dgianstefaniWith a 4090 my dude.

I know, in CPU-limit, but in many games you don't need a 4090 to be CPU-limited, especially if your monitor "only" does 144-165Hz, like mine.

dgianstefaniget 25-30% lower latency.

An how does that translate into fps? How much time did you have to invest?

Here are my EXPO-timings, unoptimzied.

dgianstefaniwith a mid range GPU over a cheaper CPU/platform

Not Midrange, but if the price premium for a faster GPU is higher, then yes.

TigerfoxI know, in CPU-limit, but in many games you don't need a 4090 to be CPU-limited, especially if your monitor "only" does 144-165Hz, like mine.

An how does that translate into fps? How much time did you have to invest?

Here are my EXPO-timings, unoptimzied.


Not Midrange, but if the price premium for a faster GPU is higher, then yes.

Do you understand that a faster CPU provides literally zero benefit when you are GPU bottlenecked?

Thus, spending more on a CPU when you are GPU limited (7800XT) at ultrawide or high resolution gaming (with high settings) is completely, functionally, pointless.

dgianstefaniThus, spending more on a CPU when you are GPU limited (7800XT) is completely, functionally, pointless.

I understand, but I wouldn't buy a 7800XT.

TigerfoxAn how does that translate into fps? How much time did you have to invest?

You could get 80% of the improvement with a 5 minute tune copying some conservative timings and doing a quick stability test.

RAM speed can affect CPU performance in games by 10-30%, depending on the game. Tarkov for example and other simulation games are incredibly CPU/Memory bound.

E.g. a recent RAM review that saw a 10% FPS increase by going from 6400 XMP to 6800 tuned.

TigerfoxI understand, but I wouldn't buy a 7800XT.

Then what are you arguing about?

The context is my original discussion with AusWolf, i.e. CP at 1440p ultra at 45 FPS.

Modern games use 1-3 GB/s of bandwidth for RAM.

Literally everything is about latency. System latency, and CPU latency per frame. Your frametimes are much higher than your RAM latency (ms vs ns), but it's a complicated thing. Moving to 50-55 AIDA latency tuned compared to 60-65 ns stock (total memory) or 150 ns from 300 ns tRFC latency are huge improvements, and are directly measurable when you are CPU limited.

Even when you're not CPU limited, your min FPS goes up, which is almost guaranteed (excepting extreme examples 45 FPS Cyberpunk) to be noticable.

With modern architectures such as Zen, the infinity fabric is directly linked to more than just RAM performance, so by tuning latencies, you are improving all aspects of the CPU's performance, IO performance, etc.

TigerfoxThanks, but that benchmark isn't helpful at all. 1) It only test up to DDR5-6000 with low and high timings on AM5, but I was interested in the benefit of higher clockspeeds like 6200, 6400 and much higher.

Because those are the RAM speeds that you're pretty much guaranteed to run 1:1 with the IMC. Like I said, in 1:2 mode, most of your RAM speed advantage is diminished by the slow IMC.

Tigerfox2) It's only 1080p with a 4090 in memory-sensitive games, that run at 200fps+ without any use to it. Sure it will show high benefits but that doesn't mean I gain anything from DDR5-6000 CL30 in 1440p with a less powerful GPU, which is what most people have.
Again, I am ok with buying DDR5-6000 CL30 or even slightly faster RAM that works out of the box, which is what I did. I am ok with buying cheap RAM and OCing it to DDR5-6x00 CL30 if you have the time, which I don't have.
But unless I see perceptible gains in several games, in low fps for example, in 1440p or 3440x1440, with anything you can't buy for a few bucks above what DDR5-6000 costs, I'm not convinced it's woth my time. Even then, since I don't wan't to spend 1.000€+ on a GPU and thus will never have a 4090 or even 4080, I still will always be GPU-bound in 3440x1440.

Because you'll see literally zero difference with a slower GPU and/or at higher resolutions. I have a system like that, so trust me, I know.

@AusWolf : I agree on both points, that's the argument I'm trying to make, too.

Tigerfox@AusWolf : I agree on both points, that's the argument I'm trying to make, too.

