Intel 12th-Gen CPUs

[RasmusP]

I thought sim games like ACC were more limited by the GPU and CPU overclocking didn’t do much.

That's a whole different question and sadly it's generalised way too often.
Basically the cpu related fps won't change that much with graphics settings. They do a bit when dropping shadows from ultra to super low or when you disable reflections in the ini files or something.

But overall, not by that much.. Basically the cpu related fps are what they are depending on how many other cars are there and what cpu you have.

The gpu situation is completely different. Graphic settings and resolution make a massive difference.


Generalisations about the best mix of cpu and gpu are done with a certain level of quality at the average standard resolution.
That resolution is changing from 1920x1080 to 2560x1440 now.

And yes, acc will very often be limited by the graphics card but it's also quite difficult to get the minimum fps over 90 fps when using VR headsets.
You can always lower the graphics settings but even with everything at minimum, most CPUs will struggle to not drop to 70-85 fps in VR with more than 10 cars around you (no matter if AI or multilayer).

Which is my most people recommend the best cpu you can afford and then work with a graphics card that is decent enough for your needs.

More than 6 cores aren't needed for acc though (and rf2 or ac barely need more than 4 cores).

So I always recommend the 6 core CPUs (3600x/5600x/8600k onwards) and then overclocking them to the maximum.
Best bang for the buck and mostly sufficient enough.
You don't need the more cores from the more expensive CPUs.
You only gain fps from more cache and better binning and therefore higher clockspeeds.
Imo not worth the amount of money needed for them though!

 

[2stains]

My poor 10600k is my limiting factor in my 3080 triple screen build, 85fps pretty well fixed and I just raise resolution scale to get my GPU loaded.
Often wonder if a newer 12th gen is worth it for me.

 

[RasmusP]

My poor 10600k is my limiting factor in my 3080 triple screen build, 85fps pretty well fixed and I just raise resolution scale to get my GPU loaded.
Often wonder if a newer 12th gen is worth it for me.

I'd say no, not worth it.
If you really really want more fps and kinda need them to enjoy racing, then maybe.
But the jump isn't THAT big compared to the costs.
This DDR4 vs ddr5, win 10 vs 11 with the little e cores being sometimes good or bad..
The mobos are simply too expensive!

5600x would probably give you 95 fps instead of 85 fps and just dropped down to 225€ in Germany. Good b550 mobo = 115€.

12600k = 289€
Mobo = 200€+

That's 489€ vs 340€.
Although the 12600k would probably give you something like 105 fps instead 85 fps.

I'm in the same boat. The 10600k was an awesome upgrade after 9 years with my 2600k but it's still lacking...

I'm personally waiting for the AMD 6600x. Which will hopefully be based on a new socket that will be compatible with the 7600x and 8600x or whatever AMD will call them.

 

[Trebormoore84]

Depends on the Sim. If your talking about titles like iRacing, RF2 and Raceroom that only use one or two cores for graphics and are constantly CPU limited, that would be a resounding yes. Also your goal, like if you are running 240hz monitors and want to push them as fast as possible. The CPU and memory speed is what you need.

 

[Trebormoore84]

Regarding ACC it’s better optimised on the CPU and graphics requirements higher so more likely to be GPU limited. But if you have more headroom on the cpu via overclocking and memory timings when you lower the graphics settings there will be more headroom for more fps compared to lower clocked cpu. This is how the balance works.

 

[Trebormoore84]

Sorry I overlooked RasmusP posts and I’ve repeated pretty much what he already said.

Also to add on the actual op question I would spend as little as possible on a motherboard. The vrms on all of the z690 boards are overkill, seems to be marketing strategy this gen. As if you were doing 24hr rendering all core or something. I think it was the YouTuber “framechasers” who said the best value board for overclocking is the bog standard msi z690 pro-a. That guy is obsessed with cpu and memory overclocking and although he’s a bit unbearable he makes some very valid points. Had the best support for overclocked memory where gigabyte and ASUS were lacking. Bios updates fix this over time of course.

 

[blekenbleu]

I think it was the YouTuber “framechasers”

Thanks for that pointer; Gamers Nexus also liked the MSI Z690 PRO-A board,
but dislikes MSI BIOS menus. Having never undervolted and not overclocked in 10 years,
feedback about sim tuning with specific brands is of interest.

It seems generally accepted that all Z690 VRM suffice for 12600k overclocking.
Without overclocking, H670 boards seem adequate..

