8 PACK MEMORY RANGE GROWING: SAY HELLO TO 8 PACK RIPPED EDITION & 32GB KITS!!!

[MrPils]

A little research has shown that Asus have some real problems with their Ryzen 5000 bios releases on all the crosshair boards at the moment, it seems like that's what I was fighting against yesterday. There's some major complaints on Asus own forums, HWBot forums and overclock.net forums from lots of users. It definitely seems to be a memory training issue and the users complaining of issues have confirmed that their settings that were stable on the 2206 and earlier bioses now will not even post. Most have had to go down to 3600mhz or even 3200mhz to get them stable, voltage and timing changes are no help.
I find it interesting that I've got no problems at all with the recently released G.skill F4-4000C16D-32GVK kit, yet no end of issues with the older (but still October manufactured, so recently built) G.Skill F4-4000C15D-16GVK kit and the Patriot 4400 C19 kit though. Both those problematic kits are on similar A2 PCBs, though the Patriot kit has the newer capacitor layout. The fact that I can get them stable given enough attempts at retraining by changing timings gives me hope - its a shame I cant tell if the training has worked properly until I get into windows and memtest though.
Even with an older 500 series chipset board and older memory kits that previously worked with Ryzen 3000 it does appear there's still some early adopter issues with Ryzen 5000. There was the odd mention of Strix B550 boards having issues as well in the Crosshair threads I was reading, so it doesn't just seem to be the top end of the Asus range.

 

[Fyter]

B-Die Rank Comparison post #1 - G.skill F4-4000C16D-32GVK Results

Results are here for the 2 x 16Gb Dual Ranked setup. This is a retail dual channel kit imported from the US, the label says it was manufactured November 2020 so its brand new B-Die.

WOW! Thank you for putting a lot of time and effort into this. This has been a hot subject in the past month or so following zen 3 launch. I think the difference if present is shown in CPU intensive games with a good graphics card at low resolutions and high refresh rates. I am in the market to get a 5900x/5950x (which ever comes in stock first). Right now its either a 2 x 8 GB or 2 x 16 GB b die for me since I have a daisy chain topology motherboard (b550 gaming edge).

It would be great if you could do side by side comparison in warzone for both of these tuned memory kits. That games is hugely popular and is known to be very cpu intensive at 1080p and below when paired with a good graphics card. It would be interesting to see how much of an impact it has on 0.1% and 1% low FPS. Runnig tests similar to this @720p and 1080p will probably be the best idea. I know tuning alone can yield 10% better perforamnce on warzone but i havent really seen anyone test the single vs dual rank in this game.
https://www.youtube.com/watch?v=FlzJLPxbaVY&ab_channel=TechImperium

 

[MrPils]

WOW! Thank you for putting a lot of time and effort into this. This has been a hot subject in the past month or so following zen 3 launch. I think the difference if present is shown in CPU intensive games with a good graphics card at low resolutions and high refresh rates. I am in the market to get a 5900x/5950x (which ever comes in stock first). Right now its either a 2 x 8 GB or 2 x 16 GB b die for me since I have a daisy chain topology motherboard (b550 gaming edge).

It would be great if you could do side by side comparison in warzone for both of these tuned memory kits. That games is hugely popular and is known to be very cpu intensive at 1080p and below when paired with a good graphics card. It would be interesting to see how much of an impact it has on 0.1% and 1% low FPS. Runnig tests similar to this @720p and 1080p will probably be the best idea. I know tuning alone can yield 10% better perforamnce on warzone but i havent really seen anyone test the single vs dual rank in this game.
https://www.youtube.com/watch?v=FlzJLPxbaVY&ab_channel=TechImperium

I'm having issues with the Asus 5000 series bioses at the moment - they're not stable with any of my 2 x 8gb kits unless they're clocked and timed well below their capabilities. Brute force retries will eventually allow the board to boot into windows with stable memory training, but after powering off or changing any timings those stable settings are lost. I managed to get a lot of the comparison work done when i did get a stable boot into windows and the difference between the two setups was largest in the Timespy 1080p cpu test - the maxed out 2 x 16gb dual rank setup got 30.00fps and the maxed out 2 x 8gb single rank setup got 30.97fps. This was a difference of around 3% in a test heavily favoured to show memory bandwidth differences. In an actual game where a GPU is more involved the difference will be lower, you are looking at 1 or 2 fps at most.

This means that if you don't need more than 16gb for maximum performance you shouldn't use a dual rank configuration (2 x DR or 4 x SR) as 2 x 8gb single rank is the fastest configuration as long as you are able to manually max out the capabilities of the memory sticks. If you are using Auto and XMP timings then it becomes down to the programming of the memory sticks / motherboard bios optimisation and is nothing to do with the configuration of ranks.

