The AMD Ryzen All In One Thread /Overclocking/Memory Speeds & Timings/Tweaking/Cooling Part 2

Seems about right, x3d cpus run lower clock speeds. My cpu boosts to 5.65ghz even in games and in cpu heavy games like Spiderman drops to 5.55ghz. It should get more stable and better when my arctic liquid 360 gets here.

Did you try playing with pbo+curve optimizer? These are my settings.

 

Interesting discussion about 7600x3d and if amd will ever release one. Its just a 7900x3d without non v-cached ccd but gaming performance would be the same.

 

That's cool you're able to get -30 on your curve optimiser, I can do -30 fine running benchmarks and run Prime95 but crashes within minutes playing RDR2. For now settled on 20 and everything seems fine (cbf'ed wasting too much more time on it for now).

Only just finished my 7700X/B650E-E build too this week and man never had this much stress building and setting up a PC in my life.

 

1.6v ddr5? I smell burning from here.

 

Some games are just funny when it comes to stability, i remember BF4 being a beast for stressing OCs (CPU or GPU i cant remember).
For me with my 6900xt it was Metro EE, i could run every stress testing 3D bench no issue but Metro EE would sniff out any weakness.

 

Absolutely. Now that I've got the system setup properly (I think haha) I am enjoying it quite a bit. Even at stock on air cooling I was able to get 5.5Ghz boost in game and +4.9GHz all core in Cinebench (and now 5.15GHz with the curve optimiser setting applied), very nice to see coming from the 3700X which struggled to boost to the advertised 4.4Ghz. Gaming's buttery smooth at 1440P with the RTX 3080.

Currently tweaking ram as my setup refuses to boot to the XMP 6200MHz speed so had to drop it down to 6000MHz and now trying to tighten up timings instead. Copied Buildzoid's secondary/tertiary timings settings for 6000MHz SK Hynix memory and I was genuinely surprised to see it making actual differences. CP2077 for example at 1440P before tweak I can run at max settings with psycho RT and DLSS on quality and keep fps mostly between 50-60 with occasional dips down to mid 40's. After tweak now I don't see fps go below 50 which is awesome.

 

My Asrock 570x Taichi is better at tight timings than it is pushing the RAM speed. The corsair 3200 I had didn't like running much above 2933 but it it did so at tight timings at 1T. Now I have 3600mhz RAM im at 3200 but again very tight timings.

 

Yea seems like the Ryzen platform particularly AM5 has issues with higher memory speeds...going to see if I can get my kit down to cl30 haha which I am sure will be faster than my stock XMP at 6200MHz anyway.

 

I've been lowering some voltages and this is the result so far.

 

Its the zen4 limitation so with zen5 and ryzen 8000 imc will be better and memory frequency will go up and we know platform likes fast memory. Thats why i went with 7600x this year as a place holder and will get 8700x or 8800x3d for much more performance next year.

 

When will AMD not have limitations? I've been hearing about memory speed limitations since Zen+ Figured they'd be up-to-par by now.

 

The issue with Zen is not the memory controller, t's the Infinity Fabric.
CPUs like the 5700G can clock memory as well as Intel on DDR4.
Buildzoid claims the best DDR4 memory controller is AMD APUs.

 

Like @Horus-Anhur said its actually infinity fabric limitation to be exact as it has to run 1:1 with the memory.

2100mhz is the max as it right now thats 6200 but i could not make mine stable. In some games it is and in some it crashes so backed down to 2000/6000, for now.

 

There is something that doesn't quite sit right with that, and that is that the data fabric is double data rate(256bit*2) on zen4, so at the stock 1800mhz fabric , the max theoretical bandwidth, would be 115gb/s , 128gb/s for an fclk of 2000, there does appear to be some sort of penalty from amd running the memory controller and the fabric asynchronous, but it doesn't really explain the huge delta, with the uclk and mclk at 1:2 (equivalent to gear 2 on intel ) bandwidth should still scale at least, though I've never actually seen anybody test that, and it doesn't appear that doing so actually yields much increase in maximum achievable memory speed. so there must be something else going on, by comparison the 5000 series apus can achieve very high memory speeds in 1:2 mode(6000mt/s+), but there isn't bandwidth to be gained since the data fabric isn't double data rate so even at 2400mhz fclk you're limited to 76.8gb/s ,

 

I don't know what is happening, but I noticed that since Zen4 launched.
Zen CPUs with 2 CCX have much higher memory bandwidth than single CCX.
Recently some people found a few 7600X with dual CCX, and this had a lot more memory bandwidth than the 7600X with a single CCX.

 

I tried some more tuning last night and interestingly enough changing the primary timings didn't seem to have the same impact (at least from just eyeball test watching frame counter in game using RIvaTuner) as it did when I first adjusted the sub timings following Buildzoid's settings...then I stumbled on another video of his ranting about why primary timings don't matter lol.

I'll preface by saying I didn't spend too much time because it's getting late and I have to get up early for work (20 hours over this weekend *yawn*) but keeping voltage at 1.35V I first tried dropping CAS to 30 (from 36) and ran a single pass of MemTest86 which passed with no errors. I then returned the CAS back up to 36 and tried to drop both tRCD and tRP (I know, lazy) down to 36 from 39 which booted into Windows fine and quickly ran a few games fine before starting Memtest86 and letting it run before going to bed. Woke up this morning and saw that the test completed with 12 errors which all happened in test 7. Any thoughts? I'm not particularly experienced at tweaking particularly with ram haha.

 

I have no idea about erros in Memtest86.
If it was with TestMem TM5 1usmus, it would be easier to know what these errors mean.