proof I asked from you hours ago. I've seen benchmarks using much faster RAM with that CPU than 6000 and the difference wasn't even worth the price, it's nowhere near 20% in the majority of scenarios. You're arguing over a few frames.

Usually that's because they just grabbed the XMP/Expo profile in bios for the faster ram and then ran with it thinking they were configuring the system correctly. Yes if AMD systems use faster ram on a divider then it's not faster than using DDR5 6000 @ 1:1. If people configure their system correctly to match infinity fabric clocks to ram speed when increasing the ram speed then the benefits are significant. Unfortunately most people that review or test computer hardware don't understand how to do this with AMD because it's not a requirement with Intel so they just don't know about it or don't know how to do it.

[r.linder]

Ursprünglich geschrieben von Ontrix_Kitsune:

Bunch of repeated nonsense

No, it doesn't. You can say whatever excuse you can come up with, doesn't change the facts.

And no, I was following the QVL for Matisse processors because I was using a Matisse processor. It wasn't stable because the QVL isn't immutable, stability will vary from CPU to CPU as not all CPUs are created equal and that includes the IMC.

For the third time: where's the evidence that going well above 6000 actually makes a big difference? I've seen benchmarks running well over 7000 and the difference was as minor as people that actually know what they're talking about typically expect. If you can't actually provide links, then I'm just going to assume that you're nothing more than a troll and just trying to waste people's time.

[Ontrix_Kitsune]

Ursprünglich geschrieben von r.linder:

And no, I was following the QVL for Matisse processors because I was using a Matisse processor. It wasn't stable because the QVL isn't immutable, stability will vary from CPU to CPU as not all CPUs are created equal and that includes the IMC.

Did you make sure to get the EXACT model number of the ram kit EXACTLY as shown on the QVL list? Or did you just assume "Oh there's Corsair ram on the QVL list, any Corair kit will work with this board" ? To maintain stability with faster ram in the QVL list you must match the ram kit in the list EXACTLY. No assuming. If you don't do that then it won't be stable.

Ursprünglich geschrieben von r.linder:

I've seen benchmarks running well over 7000 and the difference was as minor as people that actually know what they're talking about typically expect.

I literally just explained that do you. When increasing the ram speed they must also increase the infinity fabric clock and make sure it is 1:1 matched with the faster ram speed to see any gains in performance. If people just go into bios and load the XMP/Exp profile for 7000 mhz and F10 and leave then benchmark it then they aren't doing it correctly. If you see any benchmark or review that is showing the same performance with 7000+ Mhz ram as 6000 Mhz ram on Ryzen then they didn't bring up the infinity fabric clock and they tested it wrong. That's a very easy indicator that who ever did the test didn't configure it correctly.

You stating that you think faster ram speed doesn't gain performance with Ryzen demonstrates that you don't even understand at all how Infinity Fabric clocks work on Ryzen. If you did know you wouldn't be saying that.

Ursprünglich geschrieben von r.linder:

I'm just going to assume that you're nothing more than a troll and just trying to waste people's time.

Speaking of wasting people's time: Every single thing you have written in this thread is complete nonsense. You know it is. You're just spouting random things in an attempt to mislead people and waste everyone's time on purpose. You're been trolling everyone in this thread since your very first comment here. I felt compelled to make sure and correct your nonsense so people won't actually believe your words.

[r.linder]

Ursprünglich geschrieben von Ontrix_Kitsune:

Trolling and nonsense

No, where is the actual link that shows actual data instead of you just yapping?

[A&A]

Ursprünglich geschrieben von Ontrix_Kitsune:

I literally just explained that do you. When increasing the ram speed they must also increase the infinity fabric clock and make sure it is 1:1 matched with the faster ram speed to see any gains in performance.

https://www.igorslab.de/en/ryzen-7000-tuning-guide-infinity-fabric-expo-dual-rank-samsung-and-hynix-ddr5-in-practice-test-with-benchmarks-recommendations/

Nope

[Ontrix_Kitsune]

Ursprünglich geschrieben von A&A:

Ursprünglich geschrieben von Ontrix_Kitsune:

I literally just explained that do you. When increasing the ram speed they must also increase the infinity fabric clock and make sure it is 1:1 matched with the faster ram speed to see any gains in performance.

https://www.igorslab.de/en/ryzen-7000-tuning-guide-infinity-fabric-expo-dual-rank-samsung-and-hynix-ddr5-in-practice-test-with-benchmarks-recommendations/

Nope

Check your tests again. That page shows that using a Ryzen 7000 series system with ram speed at 6400 Mhz 1:1 with matched infinity fabric clocks has a score of 19,073 in 3dmark time spy while 6000 mhz 1:1 matched infinity fabric clocks has a score of 16,185. Which is a gain of +17.84%. This is far more than "marginal" and definitely more than "just 2 FPS" like rlinder wants everyone to believe. It's even more significant comparing 6000 mhz to 7000 mhz when the infinity fabric clocks are matched.

