DDR5 vs. DDR4 in gaming perf.

[opheen]

The main issue with the OP question is most people don't have a 13900k with really good DDR5/DDR4 kits and motherboards that support both to actually play around enough with them to see if it matters with a 4090.


For someone who doesn't want to invest a ton of time a 13900k with a decent 7200 kit or a 7800X3D with a decent 6000 kit is likely the best option.

Ray traceing favor DDR5 for sure, but you need DDR5 2 x 32gb DualRank 6600++ or DDR5 2x16gb Singlerank 7400++ to be on pair or slightly faster than DDR4 2x16gb DualRank at 4000 c14 \ c15 Gear 1 and even higher to rly say that DDR5 outperform DDR4

 

[oxrufiioxo]

Feels like you're only realistically running 7200 on 1DPC boards, though. The very few Z690 XOC boards or the equally good stuff on Z790... Ouch... Still very happy with the 6400 on my board. It's quite snappy.

I only glanced at the Z790 Hero around launch I remember there being like 5-6 32GB kits at 7200 on the QVL when using a 13 gen processor

 

[Dr. Dro]

I only glanced at the Z790 Hero around launch I remember there being like 5-6 32GB kits at 7200 on the QVL when using a 13 gen processor

The kit I bought is the highest frequency one on the Z690 Ace's 13th Gen QVL, it failed to pass memory tests in XMP regardless of my CPU SA and VDDQ voltages

6600 failed less often but I only achieved stability at 6400. Granted I didn't spend days trying to figure out the higher clocks, but seemed like it was going to be very difficult so I decided to settle there for now.

 

[opheen]

Looking forward to Z790 Unify X that will be my first DDR5 board , i think and hope msi will wait with release until they have a noticeably better board than APEX - kingpin - tachyon

 

[Dr. Dro]

Looking forward to Z790 Unify X that will be my first DDR5 board , i think and hope msi will wait with release until they have a noticeably better board than APEX - kingpin - tachyon

@fevgatos has the Z690 Unify-X and he speaks well of it. The ACE is more of a robust, feature loaded E-ATX board for daily settings, so it's quite acceptable IMO that it doesn't go as far.

I only gave up because I thought the ROI on time spent wouldn't be worth it, and it was proving quote uncooperative with no traces or any given pattern to failures. After I found the SA voltage sweet spot (reduced from 1.2 stock to 0.9, even 0.95 was unstable), it'd run memtest for 3 hours before a problem, or have one manifest in just 15 minutes. Those are the worst to figure out.

 

[opheen]

@fevgatos has the Z690 Unify-X and he speaks well of it. The ACE is more of a robust, feature loaded E-ATX board for daily settings, so it's quite acceptable IMO that it doesn't go as far.

I only gave up because I thought the ROI on time spent wouldn't be worth it, and it was proving quote uncooperative with no traces or any given pattern to failures. After I found the SA voltage sweet spot (reduced from 1.2 stock to 0.9, even 0.95 was unstable), it'd run memtest for 3 hours before a problem, or have one manifest in just 15 minutes. Those are the worst to figure out.

ACE is a damn good looking board with a shitload of nice futures, Secondary and tertiary timings is where the big perfromance gain is anyway and will give you 90-95% performance of potensially max anyway.

 

[Dr. Dro]

ACE is a damn good looking board with a shitload of nice futures, Secondary and tertiary timings is where the big perfromance gain is anyway and will give you 90-95% performance of potensially max anyway.

Yeah, it's amazing! As for the timings, I'm just running the XMP set's CL34 preset. Over time once the whole "in love" period is over, I'll begin tweaking it out. It's just so fast, and I'm coming from a 5950X which is no slouch... I love it!

 

[opheen]

The kit I bought is the highest frequency one on the Z690 Ace's 13th Gen QVL, it failed to pass memory tests in XMP regardless of my CPU SA and VDDQ voltages

6600 failed less often but I only achieved stability at 6400. Granted I didn't spend days trying to figure out the higher clocks, but seemed like it was going to be very difficult so I decided to settle there for now.