I don't think Warzone has a built in benchmark does it? Its certainly not possible to accurately measure differences this small without that feature. I have Valhalla and Tomb Raider that both have this function, but both use a different engine to COD so they aren't comparable. I've got Cold War but haven't yet installed it - not sure if this has a benchmark feature built in?

 

[FireTwo]

Woomack did a lot of testing. 4200 is about the max you can expect unless you have a really great board. The Msi z490i Unify is around this for reference.

4400 C17 requires 1.6v if you have the board to clock it. I guess evga ones?

4200 16-17-17-38 1.50V is around the max you will get most of the time with these.

https://www.overclockers.com/forums...on-2x16GB-DDR4-3600-CL16-TDPPD416G3600HC16GBK

my viper4400 with msi unify used to run 4400 all day long with two sticks in cl17/17/17/34 with 1.45v
in the apex 4500 was stable.

just got two sticks of this ram now 2x16gb booting fine at 4400mhz cl17 but needs like 1.5v to be stable in HKEPC RunMemTest

 

[MrPils]

my viper4400 with msi unify used to run 4400 all day long with two sticks in cl17/17/17/34 with 1.45v
in the apex 4500 was stable.

just got two sticks of this ram now 2x16gb booting fine at 4400mhz cl17 but needs like 1.5v to be stable in HKEPC RunMemTest

You can't compare clocks and timings achieved on 2 x 8gb sticks to 2 x 16gb sticks, 2 x 16gb kits need to be binned far tighter. Even the Patriot 4400's have a massive range of abilities, some only post 14-14-14-28-1T 1.35v to 3400mhz whereas others go right up to 3666mhz (or even beyond, but out of 22 kits I didn't get any that posted higher than that). My top two sticks do 4466 C16 1.6v for cpuz validations, the worst sticks wont even boot to windows at 4400 C18 1.6v.

Quality will vary massively with any 3600 C16 kit, its a pretty low bin to be aiming for 4000+ c16/c17 at. Just like you can get godlike 3200mhz C14 8gb sticks, you can get oc trash too. Sounds like your kit is capable of being binned far higher than the speed grade it was put in - you got lucky . I guarantee they wont all do that.

 

[Fyter]

I'm having issues with the Asus 5000 series bioses at the moment - they're not stable with any of my 2 x 8gb kits unless they're clocked and timed well below their capabilities. Brute force retries will eventually allow the board to boot into windows with stable memory training, but after powering off or changing any timings those stable settings are lost. I managed to get a lot of the comparison work done when i did get a stable boot into windows and the difference between the two setups was largest in the Timespy 1080p cpu test - the maxed out 2 x 16gb dual rank setup got 30.00fps and the maxed out 2 x 8gb single rank setup got 30.97fps. This was a difference of around 3% in a test heavily favoured to show memory bandwidth differences. In an actual game where a GPU is more involved the difference will be lower, you are looking at 1 or 2 fps at most.

This means that if you don't need more than 16gb for maximum performance you shouldn't use a dual rank configuration (2 x DR or 4 x SR) as 2 x 8gb single rank is the fastest configuration as long as you are able to manually max out the capabilities of the memory sticks. If you are using Auto and XMP timings then it becomes down to the programming of the memory sticks / motherboard bios optimisation and is nothing to do with the configuration of ranks.

I don't think Warzone has a built in benchmark does it? Its certainly not possible to accurately measure differences this small without that feature. I have Valhalla and Tomb Raider that both have this function, but both use a different engine to COD so they aren't comparable. I've got Cold War but haven't yet installed it - not sure if this has a benchmark feature built in?

What you say make senses, overclocking the max of 2 single rank b die is likely to yield better overclock results compared to 2 dual ranks. Whether its easier on the memory controller or less chance of having bad bin chips on the stick, that i am not sure of. My b550 board will be turned into a streaming pc when zen 4/meteor lake hits, 5900x will be sold and i will probably get a 3700x/3900x for dedicated streaming pc, thus getting 16gb might be a better idea since dedicated streaming PCs wont be benefitng much from too much ram. 16GB should be sufficient for the next 1-2 years i supposed.

Nevertheless, woud love to see the tuned single rank vs dual rank on warzone. Unfortunately, it does not have a dedicated benchmark mode but you can do a similar run to the one on the video. If there is a difference of 5-15% then we know one is better than the other if within 1-2% that could be margin of error.

https://www.youtube.com/watch?v=FlzJLPxbaVY&ab_channel=TechImperium
or this
https://www.youtube.com/watch?v=5XF-rirft8U&ab_channel=BENCHMARKSFORGAMERS

I am not sure whether i should spend on dual ranks and buy a 5900x or get 16gb and put the money towards 5950x. The second video shows that the game benefits from the 16 cores, however clock speeds are consistent with one another.

