i9-9900KS was melting VRMs, so I went Team Red

[Haro]

5 hours ago, ValkyrieStar said:

CPU CCDx (Tdie) is the spontaneous measurement of each die (im presuming it's hotspot temp),

CPU Die (average) reading is a smoothened out version of much the same,

CPU (Tctl) is just some mad yoyo wierd ass thing that some derp coded and they forgot to fix ever.

Usually I found that Tdie is the most accurate within ryzen master and the TCTL to be the most inconsistent

And I kind of never really looked at cpu die as whenever I monitor a ryzen cpu I only look at the general temps or the tdoe or tctl 

5 hours ago, ValkyrieStar said:

insanely meltingly high temperature of 57c (/s)

Better than 130c I guess 

5 hours ago, ValkyrieStar said:

With PBO disabled and AutoOC off it hovers around the 3.9-3.95GHz mark, PBO limits disabled about 4.0GHz, then AutoOC also set to +200MHz gets about 4.1GHz avarage.

And that's with encoding ?! Holy I didn't think the 3600 was that multitasking of a beast u Guess the stupid good ipc count is doing its job lol

 

5 hours ago, ValkyrieStar said:

With lighter threaded loads, all cores can achieve 4.2GHz with no AutoOC, and 4 of 6 cores hit 4.4GHz with AutoOC to +200MHz, the last two cores not far behind either. I must not have gotten too bad a sample to get that big of a boost with AutoOC. Single core perf on Cinebench R15 is almost on par with the 9900K/KS with 209, multithreaded isn't too far off either, with 1660.

Wow the 3600 is a beast 

I always looked at it's downsides such as the bad imc having to deal the meh FCLK having to deal the boards AGESA code and the wakyness if amd

But this is amazing 

I'm glad your up and running against

[ValkyrieStar]

On 6/4/2020 at 8:18 AM, TofuHaroto said:

Usually I found that Tdie is the most accurate within ryzen master and the TCTL to be the most inconsistent

And I kind of never really looked at cpu die as whenever I monitor a ryzen cpu I only look at the general temps or the tdoe or tctl 

Better than 130c I guess 

And that's with encoding ?! Holy I didn't think the 3600 was that multitasking of a beast u Guess the stupid good ipc count is doing its job lol

 

Wow the 3600 is a beast 

I always looked at it's downsides such as the bad imc having to deal the meh FCLK having to deal the boards AGESA code and the wakyness if amd

But this is amazing 

I'm glad your up and running against

Yep, it sits nicely at about 4.1ghz consuming about 100w, actual temps are a little on the higher than expected side tbh, at around 72-75c ish. Full bore prime95 doesn't push it much over 85c though, at about 4.05ghz and 130w (1.27v), this particular chip seems to be quite power hungry from the looks of things, probably why it ended up with no X at the end of its name. I don't feel like i'll be able to get an all-core of more than 4.2ghz (i reckon that's as far as i'd get with about 1.25-1.275v being the max it'll take before being too hot), and in which case i'd be sacrificing the 4.4ghz lighter load speeds (i did a quick test of an overclock though, since there's no P-States, it just sits there ramming voltage into the chip) in favour for slightly better all core load speeds, despite not caring about this CPU a whole lot, i'd rather let it do it's own thing.

 

I spent a few hours yesterday tweaking memory, i managed to get it from the stock 16-18-18-36 @ 3200 to a nice 14-16-16-32 @ 3466 with no additional voltage, performance in benches was unaltered, and i was met with some wierd instability that memtest86, prime95 large ffts, hci memtest all failed to find, this instability wouldn't go away when dropping it to 14-16-16-32 @ 3200 or by increasing voltages, so i bit the bullet and ditched that and just went back to XMP rated timings, which by chance randomly started scoring better than they did before..

 

The only main gripe i have/had/ish with this board (and it seems with AMD boards in general) is that the POST times can send you to sleep. Default was something like 24 seconds, which was just absolutely painfully slow, memory fast boot does literally nothing (idk why, i tested cold boot, reboot, and warm boot and memory fast boot didn't affect it at all). I eventually got it down to around 16.5 seconds by disabling CSM, disabling the wifi, and disabling audio which i don't use (i have usb external card), so that was a nice improvement. I might email MSI see if they have any comment about why Memory Fast Boot isn't actually doing anything.

 

I figured i'd try setting the PCIe gen to gen3 for all the slots since i have no 4.0 devices, and that dropped post time to 10.6 seconds, very nice! Then i realised this also included the cpu-chipset link, and my two NVMe drives were being bottlenecked a little, i set the chipset up to gen4 and the post time didn't change at all, i then set all the settings back up to gen4 (rather than auto) and it maintained the nice 10.6s boot time, wierd. On a side note, don't enable the "pcie 4gb address mode crypto" rubbish, it increased post times by like 5 seconds and had some other non favourable side effects.

