I am new to the whole Infinity Fabric issue.
On the subject of memory for a 5800X3D what would be your recommendation for 64GB?
I read all kinds of suggestions, 4x16, 2x32, SR, DR, etc… Thought maybe you could give me the best solution with price/performance in mind.
Thanks
I would go for a 32x2 GB DDR4 3600 CL16 kit - it’s the best in terms of performance vs value. Getting a 4x16 GB kit might be cheaper, but running 4 sticks is harder on the memory controller and not gauranteed you can hit 3600 Mhz stable. The whole single rank vs dual rank is irrelevant if you’re wanting 64 GB total capacity, as almost all 16GB and larger RAM sticks are dual-rank. Running with 2x dual-rank sticks or 4x single-rank sticks will give you the best performance due to memory interleaving.
Most Ryzen 5000 with a 1:1 coupled infinity fabric (FLCK=MEMCLK) can handle up to DDR4 3600 Mhz easily, while 3800 Mhz can work on some chips, and quite rare to reach up to 4000 Mhz stable. It’s technically possible to run even higher memory speeds for Ryzen 5000, but this would require you to uncouple the infinity fabric (so FLCK doesn’t equal MEMCLK) and imposes a significant latency penalty. There is not much performance gain from running beyond DDR4 3600 for Ryzen 5000.
Spot on analysis. It’s exactly why 16GB (8x2) of my PC-3200 samsung b’s could run stable at CL16 4000MHz (4400 unstable) yet with 4 sticks I’m limited to 3333MHz with more or less the same CL14 jedec timings they came with.
Many thanks for taking the time to answer.
Am I correct to think that:
In theory, performance wise, 4x16gb SR memory sticks @4000 would provide the best use of IF without special settings,
but these are rare & expensive, so from a value perspective, going for 2x32 @3600 CL16 is the sweet spot?
Just want to make sure I understand the trade off.
If I was building a new ddr4 rig 2x32 @3600 CL16 is exactly what I would choose now (but I would still research and pay a little extra for a reputable brand if I had to).
Not exactly. You will likely not be able to hit 4000 Mhz with four sticks (four sticks = more stress on IF/memory controller) assuming your Ryzen chip can even run the IF at such a high frequency. If you’re interested in hitting 4000 Mhz, I would purchase a 2x32 GB DDR4 4000 kit. Worst-case scenario, you can downclock the memory to 3600 or 3800 Mhz if your Ryzen IF cannot handle this frequency.
Yes, as pretty much every Ryzen 5000 chip can hit 3600 Mhz stable.
I would agree with the first part - get 2 x 32 gb of 3600 mhz, this way when you want to upgrade to 128 gb, just buy 2 more sticks from the same manufacturer. As for “harder on the memory controller”, if you’re not overclocking, it’s designed for this. I’m running 128 gb at XMP2 profile at the stock 3600 mhz with no problems - it’s all within the design spec.
Just don’t use an older motherboard - many of them weren’t designed for more than 64 gb. Check with the mb manufacturer. There are even people still running older motherboards limited to 32 gb of ram, so just because you can physically stick it in the slot is no guarantee.
Given that modern operating systems (Linux, FreeBSD, Windows) all use any extra ram as a hidden “memory object cache” to improve performance, it’s included in the OS, might as well benefit from it. Ram is dirt cheap compared to during the chip shortage, when I bought my 128 gb.
Might also explain part of why I’m able to run the sim at 16xk x 4k resolution (see the pics in the 770 discussion).
No, as has already been explained to you abundantly clear in the past, four sticks is not ideal for running 3600 Mhz. Just because your memory controller is able to handle it, doesn’t mean others are able to. You can find posts all over Reddit with users complaining they can’t hit the XMP frequency with four sticks - I’d suggest you research the topic to be better informed.
I have already linked the relevant parts from the DDR4 overclocking guide directly going against what you keep putting forth. It’s common knowledge that four sticks is harder on the memory controller. Just because you seem to be unaware of it doesn’t mean it’s not factual.