TigerfoxMore than 20fps average above any other AMD CPU, even much more expensive ones, and the absolute top of the ranking above any unleashed Intel CPU is no reason? The 7800X3D is so good in many games, it's hard to justify not to buy one if you are gaming a lot.

Sure, if you're on really limited budget then it might be to expensive, as might AM5 altogether. But I would even skimp more on RAM than on the CPU if that would get me the 7800X3D, which is less dependent on fast RAM, as you said yourself.

Your argument is still to absolute. I am talking about investing time and/or money into RAM faster than DDR5-6000/6200/6400 CL30/32/34 with EXPO, which you can buy for not much more than any old DDR5-5x00 and which works out of the box. There is next to zero benefit to going faster on AM5 as of now and it's not worth the time needed to get it stable.

It's hard to say what to invest into when you're on a thight budget.

1:
The "more than 20 FPS than any other AMD CPU" is only when paired with a top end GPU, so these CPUs are being benchmarked in situations where they're actually the bottleneck. You're talking about "next to zero benefit" of having faster RAM (while confusing MT with "faster"), while seemingly missing the point that there's also literally zero benefit of pairing a top end CPU with a mid range card (besides the exceptions I mentioned), just spending more money on no tangible performance. No. You're not going to see "20 FPS more" just because you have a 7800X3D, your GPU actually needs to be able to put out enough frames that this difference becomes evident. It's like running out of memory. There's literally zero benefit of more RAM/VRAM capacity unless it's actually utilized. No different to having a faster CPU.
2. "Not worth the time to get it stable" What time? Changing a few values in BIOS? It's literally free performance. As opposed to how "worth" the time is spent gaming? This is an enthusiast forum, where people are into tuning.
3. It's not hard to say what you should invest in. You pick an appropriate resolution, frame rate target, and quality setting for your budget, then build a system that can deliver that. Pairing a CPU capable of 500+ FPS with a GPU capable of 100, if that, for the resolution and quality settings you'll be using, is a mistake. There's very clear, objective, and borderline obvious ways to spec a PC. RAM tuning/spec is a critical/free part of that decision making process.

Tigerfox@AusWolf : I agree on both points, that's the argument I'm trying to make, too.

Then why do you want to see reviews with such high RAM speeds? I mean, by agreeing with my points, you already acknowledge that it's pretty much pointless due to the IMC running at 1:2, so...? :confused:

How will the 3 chip solution works? PCIE4x4 is 64Gbps max theoretical. Will Zen5 have a separate SOC USB4 link?
AMD going for PCIE5x4 downstream?

@dgianstefani : So, what are you saying? That every gamer that doesn't have a 4090 should go for a 7500F or 7600? 7800X3D is only top end for gaming, it's not that much more expensive than 7700X and the price difference doesn't get you from a 4070S to a 4070TiS, let alone from there to a 4080S or 4090.
I was contemplayting either buying a 4070Ti, or a used 3080Ti or 3090, or maybe a 7900XT when I build my rig back in summer 2023. Since I had no time for gaming, I didn't buy anything and now that I'm back to living in one place all of the time, the RTX 2070 from my old build is sufficient for the old 1200p@60Hz monitor. It might not for 3440x1440@144Hz. If I had to buy now, I would choose 4070TiS for 16GB with good RT-capability, but I'll rather wait for RTX 5070.

I still don't understand when manual RAM tuning would bring tangible benefits. At 1440p, 3440x1440 or even UHD, nearly everyone with at least a Ryzen 5 5600 will be GPU-limited, especially with Raytracing, so they will not see any effect of manual tuning. I believe you the result's are visible in latency, I just don't believe there is a tangible effect on games at such high resolutions.
Only use I see is absolutely maximizing fps in a not GPU-limited game. Even then, I would bet you will benefit more from first optimizinh CPU-clock an OCing the GPU.

I will try out more agressive memory presets on my board, but I don't have the time to manually tweak timings. Even then, I won't benefit on my old monitor.

@AusWolf : because @dgianstefani keeps telling me they're worth it. I don't believe so and I have yet to even see valid benchmarks showing tangible differences for either DDR5-6000 with tighter timings or anything faster in 1440p and above.