 

[FireTwo]

this stuff responds well to volts 1.5 needed tbh.

 

[MrPils]

What you say make senses, overclocking the max of 2 single rank b die is likely to yield better overclock results compared to 2 dual ranks. Whether its easier on the memory controller or less chance of having bad bin chips on the stick, that i am not sure of. My b550 board will be turned into a streaming pc when zen 4/meteor lake hits, 5900x will be sold and i will probably get a 3700x/3900x for dedicated streaming pc, thus getting 16gb might be a better idea since dedicated streaming PCs wont be benefitng much from too much ram. 16GB should be sufficient for the next 1-2 years i supposed.

Nevertheless, woud love to see the tuned single rank vs dual rank on warzone. Unfortunately, it does not have a dedicated benchmark mode but you can do a similar run to the one on the video. If there is a difference of 5-15% then we know one is better than the other if within 1-2% that could be margin of error.

https://www.youtube.com/watch?v=FlzJLPxbaVY&ab_channel=TechImperium
or this
https://www.youtube.com/watch?v=5XF-rirft8U&ab_channel=BENCHMARKSFORGAMERS

I am not sure whether i should spend on dual ranks and buy a 5900x or get 16gb and put the money towards 5950x. The second video shows that the game benefits from the 16 cores, however clock speeds are consistent with one another.

If you don't need 32gb for anything and are only considering performance then 16gb ram and the 5950x is what I would go with if those are your options. Turn off multithreading and run with 16 cores and 16 threads, that will give your physical cores more clock headroom in terms of power and temperature and you'll be getting the highest possible silicon quality that AMD currently make. If you do ever need to go to 32gb in the future you can get another pair of 8gb sticks - under 4000mhz is easy with 4 decent sticks on a decent board so you wont be held back by running with all 4 slots populated. 4 slots in use affects how high mhz you can reach, it has negligible difference on the timings themselves unless the board you are using has some kind of issue with its second slots. Something like a Crosshair, Unify or Master would be fine for both powering a 16 core and expandability to using all 4 memory slots at up to 4000mhz in the future. Honestly even a Tomahawk should have no issues with up to 4000mhz using 4 single sided sticks - the timings will be down to what the sticks themselves can manage. Dual ranks you usually have to go to Gear Down Enabled and with 4 sticks you'll have to loosen up tRDRDDD and tWRWRDD timings from 1 to somewhere between 4 and 6. tRRD_L and tWTR_L may need to be loosened out to 6 and 12 respectively, but that is down to the trace quality to the second pair of slots on the motherboard and nothing to do with running 4 sticks / 4 ranks.

 

[Guest2]

I have a query, is it better to try and reduce the time result for easy / default mem bench in DRAM calculator, or aim for lower latency.

DRAM time result for easy is 107 seconds, latency 54.X ns. I have a feeling I could try 3800Mhz with slightly slacker timings but this will increase latency. Latency vs Read/write speed, which wins in terms of general performance?

 

[MrPils]

I have a query, is it better to try and reduce the time result for easy / default mem bench in DRAM calculator, or aim for lower latency.

DRAM time result for easy is 107 seconds, latency 54.X ns. I have a feeling I could try 3800Mhz with slightly slacker timings but this will increase latency. Latency vs Read/write speed, which wins in terms of general performance?

You can check that using Timespy cpu only with locked cpu speed, should give you a pretty good idea. If you can get the memtest easy test significantly faster then that should translate to performance in most games, if its close though (1 or 2 seconds only) then latency usually wins if you're looking for more fps.

 

[Guest2]

You can check that using Timespy cpu only with locked cpu speed, should give you a pretty good idea. If you can get the memtest easy test significantly faster then that should translate to performance in most games, if its close though (1 or 2 seconds only) then latency usually wins if you're looking for more fps.