 

[RCHeliguy]

Just put the latest firmware update on my MB and I'm now running my 4000MHz DRAM at 4000MHZ ! And it is a bit faster

My CPU and GPU temps are great. No stress at all 36C CPU 56C GPU

Should I dare try AAC ?

 

[RCHeliguy]

I'm noticing my computer is not stable at 4000 MHz and I'm currently running 2666 MHz

BTW I am noticing a substantial difference in VR. Frankly it surprised me how much of a difference I'm seeing.

Playing In Death at 120fps my GPU was staying at about 80% of the latency needed to maintain that speed with no reprojection.

At 2666 MHz at 90fps my GPU is spending 23% of the time in reprojection.

Maybe there is something else going on, but that's a pretty huge drop in performance.

I'm curious if I can manually time this 4000MHz memory at 3600 or 3800 MHz or something close while staying stable.

 

[RCHeliguy]

FYI, I'm testing my system at 3800 MHz and so far so good, but I'll need to run it a while before I'll trust it. I'm getting 0.5% reprojection now.

Raptor Lake is coming in 3rd quarter (supposedly) and is expected to be 8-15% faster in single core speeds and 30-40% in multi-threaded applications.

I just saw that Raptor Lake will be LGA 1700 compatible, so it should drop right into an existing Z690 MB.

If someone were debating replacing a Z690 DDR4 MB with a DDR5 MB version with very expensive RAM vs. a straight CPU replacement when Raptor is , it would seem the CPU "should" yield more speed. But given the differences I've seen in VR just by adjusting the speed of Memory, I'll be very curious how this all plays out.

Say 3800MHz with an i9-13900K vs. i9-12900K. How much of a bottleneck will 3800MHz memory be with a RaptorLake? would DDR5-6400 MHz memory only net a few percent, or would it be more significant?

 

[Neilski]

Playing In Death at 120fps my GPU was staying at about 80% of the latency needed to maintain that speed with no reprojection.

At 2666 MHz at 90fps my GPU is spending 23% of the time in reprojection.

Maybe there is something else going on, but that's a pretty huge drop in performance.

FYI, I'm testing my system at 3800 MHz and so far so good, but I'll need to run it a while before I'll trust it. I'm getting 0.5% reprojection now.

Holy smoke, crazy-sized drop in performance indeed. At 3800 you get 0.5% reprojection, but do you mean still at 90fps? Would be interesting to compare the average latency for 4000 at 90 fps.
It's hard to believe that something else isn't going on.

Btw, I'm no expert on memory tuning, but I'm presuming that you didn't only change the frequency; i.e. either a slower profile was used (with different timings) or you manually set up timings...

 

[RCHeliguy]

With 4000Mhz using the supplied profile I had nearly 0% reprojection at 120 fps.

At stock 2666 120fps was 100% reprojection. 90 fps was 23%.

At 3800 using supplied profile, but with the speed dropped to 3800, 90fps was 0.3% and 120fps was about 2%.

More telling was that the GPU load dropped significantly with higher speed memory settings.

 

[Neilski]

At 3800 using supplied profile, but with the speed dropped to 3800, 90fps was 0.3% and 120fps was about 2%.

Well, some games can be much more sensitive "than usual" to RAM speed, and maybe this is one of them...

More telling was that the GPU load dropped significantly with higher speed memory settings.

Hmm, is that inverted? Or does it imply a lot of aborted frames being sent to the GPU?

 

[RCHeliguy]

I'm not sure what the deal is.
I've gone to a number of websites that show some differences on certain games where 2666 - 4000 could account for 10-15%, but with most being closer to 5%.

However VR games are not in their tests. With VR there is a lot of data being sent around, more than triple screens because they are static and not displayed based on where your head is pointing 90 times a second or more. I suppose IR Tracker type systems might impact that, but I don't see any benchmarks on triple screens with IR Tracker either.

I see clear improvements for fast DDR4 memory. I am using the XMP profile provided, but pulled back 200MHz from the 4000 that was unstable for me.

I'll need to test some things in Dirt Rally, iRacing Same tracks, cars etc, and see what VR fps has to say about them.

It doesn't look like I'm allowed to try 3900. It's 3800 or 4000 which makes sense since you can't buy 3900MHz memory.

I've also got 4 x 8Gb cards. I read that gives you more bandwidth, but higher latency.

I'm not claiming to be an expert on memory timing and performance, so I don't know if that is in any way related to what I'm seeing.

This motherboard claims to have profiles up to 5333MHz and beyond it says if the memory has an XMP profile.

