to the guys asking what would be the purpose of this tools, outside of overclocking.
Gigabyte x570 gaming x (f11))+ viper steel 3600 cl17 XMP = bluecreen symphony.
Used the calculator, created a fast profile + changed IF speed + changed voltage = no more bluescreens - and the timings are better that the xmp ones (i am not good at ram oc, i just followed the labels for ram timings and then entered amd overclocking screen to change IF speed).
Tested the new release, small improvements to the fast settings + now i can keep Geardown mode on > applied new timings = works like a charm 🙂
@JonasBeckman so , for the price differecence from 3600 to 4200 I gain 20 fps ! bargain 😀 ... p.s. that was just a test you showed me ! until ALL test ( gamnig or non gaming ) have that increase ... I still stand my point .
Want to report some bugs (I really valuate your hard work but this versions 1.7.1-2 are very buggy for me). 3800X Hero VI X370 Agesa 1.0.0.4B bios 7704, Corsair 2x8GB 3600MHz CL 18-22-22-42 Micron E (TPB says its A0 revision) Recommended A0 profile for Micron E-die missing, import XMP not working(doesnt read) --> manual not working, Fast presets show less Voltage req across diferent freq settings? Sometimes even compare timings doesnt read values, shows clear positions. With these bugs I`m not sure I cant trust values I see. I used 1.7.0 without problems. (one more but not urgent - Maximizing and back makes window staying wide (width stays at monitor resolution size - offscreen position)) Sry for my eng, just trying to help.
@JonasBeckman so , for the price differecence from 3600 to 4200 I gain 20 fps ! bargain 😀 ... p.s. that was just a test you showed me ! until ALL test ( gamnig or non gaming ) have that increase ... I still stand my point .
DDR4 prices going down to more reasonable levels wouldn't be a bad thing, there's a good number of titles that do benefit but of course from a price / performance perspective there's some very diminishing returns when each stick potentially costs a few hundred dollars or comparable each.
Looking forward to what the DDR5 kits can do and the latest improvements to the CPU and motherboard architecture both from AMD and Intel, maybe with some tweaking the upcoming 4000 series won't be as dependent although it is also possible it will show even bigger benefits by removing some of the current limits and bottlenecks and improvements to cache and such.
Fallout 4 even with DDR3 was what really showed me how some games and I presume also other software including the OS itself could scale really nicely with higher speed kits and since then just about any memory intensive game especially open world or those relying extensively on data streaming benefits but once you pass around the 3200 range for DDR4 kits the prices really ramp up too so these high-end modules become incredibly costly for a smaller benefit from that point.
(It's nice to have but within a reasonable pricing range and compared to the rest of the cost of the hardware for the system.)
Exactly , not to mention that all I could find in my country stores when I built this PC was 3600 CL18, and to buy a good 3600Mhz CL16 it was to expansive with an estimation date of arrival of 3 - 4 weeks . ... After you pass 3600 Mhz ... RAM prices just skyrockets
I just love this tool. Such a time saver. I have my Ryzen 1800x running with 3200Mghz cas 14 with some pretty tight sub timings due to this tool. My CPU just can't handle any higher frequency.
Are you high? It was always like that. There is no error if you pay attention to which exactly parameter you edit.
... and since forever those values are not what is written in BIOS from the upside down, but all other values seem "in the right place" so you can read / set them by going from the upside downwards.
So his first "issue" applies to me, too.
@mgilbert :
The issue with your tRFC is the calculator uses what "every chip may be capable of", for example 160ns for Samsung b-die. This equals to tRFC 256 at 3200MHz (my own "safe" profile).
You calculate the specs written in your RAM by a given formula, starting from 350ns instead of 160 (and different speed).
So both are correct, but the formula is different.
Rule of thumb = higher values are bad for performance but good for stability.
It used to recommend a max of 1.48V, and I'm pretty sure with the current settings my RAM needs 1.47V to be 100% stable, but now its max recommended says 1.40V which is a big LOL... and for tWRRD it used to recommend 3, now it's 1. Pretty sure my PC will implode into blue screen oblivion if I try 1 on that timing but I'll give it a shot later.
[spoiler]
https://i.imgur.com/a8eeTCb.png [/spoiler]
You would be surprised that some of the settings need a lot more voltage than others.
The trick is finding out which one.
For example, I am now at 1.385, down from 1.43 !!! Just by simply adding 1 to rRDWR ( 9 instead of recommended: 8 ), keeping every other setting the same.
Already 30 hours of using the computer, no issues, no crashes, no errors, nothing, works perfectly at this new lower voltage.
---
This program's recommendations are just that, not absolute values.
Some testing and experimentation on your side is needed as well.
Who knows, with just tiny changes you might drop from your 1.48 to just 1.4 and remain stable with no loss in performance.
You would be surprised that some of the settings need a lot more voltage than others.
The trick is finding out which one.
For example, I am now at 1.385, down from 1.43 !!! Just by simply adding 1 to rRDWR ( 9 instead of recommended: 8 ), keeping every other setting the same.
