Share your AIDA 64 cache and memory benchmark here

[freeagent]

Thanks for the kind words man, I appreciate it. I don’t think it’s a record, if you gave me your CPU it should do the same thing..

 

[Space Lynx]

Thanks for the kind words man, I appreciate it. I don’t think it’s a record, if you gave me your CPU it should do the same thing..

I meant the ram FCLK at 2100... and stable. I never seen a FCLK that high before.

 

[freeagent]

I meant the ram FCLK at 2100... and stable. I never seen a FCLK that high before.

Oh no it’s not stable.. I can run stuff but 24/7 probably not a good idea. That’s why I was asking about that new agesa update.. I can pass everything at 2k but error at 2033 and up in occt

 

[Space Lynx]

Oh no it’s not stable.. I can run stuff but 24/7 probably not a good idea. That’s why I was asking about that new agesa update.. I can pass everything at 2k but error at 2033 and up in occt

ah ok, dang, that would been awesome. i just run XMP these days. 3200 cas 14-14-14 - then i reboot, go into bios, change voltage to 1.310v for extra cool temps. and its all stable still. /shrug

 

[tabascosauz]

I meant the ram FCLK at 2100... and stable. I never seen a FCLK that high before.

You must have not read very far back in this thread lol, it's pretty much all we do here:



Ask @glnn_23 nicely and the madlad will probably be happy to provide a 4533 1:1 run for you:

Running a different Ryzen 4650g here in an Asus ROG Impact.


View attachment 200283

 

[freeagent]

Oh yeah it’s fast just at stock. That’s what my ram is, 3200 c14. You have me questioning whether or not I should sell it lol. I have buyers lined up for my 3600xt and 5600x tomorrow.

 

[glnn_23]

I have sold my Ryzen 4650g now but here is some stability testing I did at 4533c16 1:1

 

[oobymach]

Swapped my 16gb b-die kit back in, here's where I'm at now, and it's almost stable.

Update: couldn't boot at 12tcl but did lower some of the others a bit and passed some p95 and memtest5 block mirroring but like every other test if I run p95 and load 50 tabs in firefox while playing a music vid in one tab it skips but at this point I'm not sure if it's an issue with firefox since I get it even at super stable settings so going to run with these for a bit and see how things go.

 

[tbob22]

Update on my DR E Die, getting very close to the limit at 1.4v. Tested nearly every subtiming until unstable. Similar speed to previous 1t settings but now quite stable (8h TM5 Extreme, 8h P95 LargeFFT, 16h MT86).

 

[Taraquin]

Update on my DR E Die, getting very close to the limit at 1.4v. Tested nearly every subtiming until unstable. Similar speed to previous 1t settings but now quite stable (8h TM5 Extreme, 8h P95 LargeFFT, 16h MT86).View attachment 202252

Try trrds 7, trrdl 9, tfaw 28, trtp 8, that should work and improve perf. I would consider loosening twrwrdd to 7 and trdrddd to 5 as they can increase volrage req but do little to perf.

 

[freeagent]

Just messin around..

 

[dany8988]

Hi, these are Crucial Ballistix Max RGB 4400 CL19, but my Asus z490-f won't boot and i have relaxed timings to CL17 and frequency at 3900.

 

[Mathragh]

Using Hynix CJR B2 sticks at 1,4V.
First serious attempt at getting better latencies and speed, as this chip profits a lot from better ram.
Used DRAM Calculator to guide me a bit, although it only supports my RAM up to 3800 so i just took those timings wherever possible. Tried to lower most of the main timings, didn't seem to get much better than this.

Anyone got any tips for potentially improving things further?

 

[tbob22]

Try trrds 7, trrdl 9, tfaw 28, trtp 8, that should work and improve perf. I would consider loosening twrwrdd to 7 and trdrddd to 5 as they can increase volrage req but do little to perf.

Unfortunately it seems to be a bit slower with those settings (trrds 7, trrdl 9, tfaw 28, trtp 8), seems to be stuck around 53900-54100 read and 53000-53200 copy. Tried 10+ runs, rebmooting, making sure nothing else was running, etc. and couldn't get the same results as previous timings, reverted back and it immediately jumped back up.

 

[tabascosauz]

Using Hynix CJR B2 sticks at 1,4V.
First serious attempt at getting better latencies and speed, as this chip profits a lot from better ram.
Used DRAM Calculator to guide me a bit, although it only supports my RAM up to 3800 so i just took those timings wherever possible. Tried to lower most of the main timings, didn't seem to get much better than this.