If you look at this video using an i9-12900K like mine, it shows slower memory yielding faster framerates. Go figure!

I do wonder if some VR games are pushing enough bandwidth that it starts to become more important than the latency.

I could remove 2 of my 4 sticks to see if my 4x8Gb configuration is having any impact on this. I've also seen video showing a negligible difference between 2 or 4 sticks.

 

[RasmusP]

on certain games where 2666 - 4000 could account for 10-15%

Mostly the performance really drops when going below 3200 MHz. Here's what my ACC benchmarking with a big AI grid, doing one lap gave as results:

So using 2666 really kills fps from my experience!

It doesn't look like I'm allowed to try 3900. It's 3800 or 4000 which makes sense since you can't buy 3900MHz memory.

You can change some "multiplier" iirc. I'd need to check the bios.. But basically you can either use 100 or 133.
With 133 you should see 3866 MHz as an option!

I've also got 4 x 8Gb cards. I read that gives you more bandwidth, but higher latency.

I'm not sure about the latencies but from what I know, most 8 GB sticks are single rank, while most 16 GB sticks are dual rank.
If you put in 4x 8 GB sticks, you basically get dual rank too. 4x 16 GB sticks would give you quad rank.

While dual rank gives you more fps, for a reason I would have to look up, quad rank has too much overhead and slows you down.

However, 4 sticks are definitely less stable than 2 sticks. So you could test 2 sticks vs 4 sticks to find out if dual rank improves anything for you or try higher clockspeeds with only 2 sticks and check if it's stable.

 

[RCHeliguy]

I was able to get to 4000MHz with only 2 sticks.

Currently 2x16Gb sticks only goes up to 4600MHz. There are 5333MHz 8Gb sticks so far.

It looks like DDR5's biggest improvement is being able to address more memory. It has more bandwidth and latency, so it needs higher.frequencies to break even.

However, if like your testing my system has reached a plataeu and getting above 3800MHz yields little. There is no point. However I saw better performance at 4000, so.....

 

[RCHeliguy]

Just ordered some Kingston Fury DDR4 5333MHz memory just to see what happens.

I do realize this may be a complete waste of money to satisfy my curiosity.

Any bets on whether this is a waste or may actually bump my performance in VR ?

The memory is not expected to land for a week or so.

Edit: Changed my mind and cancelled my order. No point in dropping money when what I have is working well. I'll wait until next fall and decide what to do then.

 

[Deleted member 1066209]

Just ordered some Kingston Fury DDR4 5333MHz memory just to see what happens.

I do realize this may be a complete waste of money to satisfy my curiosity.

Any bets on whether this is a waste or may actually bump my performance in VR ?

The memory is not expected to land for a week or so.

Edit: Changed my mind and cancelled my order. No point in dropping money when what I have is working well. I'll wait until next fall and decide what to do then.

I think it would boost your minimum FPS, giving you a noticeably smoother and enjoyable experience in VR. My experiences are only limited to DDR4 @ 4266Mhz, however, so I may just be wrong.

 

[RCHeliguy]

I think it would boost your minimum FPS, giving you a noticeably smoother and enjoyable experience in VR. My experiences are only limited to DDR4 @ 4266Mhz, however, so I may just be wrong.

Very possible.
I think I'll wait until Fall and see if they have 2 x 16Gb in 5333MHz available vs. 4x 8Gb in that same speed. 5333 is an interesting speed because it ends up with a base speed of 2666MHz.
Then maybe I can drop in a Raptor Lake CPU for another 10-15% single core improvement since it's pin compatible.

 

[Spinelli]

Just ordered some Kingston Fury DDR4 5333MHz memory just to see what happens.

I do realize this may be a complete waste of money to satisfy my curiosity.

Any bets on whether this is a waste or may actually bump my performance in VR ?

I see you cancelled the order but still want to say to you and anyone else reading this thread that 5300 MHz DDR4 is almost certainly going to require running Intel 11th and 12th gen in gear 2 mode.

You also may have to pump VCCSA and VCCIO voltages to quite high levels which are possibly not too safe for 24/7 usage in order for the system to even boot at those speeds let alone run stably . I say "may" because you might not have to since it'll be in gear 2 mode which is much easier to run.

I'd personally recommend a 3600 - 4000 MHz dual-rank gear 1 setup, making sure to buy very good sticks - Samsung B-die - so you can gain lots of performance (bandwidth and latency) by lowering all the timings: primary, secondary, tertiary. Often, the "other" timings can make even bigger differences to performance than primary timings.