@danc : What are you talking about? Up to recently, USB4 on AM5, either onboard or via AIC was in fact TB4 via JHL8540 "Maple Ridge", which is Gen3x4. Since recently, there are AIC with real USB4 via Asmedia ASM4242, which is Gen4x4. I have yet to see benchmarks, but it might be slightly faster than JHL8540.
X870(E) and B850(E) will get one of those, so no problem linking those to the promotory21-chipset, as has been done one some high-end X670E-boards. There is no actual controllerchip known to be in developement with USB4v2 yet, but it recently became known that Intel's first TB5-controllers, Barlow Ridge, will only be Gen4x4. I don't know how one is to transfer 80Gbps bidirectional, let alone 120Gbps unidirectional over a 64Gbps PCIe-connection, but we don't actually know how efficient Thunderbolt is, because it may always have been limited by PCIe-connections.

However, AMD has desginated one Gen5x4-link iin Raphael-CPUs for USB in some block diagramms, which until now has been used for M.2 on almost all AM5-Boards.

Tigerfox@AusWolf : because @dgianstefani keeps telling me they're worth it. I don't believe so and I have yet to even see valid benchmarks showing tangible differences for either DDR5-6000 with tighter timings or anything faster in 1440p and above.

Ah, I see. :)

Well, that depends on your use case and rest of system specs, I guess. Having a 1440 UW screen and a 7800 XT, it is definitely not worth it in my case. But then, neither is the 7800X3D - I only bought it because YOLO.

If I could go a year back in time, I'd just buy a 7700 (non-X) and call it a day. It's more than enough for any sort of casual gaming, and it's way easier to cool than any X or even X3D model.

TigerfoxI will try out more agressive memory presets on my board, but I don't have the time to manually tweak timings.

@dgianstefani : I don't seem to have any BIOS-settings to automatically tighten timings a bit (there is no difference between EXPO I, EXPO II and EXPO enhanced).

Tigerfox@dgianstefani : I don't seem to have any BIOS-settings to automatically tighten timings a bit (there is no difference between EXPO I, EXPO II and EXPO enhanced).

Don't bother with auto stuff. You run the risk of too high voltages or unstable settings. Leave it.

If you aren't comfortable doing these things manually don't do it.

Tigerfox@danc : What are you talking about? Up to recently, USB4 on AM5, either onboard or via AIC was in fact TB4 via JHL8540 "Maple Ridge", which is Gen3x4. Since recently, there are AIC with real USB4 via Asmedia ASM4242, which is Gen4x4. I have yet to see benchmarks, but it might be slightly faster than JHL8540.
X870(E) and B850(E) will get one of those, so no problem linking those to the promotory21-chipset, as has been done one some high-end X670E-boards. There is no actual controllerchip known to be in developement with USB4v2 yet, but it recently became known that Intel's first TB5-controllers, Barlow Ridge, will only be Gen4x4. I don't know how one is to transfer 80Gbps bidirectional, let alone 120Gbps unidirectional over a 64Gbps PCIe-connection, but we don't actually know how efficient Thunderbolt is, because it may always have been limited by PCIe-connections.

However, AMD has desginated one Gen5x4-link iin Raphael-CPUs for USB in some block diagramms, which until now has been used for M.2 on almost all AM5-Boards.

I believe JHL8540 and subsequent USB4 controller, is not "free", they take up the limited PCIE lanes from your consumer cpu.

@danc : That's what I said. PCIe-lanes from CPU or chipset. I would think chipset in almost every case. But neither JHL8540 and ASM4242 available now, nor Barlow Ridge available in the future require Gen5x4, so Gen4x4 (Gen3x4 in case of JHL8540) from chipset is enought.
TB4 and USB4 respectively are free on Intel and AMD mobile CPUs and even AMD 8x00G Desktop APUs (only usable on one Gigabyte B650E-board, yet), though.

Tigerfox@AusWolf[/USER] : because @dgianstefani keeps telling me they're worth it. I don't believe so and I have yet to even see valid benchmarks showing tangible differences for either DDR5-6000 with tighter timings or anything faster in 1440p and above.

PlayStation / XBox calling your name :)
You just didn't need anything above them.