Finding the sweetspot is the key, went from 4000c15 to 4266c16 stable and 4400c17 max boot able, exact 1.335 SA and 1.32 VDDQtx 0.05v up or down from that and i am unstable or issues with training

suprised over that it is day and night difference with 0.05v

 

[Dr. Dro]

Finding the sweetspot is the key, went from 4000c15 to 4266c16 stable and 4400c17 max boot able, exact 1.335 SA and 1.32 VDDQtx 0.05v up or down from that and i am unstable or issues with training

suprised over that it is day and night difference with 0.05v

The higher my SA voltage the more crash happy my 13900KS becomes. It's insane. Sub 1V range for it to be happy and the lower the better.

 

[opheen]

The higher my SA voltage the more crash happy my 13900KS becomes. It's insane. Sub 1V range for it to be happy and the lower the bettter.

yeah, i almost gave up after tryin out so many different volt combinations , and accept that my IMC was dogshit

My 13600k also does 4266 c16 with totally different voltages 1.345v SA 1.29v VDDQtx

 

[JustBenching]

Looking forward to Z790 Unify X that will be my first DDR5 board , i think and hope msi will wait with release until they have a noticeably better board than APEX - kingpin - tachyon

There will not be a z790 unify X most likely.

The higher my SA voltage the more crash happy my 13900KS becomes. It's insane. Sub 1V range for it to be happy and the lower the better.

For anything around 6000 use 0.95 SA. That's enough for my 12900k, it should be more than enough for your 13900ks

 

[Dr. Dro]

For anything around 6000 use 0.95 SA. That's enough for my 12900k, it should be more than enough for your 13900ks

0.95 crashes for me at 6400, 0.9 is fine. Really weird.

 

[JustBenching]

@fevgatos has the Z690 Unify-X and he speaks well of it. The ACE is more of a robust, feature loaded E-ATX board for daily settings, so it's quite acceptable IMO that it doesn't go as far.

I can't say it's the best board I've ever owned. It aces memory clocking considering it's a z690, but on the CPU side lots of features are lacking or not working properly. Even those that do (like using TVB to overclock / underclock based on temps) is much harder to do than the z690 apex I had.

Overall I'm happy of course, but MSI needs to show some more love to the bios. Either that, or im a complete tool when it comes to using their bios, cause I had similar issues back with the z490 ace

0.95 crashes for me at 6400, 0.9 is fine. Really weird.

Νever had that personally but I've heard some stories bot voltage resonance or whatever it's called, that SA voltage needs to "synchronize" with other voltages so one specific voltage might not work.


What are you testing stability with? TM5?

 

[Dr. Dro]

I can't say it's the best board I've ever owned. It aces memory clocking considering it's a z690, but on the CPU side lots of features are lacking or not working properly. Even those that do (like using TVB to overclock / underclock based on temps) is much harder to do than the z690 apex I had.

Overall I'm happy of course, but MSI needs to show some more love to the bios. Either that, or im a complete tool when it comes to using their bios, cause I had similar issues back with the z490 ace


Νever had that personally but I've heard some stories bot voltage resonance or whatever it's called, that SA voltage needs to "synchronize" with other voltages so one specific voltage might not work.


What are you testing stability with? TM5?

I've run bootable memtest86+ (4h), tm5 anta777 (all passes twice over), occt memory (2 passes), prime95 blend (2h) to validate.

Higher SA or DRAM clocks error out in bootable memtest, so didn't even move forward with those. As for VDDQ and VDD2 both are 1.35v

 

[JustBenching]

I've run bootable memtest86+ (4h), tm5 anta777 (all passes twice over), occt memory (2 passes), prime95 blend (2h) to validate.

Higher SA or DRAM clocks error out in bootable memtest, so didn't even move forward with those. As for VDDQ and VDD2 both are 1.35v

Wow that's overkill testing. Tm5 is usually enough - with 10 passes - and then you have to test something that also uses the GPU so your ram temps climb to the 55s+ and check stability there.

 

[opheen]

My friend have 13900KF z690i unify and g.skill 2x16gb 7800c36(A-Die) and he runs 8000c38 without any issues on z690i unify , we both run with e-cores disabled and static allcore oc , and yes msi had some rly interesting bios'es , Missing essential settings like RingDownBin, TVB , M.2 driver missing for EFI so if you only have 1 M.2 SSD you needed to add a driver too EFI so you where able to install Windows, etc
Karhu 10000% -TM5 ABSOLUTE 6 cycle- y-cruncher - It's good enough for what I use it for

 

[JustBenching]