Anyone got any tips for potentially improving things further?

It's CJR, 4133 17-19-21 @ 1.4V is already pretty admirable. tRFC ~250ns is also basically the limit for CJR. It's your CPU that is running wack; I don't know why you're not running synced, I've not yet seen a 4650G/4750G that can't do at least 4200-4400MT/s at 1:1, and you're not even using the iGPU so the IF load is lower too. Just don't limit yourself to 1.1V VSOC.

Also, what's up with VDDP in the screenshot?

As for the "a lot from better RAM", that is generally only true of the iGPU. When paired with a Vega 56 you won't see nearly as much performance improvement if any at all, Renoir L3 is still small.

 

[freeagent]

Well.. I don’t think my cpu enjoys ventures up to the 2K area very much.. it can do it.. but I will wait to explore what it needs voltage wise.. don’t want to thrash it too much just yet.. must wait for the new stepping to drop before I do that

I kid..

But this cpu can handle 4 sticks up to 1900.. so that’s pretty decent..

Also testing PBO +150 -30 co.

 

[Mathragh]

It's CJR, 4133 17-19-21 @ 1.4V is already pretty admirable. tRFC ~250ns is also basically the limit for CJR. It's your CPU that is running wack; I don't know why you're not running synced,

Thanks! With whack and out of sync you mean the MCLK, FCLK and UCLKC? I have most power saving features still turned on. I suppose I'll get better benching scores with everything running full tilt all the time? Would safemode work?

I've not yet seen a 4650G/4750G that can't do at least 4200-4400MT/s at 1:1, and you're not even using the iGPU so the IF load is lower too. Just don't limit yourself to 1.1V VSOC.

I currently have VSOC on auto as manually setting it to 1.1 doesn't boot at all. Still have to find out what's up with that, or try higher voltages I suppose.

Also, what's up with VDDP in the screenshot?

Don't know, the added HWiNFO screenshot shows it correctly.

As for the "a lot from better RAM", that is generally only true of the iGPU. When paired with a Vega 56 you won't see nearly as much performance improvement if any at all, Renoir L3 is still small.

Ideally I'd have both the iGPU and Vega56 available for use in Windows, but I've yet to find out how I can let windows see them as switchable graphics.
I always figured with a smaller amount of L3, DRAM speed and latency matters a bit more since the CPU needs to hit the RAM more often.


Thanks for the feedback!

 

[Taraquin]

Unfortunately it seems to be a bit slower with those settings (trrds 7, trrdl 9, tfaw 28, trtp 8), seems to be stuck around 53900-54100 read and 53000-53200 copy. Tried 10+ runs, rebmooting, making sure nothing else was running, etc. and couldn't get the same results as previous timings, reverted back and it immediately jumped back up.

How was latency? Tried tWR 12, tRTP 6? Read and write speed doesn't always tranfer to real life performance, but it is unusual that liwer trrd and tfaw decreases perf.

 

[tabascosauz]

Thanks! With whack and out of sync you mean the MCLK, FCLK and UCLKC? I have most power saving features still turned on. I suppose I'll get better benching scores with everything running full tilt all the time? Would safemode work?


I currently have VSOC on auto as manually setting it to 1.1 doesn't boot at all. Still have to find out what's up with that, or try higher voltages I suppose.


Don't know, the added HWiNFO screenshot shows it correctly.

Ideally I'd have both the iGPU and Vega56 available for use in Windows, but I've yet to find out how I can let windows see them as switchable graphics.
I always figured with a smaller amount of L3, DRAM speed and latency matters a bit more since the CPU needs to hit the RAM more often.


Thanks for the feedback!

I see. From the AIDA results I'm not sure if it's 1:1. The APUs are peculiar in that when IF is desynced, they keep the full R/W bandwidth, but still the Copy bandwidth suffers and so does latency. But Copy is usually a bit lower as well, so it's unclear.

Here's 4000CL16:



But yeah Renoir seems to be notorious for downclocking FCLK all the time. I have DF Cstates disabled and SOC/Uncore OC enabled, and it still fluctuates all the time.

Do disable TSME in your BIOS though (I think buried somewhere either in AMD CBS or AMD OC), it accounts for 5-10ns of latency. You should be in the 50ns range with your settings.

 

[Mathragh]

I see. From the AIDA results I'm not sure if it's 1:1. The APUs are peculiar in that when IF is desynced, they keep the full R/W bandwidth, but still the Copy bandwidth suffers and so does latency. But Copy is usually a bit lower as well, so it's unclear.