Thank you for sharing that article with us so I didn't have to.

[r.linder]

"At RAM clock rates beyond 6000 Mbps, the UCLK automatically goes into 1/2:1 mode, better known as 1:2. For example, if you activate an EXPO profile with DDR4-6400, the corresponding values are 3200 MHz for MCLK and 1600 MHz for UCLK. The FCLK remains capped at 2000 MHz, at least as long as it is not changed manually. For RAM clock rates below DDR5-6000, an average value is automatically set for FLCK between the base clock of 1800 MHz and the maximum of 2000 MHz, e.g. 1900 MHz for DDR5-5600."

The fabric will automatically go out of 1:1 mode as soon as you exceed 6000 MT/s. That's by AMD's design and found through thorough testing done by igorslab. With all that yapping about people not understanding how the IF works, you've gone and proven that you don't even know it yourself.

Using 3DMark as a reference for getting "17%" for just 400 MHz is laughable, synthetic benchmarks are completely different from real world gaming performance and you still haven't provided an actual source that consistently gets you "+20%" more FPS or justified why people should be spending a lot more on DDR5-8000 and expensive motherboards geared for memory overclocking, since it's pretty much guaranteed that most AM5 motherboards are not going to be able to run that spec easily, if at all. You're essentially trying to force a 0.1% user demographic down people's throats when AMD recommends 6000 MT/s as the sweet spot and will give people far less trouble, especially on a budget.

The only people actually buying these really expensive RAM kits are enthusiasts, not normal users. And they're more often buying higher spec for Intel systems, not AMD.

[A&A]

Ursprünglich geschrieben von Ontrix_Kitsune:

Check your tests again. That page shows that using a Ryzen 7000 series system with ram speed at 6400 Mhz 1:1 with matched infinity fabric clocks has a score of 19,073 in 3dmark time spy while 6000 mhz 1:1 matched infinity fabric clocks has a score of 16,185. Which is a gain of +17.84%. This is far more than "marginal" and definitely more than "just 2 FPS" like rlinder wants everyone to believe. It's even more significant comparing 6000 mhz to 7000 mhz when the infinity fabric clocks are matched.

Thank you for sharing that article with us so I didn't have to.

lol
I shared only to see the AMD sheets saying that the infinity fabric is no longer 1:1:1
How about the benchmark? I saw a 13% improvement in 3dmark (who knows how their score system works) not 17%, but the caveat is to make the motherboard work with a 1:1 mclk and uclk ratio and OC IF by yourself. Heck, I can see that the OP certainly doesn't even want to do anything other than turn on EXPO. Also these benchmarks are made on a system with 2 dimms, not 4.
+ the frequency increase from 6000 to 6400 is not even 7%, so the rest of the math is meaningless.

Gaming benchmarks show it very well. The minimum FPS difference is on a par. 1+- is stupid.
https://www.igorslab.de/en/ryzen-7000-tuning-guide-infinity-fabric-expo-dual-rank-samsung-and-hynix-ddr5-in-practice-test-with-benchmarks-recommendations/9/

[Ontrix_Kitsune]

Ursprünglich geschrieben von r.linder:

The fabric will automatically go out of 1:1 mode as soon as you exceed 6000 MT/s. That's by AMD's design and found through thorough testing done by igorslab. With all that yapping about people not understanding how the IF works, you've gone and proven that you don't even know it yourself.

You're the one here claiming you know anything and everything about AMD and how they work so surely you do realize that we're supposed to set the clocks manually for higher ram speeds with AMD. Surely you know that. You know literally every single thing about AMD and how to tune AMD systems after all.

[r.linder]

Ursprünglich geschrieben von Ontrix_Kitsune:

Ursprünglich geschrieben von r.linder:

The fabric will automatically go out of 1:1 mode as soon as you exceed 6000 MT/s. That's by AMD's design and found through thorough testing done by igorslab. With all that yapping about people not understanding how the IF works, you've gone and proven that you don't even know it yourself.