I’ll give it a try, pbo disabled with current setting then attempted 3800 / 4000 with slacker timings.
Time consuming stuff. Out of interest, do you work in RAM sales / R&D for a company? Your tests must take a lot of time. Things like testing 22 RAM kits with all different timings must take weeks

 

[MrPils]

I’ll give it a try, pbo disabled with current setting then attempted 3800 / 4000 with slacker timings.
Time consuming stuff. Out of interest, do you work in RAM sales / R&D for a company? Your tests must take a lot of time. Things like testing 22 RAM kits with all different timings must take weeks

I worked with Foxconn engineers on tuning their motherboards for a couple of years, the Blood Rage and Katana took some of my input. Unfortunately management changes happened at Foxconn and they abandoned the enthusiast sector just when they were getting some good stuff out the door. Family has happened since then, now I'm a boring network admin and this is just hobby level stuff

It's pretty quick binning B-Die for quality, just set the timings manually - Primaries to 14-14-14-14-28-42-1T and TRRD/TFAW/TWTR to 4-4-16-4-8 (the rest of the timings don't matter) @ 1.35v and see how high it goes before you get a no post. 2 minutes per stick at most, less for the bad ones lol. That was a rare opportunity I had in that we needed 20 kits of memory for upgrades at work and the patriot 4400 C19 was cheaper on offer than we were paying for the usual Kingston trash we get. I bought two kits myself and quick tested the lot, kept the best 4 sticks . Think it took me around an hour in total to bin through 44 sticks.

 

[Guest2]

Nice. I would take me a week!
Certainly an interesting job but need to be patient which sometimes i'm not and get blood rage

Results...

I disabled PBO and left my settings as current (CAS14, 1.43v, DDR3600, FCLK 1800 - 108 second easy DRAM membench, 54.1 ns latency) 12950 timespy cpu score
I then tried PBO disabled, CAS16, sub timings auto, 1.45v, DDR4000, FCLK 1800 - errors in easy DRAM membench
Then PBO disabled, CAS16, sub timings auto, 1.45v, DDR3866, FCLK 1800 - errors in easy DRAM membench
Then PBO disabled, CAS16, sub timings auto, 1.45v, DDR3800, FCLK 1900 - didnt post. rebooted and ran into problems. Powered off and reset back to my working CAS14, 1.43v, DDR3600

I'm missing something obvious or have bad sticks. I doubt the latter.

Certainly dont have the same patience for this as you guys. Think i'll be happy with my new system and leave it as is, at least until a new non beta bios is available. The Meg unify bios before the beta im using didnt even work with default settings and gave me random freezes so it could be a bios thing

 

[retrop]

Just received two sets of the 3600 cl14 stuff.

I've been playing around this morning and had no problems with the xmp settings when running just 16gb after adjusting a few voltages but I'm yet to get 32gb dialed in.

The best I can get booted and in to Windows to do some benching so far is 15-15-15 but this still blows the 3000 16-18-18 stuff I'm coming from out of the water.

z390 aurous ultra, i9 9900k 4.7 cache, 4.8 all core oc (daily runner oc as my chip isn't the best of the bunch).

Some quick and dirty numbers so far from aida64 benchmarks.

gb - freq - timings - latency (avg) - read - write - copy
32g - 3000 - 14-15-15-35-630-2t - 49.8 - 42218 - 45578 - 42319
16g - 3466 - 14-15-15-35-630-2t - 43.25 - 49988 - 50533 - 45163
16g - 3600 - 14-15-15-35-630-2t - 41.775 - 52644 - 52565 - 48052
32g - 3600 - 15-16-16-35-630-2t - 45.15 - 38600 - 37399 - 41476
32g - 3600 - 15-15-15-32-630-2t - 44.65 - 51719 - 54825 - 51130

 

[Guest2]

Just received two sets of the 3600 cl14 stuff.

I've been playing around this morning and had no problems with the xmp settings when running just 16gb after adjusting a few voltages but I'm yet to get 32gb dialed in.

The best I can get booted and in to Windows to do some benching so far is 15-15-15 but this still blows the 3000 16-18-18 stuff I'm coming from out of the water.

z390 aurous ultra, i9 9900k 4.7 cache, 4.8 all core oc (daily runner oc as my chip isn't the best of the bunch).

Some quick and dirty numbers so far from aida64 benchmarks.

gb - freq - timings - latency (avg) - read - write - copy
32g - 3000 - 14-15-15-35-630-2t - 49.8 - 42218 - 45578 - 42319
16g - 3466 - 14-15-15-35-630-2t - 43.25 - 49988 - 50533 - 45163
16g - 3600 - 14-15-15-35-630-2t - 41.775 - 52644 - 52565 - 48052
32g - 3600 - 15-16-16-35-630-2t - 45.15 - 38600 - 37399 - 41476
32g - 3600 - 15-15-15-32-630-2t - 44.65 - 51719 - 54825 - 51130

Nice, thats some very low latency averages for CL14. With my CL16 set at CL14 (14-14-14-28-252-1st) 54 ns is the lowest I see.
I wonder if the Intel board / cpu has anything to do with it
Why are you using 2t instead of 1t? I presumed 1t was always better but never experimented setting 2t to be honest

 

[Shaolin Chicken]

I worked with Foxconn engineers on tuning their motherboards for a couple of years, the Blood Rage and Katana took some of my input. Unfortunately management changes happened at Foxconn and they abandoned the enthusiast sector just when they were getting some good stuff out the door. Family has happened since then, now I'm a boring network admin and this is just hobby level stuff

I used to have the Foxconn Blood Rage motherboard, I really liked it.
Remember it being expensive in it's day, think I paid around £250.00 for it. It was the X58 platform, so going back some now.