Here's 4000CL16:

View attachment 202515

But yeah Renoir seems to be notorious for downclocking FCLK all the time. I have DF Cstates disabled and SOC/Uncore OC enabled, and it still fluctuates all the time.

Do disable TSME in your BIOS though (I think buried somewhere either in AMD CBS or AMD OC), it accounts for 5-10ns of latency. You should be in the 50ns range with your settings.

Thanks!! that's some great info

I believe I did specifically disable TSME after reading about it somewhere, but I'll dig around and see what else can be done.

Edit: @tabascosauz
After tinkering for too long when I should be doing something else:
Apparently my motherboard doesn't automatically sync the FCLK with the mem clock, and so it has never run on the same frequency.
When I do manually sync the frequency, suddenly the system is not stable at all anymore. It's probable that the DRAM has actually never really been properly stressed because of the desynced clocks, and only now it shows.

Sadly that in effect means that I have to do the whole overclock all over again, as I cannot trust my previous results.

BTW VSOC on my motherboard only reliably is set via the CBS menu by editing the hex value it seems, the tooltip tells me that this value will be used above certain FCLKs.

Currently with normalised tRCDWR and tRC timings I can get up to 3933MHz with properly synced clocks, but 4000MHz instantly fails memtest at the first stage during address checks.

At least the latency is already much better! eventhough the read and write values are worse. Don't mind the weird cache values, pc wasn't completely idle during that test.

Thanks again for putting me on the right track.

 

[tbob22]

How was latency? Tried tWR 12, tRTP 6? Read and write speed doesn't always tranfer to real life performance, but it is unusual that liwer trrd and tfaw decreases perf.

Similar, but seemed to be slightly higher on average. Well, tfaw was 20 before (but now looking at it closer, that might be ignored).

I was actually able to keep tfaw and the other timings tighter and drop tRTP to 6 and tWR to 12, seems to have helped speeds this time (hitting as high at 54850 on read). Probably have to revisit many of these timings.

 

[tabascosauz]

I believe I did specifically disable TSME after reading about it somewhere, but I'll dig around and see what else can be done.

Edit: @tabascosauz
After tinkering for too long when I should be doing something else:
Apparently my motherboard doesn't automatically sync the FCLK with the mem clock, and so it has never run on the same frequency.
When I do manually sync the frequency, suddenly the system is not stable at all anymore. It's probable that the DRAM has actually never really been properly stressed because of the desynced clocks, and only now it shows.

Sadly that in effect means that I have to do the whole overclock all over again, as I cannot trust my previous results.

BTW VSOC on my motherboard only reliably is set via the CBS menu by editing the hex value it seems, the tooltip tells me that this value will be used above certain FCLKs.

Currently with normalised tRCDWR and tRC timings I can get up to 3933MHz with properly synced clocks, but 4000MHz instantly fails memtest at the first stage during address checks.

At least the latency is already much better! eventhough the read and write values are worse. Don't mind the weird cache values, pc wasn't completely idle during that test.

Thanks again for putting me on the right track.

That's interesting, I've not seen a stable memory profile fail for want of IF stability before. If you want to push 4000+ 1:1 most likely you will have to push to 1.15V VSOC, but seeing as the board BIOS is pepega I'm not sure how far you'll get. Might want to test memory in windows with TM5 or HCI instead, memtest86 is a pointless hassle.

Unfortunately I had the same board (B450 I Aorus Pro Wifi) with my 3700X originally and the BIOS is really subpar. It uses Gigabyte's last-gen BIOS (X570 and B550 boards run the successor version) and stayed on that old BIOS despite new firmware. There was a lot of buggy / missing / duplicate / broken features in that B450 BIOS. Add to that the fact that Renoir support was only (unofficially) added much later to B450. The 2.1V VDDP reading seems to be a pretty good indicator of that lol

In my HTPC I now have the successor B550I Aorus Pro AX and it handles the 4650G well (albeit a significantly higher end board). It handles 1:1 automatically and is smart enough to automatically desync when it is certain that FCLK has hit its limit, which for me the wall is between 4400 and 4466 (2200MHz). I usually don't have to use the CBS menu unless I'm running on the edge benching at 4400.

