• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.
  • The forums have been upgraded with support for dark mode. By default it will follow the setting on your system/browser. You may override it by scrolling to the end of the page and clicking the gears icon.

Share your AIDA 64 cache and memory benchmark here

So after tinkering with CO a bit and re reading back to some old advice I think I found my new daily settings. I think its a tasteful blend of power and performance.. its pretty tough trying to keep up with those pesky octal cores :D

To get 1900 IF with both pairs it was just like on my 3600XT.. just bump VDDG IOD a little.. I forgot that.. I bombarded myself with hardware and info and lost track of some of it.. I blame a temporarily defective short term memory, because that's how I roll..

I am using these settings for my CPU:

If its really golden

PBO 200MHz offset
-30 negative curve
+20mV offset

Set appropriate limits

EDC ~110
TDC ~75
PPT ~110

Enjoy 4.85 with peak voltage around 1.35V, in nearly every scenario, except R20 where depending on cooling, it will sit at 4.7 to 4.75 at 1.25V

But that's for another thread/topic :D

Disclaimer: May need tweaking
I've been testing for about a day and a half with no errors.. pretty spiffy :)

That L3 floats around by 20-50GB/s.. even with the cores locked.. not sure if you guys noticed that too..

Capture.PNG
 
Well, a little tweaking and we have 4x8 at 4066. Will test overnight as I need my PC at the moment, but it did do 6 runs of TM5 No problemo :D

4066 4 sticks.png
 
A few notebooks has the posdibility if tuning ram in bios, if you can do that and find out how to reset bios your scores can improve a lot :)
Well sadly my Asus G15 does not have option to tweak ram in bios.
 
If its really golden

PBO 200MHz offset
-30 negative curve
+20mV offset

Set appropriate limits

EDC ~110
TDC ~75
PPT ~110

Enjoy 4.85 with peak voltage around 1.35V, in nearly every scenario, except R20 where depending on cooling, it will sit at 4.7 to 4.75 at 1.25V

But that's for another thread/topic :D

Disclaimer: May need tweaking
I finally tried PBO+CO. -30 neg offset and +50MHz is stable in everything. I can`t use the edc, tdc and ppt values as my CPU-cooler don`t handle the temp. +100MHz+ gives me rare crashes in heavy single core scenarios. I guess it`s due to to worst active core being cycled and hitting its voltage limits at 4750MHz or more. Would a lower negative offset on worst cycled core increase SC stability? I checked CB20 and it seems it uses the best and second best core most of the time, but it somtimes hit the third and 4th core. The 2 worst cores don`t seem active at all. Best 2 cores has 143 rating, 3th and 4th have 139 and 135 in CTR 2.0. The 2 worst cores which are never used in SC it seems have 129 and 131 rating.

Well, a little tweaking and we have 4x8 at 4066. Will test overnight as I need my PC at the moment, but it did do 6 runs of TM5 No problemo :D

View attachment 198421
How did it turn out in aida?
 
The more you push the clock speed, the more EDC you need. You can open the EDC as it is purely for lighter loads - you can allow it to pull as much as it wants cos you are using -30 CO. Heavier apps will rely on the TDC/PPT to throttle down. EDC 110 is fine up to 4750, after that you experience a decrease in cache performance as cores throttle down.

Stock limits:

PPT: 88W
TDC: 60A
EDC: 90A


USE:

PPT: 90A
TDC: 65A
EDC 110A

I have cores of 125-140-137-133-140-129, I can run all cores at 5000+ using CTR with 1.38V for some really light loads ;) I need per core max frequency and then it will be much much better :D
180274734_4214447938611451_6420263172099383123_n.jpg

Aida I havent tested, I wanna pass 25 cycles on TM5. I crashed half way through last night on cycle 13, need to tweak some voltage here and there (CCD-IOD-VSOC related)
 
Last edited:
The more you push the clock speed, the more EDC you need. You can open the EDC as it is purely for lighter loads - you can allow it to pull as much as it wants cos you are using -30 CO. Heavier apps will rely on the TDC/PPT to throttle down. EDC 110 is fine up to 4750, after that you experience a decrease in cache performance as cores throttle down.