Already 30 hours of using the computer, no issues, no crashes, no errors, nothing, works perfectly at this new lower voltage.
---
This program's recommendations are just that, not absolute values.
Some testing and experimentation on your side is needed as well.
Who knows, with just tiny changes you might drop from your 1.48 to just 1.4 and remain stable with no loss in performance.
Thanks for the suggestion. But I'm not afraid of high voltage on RAM, there is a reason why RAM has lifetime warranty. It's pretty hard to kill, and 1.47V or even higher isn't dangerous for DDR4 based on everything I've read. Also I'm pushing pretty much all the timings to extremes on this RAM, it would take an insane amount of testing for me to get to that lower voltage. Not to mention many nights of Memtest 86+ running to confirm stability. So screw it, I'll take that extra drop of performance with brutal voltage. Brutal voltage for brutal timings.
I didn't even know precisely what this RAM was when I bought it, there was squat for info online (still squat), but I was definitely going to buy it because it was hands down the best I could get for anything in the same price galaxy (was given a special discount by ye olde Amazon). I only relatively recently found out they're A2 Samsung B-dies, but considering the MSRP on these things was enough to make Vegeta break his scouter, I'm not too surprised.
Early DDR3 used pretty hefty values and some of the higher clocked kits even up to 2.0v or higher whereas I believe it's now 1.25v stock on DDR4 and 1.35v on most XMP profiles but upwards of 1.45v or even 1.5v is usable though I'd imagine the higher voltages need some more testing and checking temps and all that plus the different memory modules and what type it is could need some adjusting for.
OCZ 1800 (8-8-8-24 I think.) at 2.0v up to 2.1v maximum and even seven years later it worked just fine though that too is going to be pretty variable. 😀
EDIT: Well not like I'm an expert, early on I only considered the increased power consumption and heat buildup not what additional voltage could do to the circuitry over time and faster degradation.
Smaller nodes and all that too which are more sensitive compared to the big chunky things of the 1990's when I got started with this ha ha.
EDIT: Suppose that also means recent DDR4 modules might differ from the initial batches whatever steps memory have gone down through over the years. Should have done some reading up on that come to think of it.
Tests being done on 5nm and even lower now although that might not be on the market for some time yet, will be interesting to hear what DDR5 will be using and I suppose higher speeds up to possibly 8000 but even laxer timing values.
Be careful with DRAM Calculator for Ryzen. I've had the settings it recommends work, but I've also had those settings corrupt a Windows installation beyond all recovery, requiring a clean reinstallation. Set a restore point before you change anything - but still be prepared to do a clean install if things go really wrong. And no, I'm not a noob. I've got over forty years' experience with computers, having once owned my own computer business in which I built and repaired hundreds of computers over the course of a decade.
After seeing Tech Jesus's video on RAM speed affecting real world game performance on Ryzen I tried to push my RAM clocks, but alas, the tight timings Ryzen DRAM calculator lists cannot be pulled off at 3800MHz on a kit that was rated 3600MHz. I did the tight timings for sub timings and the safe timings for most the primary timings. Raising the RAM/controller/fabric all from 1800 to 1900MHz even with the far worse timings gave me a minimum of 5% performance improvement in Nier Automata, a game that is programmed like ass when it comes to CPU usage (DX 11 and Japanese) so I expect a larger than 5% improvement in properly coded DX 12 games that use more cores.
My current settings with RAM at 1.45V showed as stable after much Mem 86+ testing, but then my PC failed to POST after I rebooted. Not sure if that was a strange fluke but I just didn't have the patience and cranked it back to 1.47V and left it there for now. I had to learn about VDDG CCD/IOD and CLDO VDDP and all that because my Asus TUF X570-Plus Wi-Fi is absolutely retarded with stock settings.
It isn't just an Asus thing either I think, MSI was also nuts with stock voltages, and the CPU voltage at "auto" is also insanity.
As soon as the Fabric clock was touched the board was pumping like 1.15V for the VDDGs which is apparently death, and the VDDP voltage also went high (I forgot what). Meanwhile the SoC voltage was 1.107V in reality because I had that manually set. Apparently the SoC voltage should always be 50mV above the VDDG otherwise ???
Anyway my preliminary manual voltage settings, all of which can hopefully go lower:
- RAM 1.47V BIOS doesn't show in Windows what it actually is.
- CCD & IOD VDDG 1.02V BIOS, 1.0182V Windows.
- VDDP 0.91V, 0.9091V Windows.
- SoC 1.13V, 1.107V Windows (auto LLC, I don't want to screw with it, since this seems safe).
Screenshot: [spoiler]
[/spoiler]
I didn't even know when gear-down mode is on only even CL is possible. I learned quite a bit having to looked up individual settings so Asus auto settings don't murder my system. This is the first time where extremely fine tuning RAM was worthwhile. I do not like how convoluted this all is.
I suspect in a few years a lot of users who just used stock/auto are going to have degraded systems, particularly memory controllers.