It was renowned for being a great overclocker, though to my shame I never bothered.

 

[retrop]

Nice, thats some very low latency averages for CL14. With my CL16 set at CL14 (14-14-14-28-252-1st) 54 ns is the lowest I see.
I wonder if the Intel board / cpu has anything to do with it
Why are you using 2t instead of 1t? I presumed 1t was always better but never experimented setting 2t to be honest

Hey cheers, yeah I think the Intel board is helping. I was getting latency in mid 50s with the 3000mhz stuff at 16-18-18.

I've only briefly tried 1t this morning with 32gb and it didn't even post. I'll have a fiddle around later but I've just been doing some game stability testing.

Assassins creed Valhalla and FS2020 seem solid so far.

 

[Awahwah]

I worked with Foxconn engineers on tuning their motherboards for a couple of years, the Blood Rage and Katana took some of my input. Unfortunately management changes happened at Foxconn and they abandoned the enthusiast sector just when they were getting some good stuff out the door. Family has happened since then, now I'm a boring network admin and this is just hobby level stuff

It's pretty quick binning B-Die for quality, just set the timings manually - Primaries to 14-14-14-14-28-42-1T and TRRD/TFAW/TWTR to 4-4-16-4-8 (the rest of the timings don't matter) @ 1.35v and see how high it goes before you get a no post. 2 minutes per stick at most, less for the bad ones lol. That was a rare opportunity I had in that we needed 20 kits of memory for upgrades at work and the patriot 4400 C19 was cheaper on offer than we were paying for the usual Kingston trash we get. I bought two kits myself and quick tested the lot, kept the best 4 sticks . Think it took me around an hour in total to bin through 44 sticks.

You be a Network bod, ditto buddy but I have to disagree, it's not boring or I've just been lucky.

 

[Blackjack Davy]

Well I'm revisiting this thread as I've got a new cpu and this is my musings:

If you can't do XMP it may not be the RAM thats at fault it could well be the IMC on cpu I have a set of 3600 CL16 T-force 8pack edition and with my old 3900 could never get faster than 3533mhz no matter how much volts you threw at it earlier bios's would appear to be ok but freeze or bug out later bios's simply wouldn't POST at all I did seriously think about returning the kit. Well I swapped out the 3900 for a 5900X and the results are night and day I can dial in the XMP and it just works. Bingo. Infact it'll go higher.

Also I've always been lead to believe all you have to do select the XMP profile and thats it, away you go I've never been overly interested in overclocking memory never seemed worth the effort for the sake of a few benchmark scores but simply selecting XMP is not enough on these chips it only sets the primary timings 16-16-16-16-36 but the rest of them - the secondarys - it leaves those on Auto - you have to enter them manually from the XMP data on the left and some of the autos are very lax indeed tFRC is 312 in XMP but on auto its 630! Seems I've learnt something afterall.

as my timings are tighter than standard i'm running 1.47v:



i'm aware they could be a bit tighter too, by the looks of things here, but I wasn't interested in the hassle so i just dialed these in from the get-go, as they're pretty much guaranteed to work.

I'll probably only revisit when the ryzen 4000 series proper lands and there's a ram speed boost to take advantage of. I've got an older 3600 which can be a bit twitchy with infinity fabric speed increases and, again, i cba

i have a full list of timings which produced an updated benchmark of 215.80 if you want them?

ps. tRDRD SCL should = 4 not 5

Interesting, I'm done as well and I've got almost the same timings and exactly the same results but for 1.392V on my 3600 CL16 kit. I guess these sets really are binned to do higher frequencies @ lower voltage.

 

[retrop]

Has anyone experienced problems with ram/cpu not working at certain frequencies from cold?

I've only been playing with my kits since yesterday but settings which seemed pretty solid with the PC warmed up seem to fail to boot when resuming from cold.

I had two pull two sticks last night after the PC went to sleep to get it to boot and this morning even the xmp settings with just the two sticks failed to boot the first few times and I had to back the frequency down to 3,466 to get in to Windows.

Pretty sure if I reboot now I've typed this and things have warmed up 3,600 will be possible again...