Stock limits:

PPT: 88W
TDC: 60A
EDC: 90A


USE:

PPT: 90A
TDC: 65A
EDC 110A

I have cores of 125-140-137-133-140-129, I can run all cores at 5000+ using CTR with 1.38V for some really light loads ;) I need per core max frequency and then it will be much much better :D
View attachment 198559

Aida I havent tested, I wanna pass 25 cycles on TM5. I crashed half way through last night on cycle 13, need to tweak some voltage here and there (CCD-IOD-VSOC related)
Thx, will try that. My cores are rated 143, 143, 139, 135, 131, 127 so I should be able to get the same score or slightly more than you.
 
So I tried the settings and so far stable. Voltage stabilizes at 1.19V in stresstesting and maxed at beliw 1.3V in sc. All core clocks at 4675-4700 and single at 4850. Consumption at 86 instead of 76 in CB20. I get 4615 score now vs 4520 with last setting and 1ns better in aida with pbo+200 vs pbo+50.

The more you push the clock speed, the more EDC you need. You can open the EDC as it is purely for lighter loads - you can allow it to pull as much as it wants cos you are using -30 CO. Heavier apps will rely on the TDC/PPT to throttle down. EDC 110 is fine up to 4750, after that you experience a decrease in cache performance as cores throttle down.

Stock limits:

PPT: 88W
TDC: 60A
EDC: 90A


USE:

PPT: 90A
TDC: 65A
EDC 110A

I have cores of 125-140-137-133-140-129, I can run all cores at 5000+ using CTR with 1.38V for some really light loads ;) I need per core max frequency and then it will be much much better :D
View attachment 198559

Aida I havent tested, I wanna pass 25 cycles on TM5. I crashed half way through last night on cycle 13, need to tweak some voltage here and there (CCD-IOD-VSOC related)

The more you push the clock speed, the more EDC you need. You can open the EDC as it is purely for lighter loads - you can allow it to pull as much as it wants cos you are using -30 CO. Heavier apps will rely on the TDC/PPT to throttle down. EDC 110 is fine up to 4750, after that you experience a decrease in cache performance as cores throttle down.

Stock limits:

PPT: 88W
TDC: 60A
EDC: 90A


USE:

PPT: 90A
TDC: 65A
EDC 110A

I have cores of 125-140-137-133-140-129, I can run all cores at 5000+ using CTR with 1.38V for some really light loads ;) I need per core max frequency and then it will be much much better :D
View attachment 198559

Aida I havent tested, I wanna pass 25 cycles on TM5. I crashed half way through last night on cycle 13, need to tweak some voltage here and there (CCD-IOD-VSOC related)
What no do you get error on in TM5? It seems that with agesa 1.2.0.1 or later too low voltage on soc/iod/ccd don't cause instability, but performance is reduced if volt is too low. I now run 1.11vsoc (1.1V effective on Gigabyte) and 1.02v iod, 860mv ccd and vddp. If I lower soc or iod even 10mv I get 1-2 more ns in aida and 1000mb less on read/write. Going even lower causes worse perf. Veii on OC-forum recommends vddp to 860-900 and ccd to 860-940. I would try to lower both if I were you and see what happends.

Running 110W on my setup causes thermal throttling, even though cooler is specced at 180W I get steaight to 100C with 4.8@1.32V, consumption sits at 115W then. 90W ppt gets temp to 80C, I gues 100W could work, but not much more.

overclocked bdie
If you post a zentimings-screenshot we can help you with other timings if you want? :) 3800cl14 with tweaked subs should be able to do less than 55ns :)

So after tinkering with CO a bit and re reading back to some old advice I think I found my new daily settings. I think its a tasteful blend of power and performance.. its pretty tough trying to keep up with those pesky octal cores :D

To get 1900 IF with both pairs it was just like on my 3600XT.. just bump VDDG IOD a little.. I forgot that.. I bombarded myself with hardware and info and lost track of some of it.. I blame a temporarily defective short term memory, because that's how I roll..