My have 13900KF z690i unify and g.skill 2x16gb 7800c36(A-Die) and he runs 8000c38 without any issues on z690i unify , we both run with e-cores disabled and static allcore oc , and yes msi had some rly interesting bios'es , Missing essential settings like RingDownBin, TVB , M.2 driver missing for EFI so if you only have 1 M.2 SSD you needed to add a driver too EFI so you where able to install Windows, etc
Karhu 10000% -TM5 ABSOLUTE 6 cycle- y-cruncher - It's good enough for what I use it for

Yeah, the unify I and it's successor, the z790 edge I are better than the unisex for memory oc. Mine caps at 7600 and from what I've read, it's the exception, most of them hit a wall at 7200 - 7400

 

[opheen]

Disable Late Command Training , Enable RoundTripLatency , Disable fastboot and set fastboot to auto after it is stable helped too keep it stable for my friends z690i unify

Yeah, the unify I and it's successor, the z790 edge I are better than the unisex for memory oc. Mine caps at 7600 and from what I've read, it's the exception, most of them hit a wall at 7200 - 7400

Yeah the z790i Edge goes a bit higher, mby cuz it have 10Layer PCB and "unisex" 8 Layer ?

My B760i DDR4 8Layer PCB and 1DCP, i think i can get 4400c17 stable in gear 1 with 13700KF paired with vipers 4400c19 SingleRank
2x8gb, i am getting closer.

 

[JustBenching]

Disable Late Command Training , Enable RoundTripLatency , Disable fastboot and set fastboot to auto after it is stable helped too keep it stable for my friends z690i unify


View attachment 294430


Yeah the z790i Edge goes a bit higher, mby cuz it have 10Layer PCB and "unisex" 8 Layer ?

My B760i DDR4 8Layer PCB and 1DCP, i think i can get 4400c17 stable in gear 1 with 13700KF paired with a vipers 4400c19 SingleRank
2x8gb, i am getting closer.
View attachment 294431

Can't do anything now cause I have my 12900k, that one hits the wall at 7200. My 13900k is gone

 

[opheen]

7200 with 12gen is nothing to complain over

 

[ir_cow]

There will not be a z790 unify X most likely.

Yep. Your choices are Z700 Apex, kingpin and Tachyon. No ASRock Aqua or MSI Unify X coming. It's late in the cycle for anything new. I wonder if Intel will do another chipset fresh with the fall 2023 CPUs. Probably not.

 

[oxrufiioxo]

Yep. Your choices are Z700 Apex, kingpin and Tachyon. No ASRock Aqua or MSI Unify X coming. It's late in the cycle for anything new. I wonder if Intel will do another chipset fresh with the fall 2023 CPUs. Probably not.

Intel has been slightly confusing on one hand they say meteor lake is on track but we also have seen multiple rumors of a raptorlake refresh...... Not sure what to expect at this point.

 

[ir_cow]

My guess (truly a guess) is the fall is a Rapter Lake refresh. Same CPUs but with new SKUs. Fall 2024 is Metor Lake and new socket.

 

[opheen]

My guess points to Refresh since it soo quiet around what intel's next move.

Yep. Your choices are Z700 Apex, kingpin and Tachyon. No ASRock Aqua or MSI Unify X coming. It's late in the cycle for anything new. I wonder if Intel will do another chipset fresh with the fall 2023 CPUs. Probably not.

You 100% sure ? b760i DDR5 z790i Edge would be the only 700series boards from MSI with 1DCP(DDR5)

 

[80251]

DDR5 Memory Performance Scaling with Alder Lake Core i9-12900K

In this article, we take a closer look at the performance scaling you can expect for various DDR5 configurations. We test from DDR5-6000 all the way down to DDR5-2400 and compare CL30 vs. CL36 vs. CL40. Last but not least, we also consider these numbers in relation to what DDR4-3600 offers.

www.techpowerup.com

"In gaming, which is more sensitive to memory timings than memory bandwidth, our DDR4-3600 configuration works even better, achieving FPS similar to DDR5-5600. The takeaway here is to focus on DDR4 unless you absolutely want to play with the new DDR5 tech. Motherboards are cheaper, too, and you can use the DDR4 memory you already have."

That test only looked at average FPS, the techspot article indicated sometimes substantial differences in 1% lows, but I believe the conclusions were the same. Both articles used the 12900k too, maybe things would be different if tested with newer AMD DDR5 CPU's?