The official supported DDR4 frequency for Ryzen 5800X3D is 3200 Mhz, so yes, running 3600 Mhz is an overclock - regardless of 2 or 4 sticks.
XMP is an overclock and not a right, although suppliers are generally sympathetic they are under no obligation should your independently aquired ram not be able to run at advertised XMP speeds in any particular rig. And as mentioned already, there are also many users that can use XMP but not with more than two occupied slots.
End of discussion.
Seriously? XMP is an Intel specification. They clearly state that running an XMP profile is not “overclocking.” It’s allowing the RAM to run with fewer wait states, enabling it to reach it’s full potential at it’s rated speed.
That many people “think” it’s overclocking is easily debunked - just go into the bios - running at the rated speed (in my case 3600 mhz) the actual clock speed using no XMP profile, XMP1 or XMP2, the clock speed does not change. Can I overclock it? Sure, but why would I? It works just fine at it’s rated speed.
And where in the world do you get “not more than 2 occupied slots?” Upgrade your motherboard to this decade, please! And your CPU. And your RAM. Pushing outdated information is pushing misinformation. People shouldn’t be making buying decisions on misinformation. And we’ve had a lot of that lately, especially surrounding the Arc, from people who never bought one or even tried it themselves.
For the record:
A. Kingston Black Fury DDR4 - 4 x 32 gb, rated at 3600 mhz, running at 3600 mhz. XMP profile 2, which is part of Intel’s approved spec for that particular ram - no overclocking involved, just reduced wait states.
B. New Asus motherboard bought last spring.
C. 2 x Arc 770LEs.
Posted pictures running at 16k x 4k in the 770 discussion forum. Because I knew that the 4090 is a dead-end, inefficient product. They failed to get the expected performance from the die shrink, and the internals will require the next card, if they don’t drastically switch their rendering methods, to have 1600 watts of power and between 48 and 96 gb of VRAM to get a 30% jump in performance. Which is why they haven’t announced a roadmap for their video cards, unlike AMD and Intel.
If Intel says my specific RAM supports XMP2, then it supports XMP2. In all 4 slots (they say not to mix-n-match different manufacturers). End of story. So yes, it’s a right of anyone who bought XMP2 Ram certified by Intel to run at XMP2, and if it doesn’t then it’s defective and should be returned under warranty.
Straight from Intel:
Intel® Extreme Memory Profile (Intel® XMP) lets you overclock1 compatible DDR4 /DDR5 memory modules to enhance the gaming features built into PCs with Intel® Core™ processors.1
Because of headroom. At 4K, you load up AI + complex airport + complex airplane + heavy weather, you will want as much cpu processing power as you can get. THAT is why the cpu tests at 1080p matter, it shows which cpu can put out the most raw cpu power in MSFS.
Obviously but hardly the point if I don’t actually play at 1080p. I am not discussing absolute performance here but bang for bucks … especially in the case of the 5800X3D if it means you must also purchase an otherwise defunct mobo.
It is precisely the point, as I mentioned specifically the use case of 4K.
Of course but I’m not a heavy hitter in that respect. Although my frames do drop considerably when low over PG it’s never below 35fps even in 4k (with dlss). I wouldn’t shell out an extra €200 just for this one scenario… Bang for bucks like I said.
I currently have strix 3080ti. Would 4070ti be worth the upgrade?
How deep are your pockets? … Frame generation would double your fps (with some anomalies) but if you are already happy with what you have there’s little point if you would really miss the cash.
I could afford a 4070ti
^The above is what I was responding to. And of course the answer is they should not. However, in this case, the premise that the cpu does not improve performance at 4K is still questionably unknown. You are basing this on the 4K chart. That would seem to be a correct way of looking at it. However, you and I do not know how loaded the test scenario was. Did they simply plop down at KAFW, taxi out, take off, and call it a day?
I want to see a test scenario with known cpu-intensive MSFS items – AI + add-on aircraft + add-on airport before I make the statement of, “Why would anyone spend x amount of extra dollars for a cpu that doesn’t improve performance at the resolution they play at?”
Anyway, I’m waiting for the more economical 7800x3d to drop along with actual user-reports before doing anything.