I am using these settings for my CPU:


I've been testing for about a day and a half with no errors.. pretty spiffy :)

That L3 floats around by 20-50GB/s.. even with the cores locked.. not sure if you guys noticed that too..

View attachment 198358
I find CO to be better than all core oc in almost every way. On Ryzen 1,2 and 3xxx I usually recommended all core oc at fixed voltage, but CO with UV+PBO on 5xxx is great. You get the idle consumption of stock, the single core perf of PBO and MC of fixed clock OC. It is possible to get slightly better consumption with all core OC during load than CO UV+PBO, but the great SC perf and idle consumption makes CO better in my optinion :) In idle\very low loads my CPU now runs 3.7GHz@840mv and uses 16-25W.
 
Last edited:
TM5 randomly stops testing on cycles 13 to 16, if it does that its an infinity fabric crash. 4 sticks is a bit more difficult than 2 sticks, needs more cldo vddp. Its just a matter of finding the right combo atm

Testing 910 - 960 - 1060 - 1160 right now
 
TM5 randomly stops testing on cycles 13 to 16, if it does that its an infinity fabric crash. 4 sticks is a bit more difficult than 2 sticks, needs more cldo vddp. Its just a matter of finding the right combo atm

Testing 910 - 960 - 1060 - 1160 right now
Okay. I think Veii recommends 40mv stepping, dunno if that can help. Iod matters most for IF, soc secondary, ccd and vddp don't affect IF at all if I remember correctly.
 
Hi people im new to the ram oc stuff please some1 would check my timing is there any suggestion bout it


new.png


cachemem.png
 
I gave up for now on the hard push with 4 dimms. I have another set of Royals in my cart to match up what I have now, but not sure if that will solve my problem. You know what I bet would solve it? Something as simple as switching positions. Put the B+W in 2+4 and the Royals in 0+3. That one little number difference in Thaiphoon is probably what bunging me up. I think some of the timings we cant see are being set to my Royals which is the higher bin in the primary slots.. The royals can be dumbed down, but the B+Ws are not the same grade..

I just don't want to blow over 200 bucks on ram lol.. But the little bump in performance that you do get from 4x8 is hard to walk away from..

CO is nice, but for absolute best performance I prefer all core. CO hasn't kept up in that regard according to my benchmarks.. but its close.

Edit:

Wifey is buying a board for the kids to use with my 3600XT, so I will be losing my black and whites anyways.. as intended..
 
4066C16 TM5.jpg


:D


The IOD fixed it. Thanks for that info @Taraquin
 
4333 is stable.......but not iGPU-stable unfortunately.

renoir 4333 17-18-18.png
4333 17-18-18.png
 

Attachments

  • 4333 membench 99s.png
    4333 membench 99s.png
    73.5 KB · Views: 116
Allow me assume the appropriate position

:respect:

Currently:

Capture2.PNG
 
Zenshot? :)


When I build a 2400G last year I noticed the same. 3533 fast preset rev E was stable in non GPU-loads, but anything above 3333 was iGPU unstable.
I have to shamefully say it’s just a slightly tweaked DOCP profile :fear:
 
Zenshot? :)


When I build a 2400G last year I noticed the same. 3533 fast preset rev E was stable in non GPU-loads, but anything above 3333 was iGPU unstable.

If I had opted for an external AC-DC brick I'd be able to run my 5450 or GT610, then the limit would probably be 4400 or 4466. Unfortunately I mounted the HDPlex brick with extra-strength tape :laugh: if i just had a riser I could still hook up a GPU ghetto-style for benching, just not gaming bc HD5450 sucks ass

As soon as the iGPU comes online, VSOC drops like a brick and there's no stopping it on the VSOC VRMs that we have. I get why it is, because that's pretty much exactly how vdroop is supposed to work, but it doesn't make it any less frustrating. If I had a way to reduce droop by ~40-50% then I'm pretty sure 4266 and 4333 would be doable (4400 still impossible at 1.2V).