You're the one here claiming you know anything and everything about AMD and how they work so surely you do realize that we're supposed to set the clocks manually for higher ram speeds with AMD. Surely you know that. You know literally every single thing about AMD and how to tune AMD systems after all.

I never said that, now you're just putting words into my mouth and making a strawman argument. You also completely ignored what A&A said.

[Ontrix_Kitsune]

Ursprünglich geschrieben von r.linder:

I never said that, now you're just putting words into my mouth and making a strawman argument.

I didn't make up anything. This entire thread and in every single comment you wrote in this thread was written with an air of confidence as if you seem to think you are the defacto expert about AMD. You have written your comments as if you know every single thing possible that there is to know about AMD. Every time I say something you claim I'm wrong and you are right. Since you are a complete expert on AMD I assume that you know exactly how to tune AMD Ryzen computers, how to set the infinity fabric clocks correctly (manually) and how to tune AMD systems to get the most performance out of every configuration. Including how to get extra performance by using faster ram when clocked correctly.

[76561199793899658]

all my cpu's for a long long while have been intel. (last time I went amd was with an k7 in 1999)
but well at least for these ryzens.. amds memory controller sucks.
guees they had to cut some corners..
never had these issues with intel...

and well this dives deeper into blueprint designs of more current cpus.. and ram than I used to go.. interesting..

still if I read this data right....

*below 6000mhz you loose optimal parity
*above 6000mhz your parity goes 2:1 and that hurts speed.
(but still it should run even if slower than 6000mhz ram no?)

*above 6000mhz to keep 1:1 parity needs to overclock the busspeed... which may or may not be stable upto 6400mhz.
->
there might be a slight possibility due 9800x3d stock rated for 5600 not 3600.. it might push a bit further than 6400??

64 gb dimms don'r excist yet as any set of 2 or 4.
the first non ecc sets of 1 and 2 are 64gb dimms are about to be released..

Crucial CT2K64G64C52CU5 (2 dimm)
Crucial CT64G64C52CU5

downside 1 : not yet for sale only announced.

downsite 2 : only announced as kits of 1 and 2.
-meaning you need at least 2 kitts of which is an unstability factor

downsite 3 : insanely slow true latency of 16.25 (where 10 or lower is the norm)

downsite 4 : 6400mhz which may not run stable.. meaning you have to underclock them which means even worse true latency..

so at the moment the market misses what we would need : a 256gb (4x64gb) 5600mhz cl28 kit
(which probably run fine)

so if I do want 256gb I need to wait till any dimms like that are on the market..

with 48gb dimms the pickings get better.

the best would be the yet unreleased (so not for sale yet)
G.Skill Flare X5 F5-6000J2838A48GX2-FX5

*downsite kit of 2.. so again need to use 2 kitts.

*but true latency of 9.33 the lowest of any 48gb dimms (and on par with the best 32gb dimms)

**it is however 6000mhz so it would be a question if 4 of those would run stable with a 9800x3d.

---
the best that can be bought right now.. is
Corsair Dominator Titanium RGB CMP96GX5M2B6600C32

with a 9.7 true latency..
but it is 6600mhz so most likely wont run stable with 9800x3d??
(especially since they are again kits of 2.. so you need 2 kits)

if I don't want to risk that..
Corsair Vengeance CMK96GX5M2B6000Z30
6000mhz, true latency 10 and amd expo (so no xmp needed)
still only kitts of 2 though..

if I really want a kit of 4..
there is only ONE 192gb kit (4x48) for sale (and none announced)
-
Corsair Vengeance CMK192GX5M4B5200C38
**
5200mhz so despite only having xmp it will most likely work with 9800x3d.
**
but with an true latency of 14.62 your performance would really be in the crapper..

so which of these would work with 9800x3d?

[r.linder]

You don't need such an absurd number of RAM with a 9800X3D, those figures are something you'd expect with at least a Ryzen 9 9950X/3D or Threadripper.

The absolute maximum that a gamer should realistically need is 64GB, there is almost never going to be instance where you need more than that, and at least 95% of the time, 32GB is enough as well. Getting more RAM than what you're actually going to use is generally a waste, getting a lot more RAM than what you're actually going to use is just a stupid waste of money that could've went into something else in the system.

With higher capacity you also generally make it a little bit more difficult to manual overclock beyond the rated spec, which is why the 256GB kits are as low as 5600, it's extremely difficult to get that much RAM running at a high frequency that a 2x16 or 2x32 kit would easily handle.