After seeing Tech Jesus's video on RAM speed affecting real world game performance on Ryzen I tried to push my RAM clocks, but alas, the tight timings Ryzen DRAM calculator lists cannot be pulled off at 3800MHz on a kit that was rated 3600MHz. I did the tight timings for sub timings and the safe timings for most the primary timings. Raising the RAM/controller/fabric all from 1800 to 1900MHz even with the far worse timings gave me a minimum of 5% performance improvement in Nier Automata, a game that is programmed like ass when it comes to CPU usage (DX 11 and Japanese) so I expect a larger than 5% improvement in properly coded DX 12 games that use more cores.
My current settings with RAM at 1.45V showed as stable after much Mem 86+ testing, but then my PC failed to POST after I rebooted. Not sure if that was a strange fluke but I just didn't have the patience and cranked it back to 1.47V and left it there for now. I had to learn about VDDG CCD/IOD and CLDO VDDP and all that because my Asus TUF X570-Plus Wi-Fi is absolutely retarded with stock settings.
It isn't just an Asus thing either I think, MSI was also nuts with stock voltages, and the CPU voltage at "auto" is also insanity.
As soon as the Fabric clock was touched the board was pumping like 1.15V for the VDDGs which is apparently death, and the VDDP voltage also went high (I forgot what). Meanwhile the SoC voltage was 1.107V in reality because I had that manually set. Apparently the SoC voltage should always be 50mV above the VDDG otherwise ???
Anyway my preliminary manual voltage settings, all of which can hopefully go lower:
- RAM 1.47V BIOS doesn't show in Windows what it actually is.
- CCD & IOD VDDG 1.02V BIOS, 1.0182V Windows.
- VDDP 0.91V, 0.9091V Windows.
- SoC 1.13V, 1.107V Windows (auto LLC, I don't want to screw with it, since this seems safe).
Screenshot: [spoiler]
[/spoiler]
I didn't even know when gear-down mode is on only even CL is possible. I learned quite a bit having to looked up individual settings so Asus auto settings don't murder my system. This is the first time where extremely fine tuning RAM was worthwhile. I do not like how convoluted this all is.
I suspect in a few years a lot of users who just used stock/auto are going to have degraded systems, particularly memory controllers.
Can you run the MEMbench test at
default, Thank you.
Can you run the MEMbench test at
default, Thank you.
Here you go:
I haven't used MemBench before. Are my results normal-ish and any suggestions? My inner-CCX latency seems a bit high for a 3900X compared to results I've seen from others. And memory latency seems crap but I suppose with those loose main timings I had to use I shouldn't be surprised at that part.
I was wondering some other things as well if anyone knows:
1 - Do the DIMM slots used matter? Does this affect how it's read by the memory controller in any meaningful way? I'm using slots 2 & 4 rather than 1 & 3.
2 - Does using all 4 slots pressure the mem controller enough to lower performance/stability like what happened with some Intel chips?
3 - Is it normal for the RAM model # listed to not match what's physically written on it/what it's branded/sold as? What I ordered, and what is on the stickers of my DIMMs, doesn't match that model #. Do manufacturers use separate model #s? This was super annoying because I was looking up what my system was showing me to see compatibility and try to find info on it (before I got Thaiphoon reader to just read the eeprom) and I was coming up with nothing.
4 - Is it normal for nothing to read RAM voltage/temps in Windows? Edit: Apparently no. Great. It'd be hilarious if these high end DIMMs, with bullshit like RGB syncing through infrared, don't send sensor info in Windows and have no temp sensor. And by hilarious I mean retarded.
[/spoiler] I didn't even know when gear-down mode is on only even CL is possible. I learned quite a bit having to looked up individual settings so Asus auto settings don't murder my system. This is the first time where extremely fine tuning RAM was worthwhile. I do not like how convoluted this all is. I suspect in a few years a lot of users who just used stock/auto are going to have degraded systems, particularly memory controllers.
I haven't used MemBench before. Are my results normal-ish and any suggestions? My inner-CCX latency seems a bit high for a 3900X compared to results I've seen from others. And memory latency seems crap but I suppose with those loose main timings I had to use I shouldn't be surprised at that part. I was wondering some other things as well if anyone knows: 1 - Do the DIMM slots used matter? Does this affect how it's read by the memory controller in any meaningful way? I'm using slots 2 & 4 rather than 1 & 3. 2 - Does using all 4 slots pressure the mem controller enough to lower performance/stability like what happened with some Intel chips? 3 - Is it normal for the RAM model # listed to not match what's physically written on it/what it's branded/sold as? What I ordered, and what is on the stickers of my DIMMs, doesn't match that model #. Do manufacturers use separate model #s? This was super annoying because I was looking up what my system was showing me to see compatibility and try to find info on it (before I got Thaiphoon reader to just read the eeprom) and I was coming up with nothing. 4 - Is it normal for nothing to read RAM voltage/temps in Windows? Edit: Apparently no. Great. It'd be hilarious if these high end DIMMs, with bullshit like RGB syncing through infrared, don't send sensor info in Windows and have no temp sensor. And by hilarious I mean retarded.