In the meantime I'm still stuck at 4200, can only work on tighter timings instead which doesn't help GPU performance even 0.1%

4200 16-16-16 stable.png
4200 new bdie stable timings.png
 
If I had opted for an external AC-DC brick I'd be able to run my 5450 or GT610, then the limit would probably be 4400 or 4466. Unfortunately I mounted the HDPlex brick with extra-strength tape :laugh: if i just had a riser I could still hook up a GPU ghetto-style for benching, just not gaming bc HD5450 sucks ass

As soon as the iGPU comes online, VSOC drops like a brick and there's no stopping it on the VSOC VRMs that we have. I get why it is, because that's pretty much exactly how vdroop is supposed to work, but it doesn't make it any less frustrating. If I had a way to reduce droop by ~40-50% then I'm pretty sure 4266 and 4333 would be doable (4400 still impossible at 1.2V).

In the meantime I'm still stuck at 4200, can only work on tighter timings instead which doesn't help GPU performance even 0.1%

View attachment 198792 View attachment 198795
Very good R, W and C-speed. tRAS 32 and tRC 48 and tRFC 288 will probably work :) tRTP trives at half of tWR. Try tRTP 6, 8 could also work. Sometimes tWR 10 and tRTP 5 works, that gave me a slight boost without requiring more voltage. tRRDL 6 and tWTRL 10 or 8 might work aswell :)
 
Very good R, W and C-speed. tRAS 32 and tRC 48 and tRFC 288 will probably work :) tRTP trives at half of tWR. Try tRTP 6, 8 could also work. Sometimes tWR 10 and tRTP 5 works, that gave me a slight boost without requiring more voltage. tRRDL 6 and tWTRL 10 or 8 might work aswell :)

Already done :laugh: these A2s are no joke

It's really interesting to see how tRFC behaves in the presence of an iGPU. I can pass endless TM5 runs @ 133ns without moving VDIMM even an inch, but the GPU really doesn't like it. 130ns no POST. 133ns tests stable but crashes instantly opening a game. 135ns test stable, runs for a while until throwing a UNEXPECTED_KERNEL_MODE_TRAP. 138ns (290) works just fine though.

It was the same with the A0s, tRFC went all the way down to 140ns stable but only about 143-145ns was actually iGPU stable.

On the A0s I couldn't do tRRDL 4, tWTRL 8 or tWR 10 without needing more VDIMM, not yet sure if it's the same here. I'll probably play with tRTP next.

@FilipM that reminds me, the old GDM vs 2T debate seems to be rather interesting. 2T is *sometimes* faster in membench by half a second or so. But GDM consistently wins in AIDA read/write by about 500MB/s and copy by about 200MB/s. GDM might also be leading to better GPU performance, but I'll need to bench that.

4200 16-16-16 133ns not gpu stable.png
 
Already done :laugh: these A2s are no joke

It's really interesting to see how tRFC behaves in the presence of an iGPU. I can pass endless TM5 runs @ 133ns without moving VDIMM even an inch, but the GPU really doesn't like it. 130ns no POST. 133ns tests stable but crashes instantly opening a game. 135ns test stable, runs for a while until throwing a UNEXPECTED_KERNEL_MODE_TRAP. 138ns (290) works just fine though.

It was the same with the A0s, tRFC went all the way down to 140ns stable but only about 143-145ns was actually iGPU stable.

On the A0s I couldn't do tRRDL 4, tWTRL 8 or tWR 10 without needing more VDIMM, not yet sure if it's the same here. I'll probably play with tRTP next.

@FilipM that reminds me, the old GDM vs 2T debate seems to be rather interesting. 2T is *sometimes* faster in membench by half a second or so. But GDM consistently wins in AIDA read/write by about 500MB/s and copy by about 200MB/s. GDM might also be leading to better GPU performance, but I'll need to bench that.

View attachment 198959
Your bin is so much better than mine, atleast 133MHz at everything the same. I could run the same as you on 4066 at 1.47V, but not further.
 
Back
Top