But in all honesty, 3800CL16 is already good enough for using a dGPU, decent CJR should still be able to do 3800 16-19-19 260ns. Theoretically yes, fast RAM and monolithic latency can recoup some of the loss of cache, but Renoir's tiny L3 means that you'll never catch up to Matisse no matter what tricks you pull, including a high static CPU OC or 4000+ RAM - you won't notice much of a real difference unless you use the iGPU. Cezanne is much better in that respect, dGPU gaming. I wouldn't worry too much, and just keep CJR in its sweet spot at 3600-3800 16-19-19.

 

[Taraquin]

Similar, but seemed to be slightly higher on average. Well, tfaw was 20 before (but now looking at it closer, that might be ignored).

I was actually able to keep tfaw and the other timings tighter and drop tRTP to 6 and tWR to 12, seems to have helped speeds this time (hitting as high at 54850 on read). Probably have to revisit many of these timings.

Tfaw is ALWAYS 4xtrrds so configure them at the same time. Trrds 4/tfaw 16, 5/20, 6/24 etc. Trrdl should be 2 higher than trrds for stability, the same for a bit more perf.

 

[tbob22]

Tfaw is ALWAYS 4xtrrds so configure them at the same time. Trrds 4/tfaw 16, 5/20, 6/24 etc. Trrdl should be 2 higher than trrds for stability, the same for a bit more perf.

Yeah, looks like having tfaw at 20 was actually causing a performance penalty, board was probably just reverting to auto or something. Just over 55k now on Read.

Haven't really done any memory overclocking since Westmere days.

 

[Mathragh]

That's interesting, I've not seen a stable memory profile fail for want of IF stability before.

I'm still not sure whether the IF is unstable, or actually the memory is. If you say CJR mem's sweetspot is around 3600-3800 then 4133MHz seems quite high on all accounts. Won't stop me from trying though of course .

If you want to push 4000+ 1:1 most likely you will have to push to 1.15V VSOC, but seeing as the board BIOS is pepega I'm not sure how far you'll get.

Regarding voltages, I'm a bit hesitant going much higher on the VSOC, what would be a decent maximum voltage I could or should try for those sweet 2000MHz+ FCLKs?

Might want to test memory in windows with TM5 or HCI instead, memtest86 is a pointless hassle.

I am somewhat scared of running windows repeatedly with unstable hardware, things might get corrupted :O running memtest from a bootable USB device ain't so bad and might be somewhat quicker to boot, will look into those others regardless.

Unfortunately I had the same board (B450 I Aorus Pro Wifi) with my 3700X originally and the BIOS is really subpar. It uses Gigabyte's last-gen BIOS (X570 and B550 boards run the successor version) and stayed on that old BIOS despite new firmware. There was a lot of buggy / missing / duplicate / broken features in that B450 BIOS. Add to that the fact that Renoir support was only (unofficially) added much later to B450. The 2.1V VDDP reading seems to be a pretty good indicator of that lol

yeah I figured a bit regarding the buggyness.
In my case it might actually make sense to do most or all the tweaking from the CBS menu then, as I suppose that ones the baseline for all other features and should be the least buggy.

In my HTPC I now have the successor B550I Aorus Pro AX and it handles the 4650G well (albeit a significantly higher end board). It handles 1:1 automatically and is smart enough to automatically desync when it is certain that FCLK has hit its limit, which for me the wall is between 4400 and 4466 (2200MHz). I usually don't have to use the CBS menu unless I'm running on the edge benching at 4400.

A new motherboard isn't really in the picture for me atm, I'd rather save up a bit for a Cezanne chip, if this board will also support it.

But in all honesty, 3800CL16 is already good enough for using a dGPU, decent CJR should still be able to do 3800 16-19-19 260ns. Theoretically yes, fast RAM and monolithic latency can recoup some of the loss of cache, but Renoir's tiny L3 means that you'll never catch up to Matisse no matter what tricks you pull, including a high static CPU OC or 4000+ RAM - you won't notice much of a real difference unless you use the iGPU. Cezanne is much better in that respect, dGPU gaming. I wouldn't worry too much, and just keep CJR in its sweet spot at 3600-3800 16-19-19.

Thanks! I still hope i'll be able to use both the iGPU and discrete GPU one day, time will tell. Currently have screens connected to both but windows will only let me switch between GPUs right after a new hardware install. Feels a bit like they're disallowing it on purpose.

Thanks for all the help. Its funny how different a system can feel when tinkering with the memory. Can give it a whole new character.

I'll prolly try to get back to 4133 or even higher in the future, now I've learned some of the practicalities of it. Will hopefully be posting back later with sub 60ns latencies!