 

I'll have a go with tweaking some other bits and bobs at some point but i'm not sure what else there even is to tweak to improve post times. I'm just used to pressing the power button on my Z390/9900KS and about 4 seconds later i was at the login screen, post times in that were 2.7s, and windows fast boot bringing up the login screen almost immediately after that.

 

4950X gonna rip through some work really nice i bet...

[Guest]

On 6/3/2020 at 6:41 PM, ValkyrieStar said:

 

For temp monitoring i've found that;

CPU CCDx (Tdie) is the spontaneous measurement of each die (im presuming it's hotspot temp),

CPU Die (average) reading is a smoothened out version of much the same,

CPU (Tctl) is just some mad yoyo wierd ass thing that some derp coded and they forgot to fix ever.

 

Huh? lol. Nice.

 

Here, let me help you.

 

-CPU (onboard sensor): This is either a temperature measured by a dedicated sensor on mainboard located inside the CPU socket (the external CPU temperature) or temperature obtained from internal CPU thermal sensor (i.e. DTS = Digital Thermal Sensor).

-Core Max: The maximum temperature among all cores in the CPU.

-CPU (Tctl): This is the T_control temperature available on AMD CPUs only. On several generations before Zen (Ryzen), this is not a reliable representation of the temperature. On AMD Zen series this is the temperature used to control cooling and is a fixed offset from the real CPU temperature. Offset is used mostly on X-series and some Threadripper CPUs; in such case two values are shown: Tctl and Tdie. If no offset is used, then only a single value is shown as Tctl/Tdie, which equals the real temperature.

-CPU (Tdie): This value is shown in case the CPU uses an offset from Tctl and represents the real temperature (Tdie = Tctl - Tctl_offset).

-CPU Package: Shown on Intel CPUs represents a 256-millisecond average value (calculated by CPU) of the hottest temperature sensor within the CPU package.

-CPU Package (TSI): Available on pre-Zen AMD CPUs is the CPU temperature obtained via TSI interface.

-Core #n (n=any number): Actual temperature of a particular CPU core.

-CPU IA Cores: Maximum temperature among all computing (x86) cores in CPU (so part of CPU except Uncore and Graphics logic).

-CPU GT Cores: Temperature of the integrated graphics part of CPU (if present).

https://www.hwinfo.com/forum/threads/cpu-temp-sensors-explanation.5597/

[ValkyrieStar]

2 minutes ago, ShrimpBrime said:

Huh? lol. Nice.

 

Here, let me help you.

 

-CPU (onboard sensor): This is either a temperature measured by a dedicated sensor on mainboard located inside the CPU socket (the external CPU temperature) or temperature obtained from internal CPU thermal sensor (i.e. DTS = Digital Thermal Sensor).

-Core Max: The maximum temperature among all cores in the CPU.

-CPU (Tctl): This is the T_control temperature available on AMD CPUs only. On several generations before Zen (Ryzen), this is not a reliable representation of the temperature. On AMD Zen series this is the temperature used to control cooling and is a fixed offset from the real CPU temperature. Offset is used mostly on X-series and some Threadripper CPUs; in such case two values are shown: Tctl and Tdie. If no offset is used, then only a single value is shown as Tctl/Tdie, which equals the real temperature.

-CPU (Tdie): This value is shown in case the CPU uses an offset from Tctl and represents the real temperature (Tdie = Tctl - Tctl_offset).

-CPU Package: Shown on Intel CPUs represents a 256-millisecond average value (calculated by CPU) of the hottest temperature sensor within the CPU package.

-CPU Package (TSI): Available on pre-Zen AMD CPUs is the CPU temperature obtained via TSI interface.

-Core #n (n=any number): Actual temperature of a particular CPU core.

-CPU IA Cores: Maximum temperature among all computing (x86) cores in CPU (so part of CPU except Uncore and Graphics logic).

-CPU GT Cores: Temperature of the integrated graphics part of CPU (if present).

https://www.hwinfo.com/forum/threads/cpu-temp-sensors-explanation.5597/

I flat out ignore the motherboards "cpu temp" sensor, as it's often inaccurate unless it's reading one of those CPU sensors, in which case, i still ignore it anyway.

 

For Zen2 (Ryzen 3000);

 

CPU (Tctl/Tdie) = a wierd averaged thing, that the SMU ramps up with any spike of Tdie (even ones that aren't picked up by monitoring software), but ramps down much slowly, at idle this is often routine stuff like system logging or whatever, and can lead to annoying as hell fans which ramp up every few seconds, this is the one which produces a sawtooth like graph.

 

CPU CCDx (Tdie) = is the instantaneous hotspot temperature of die "x", on a single die chip there'll only be CCD1, on the 3900X or 3950X there'll be both CCD1 and CCD2, it's updated at something like every 1ms. Watching this can produce a very spikey graph at load loads/idle.

 

CPU Die (average) = a moving average temperature (not sure what kind of timescale it's averaged over, maybe 500ms or something) which produces a much more reasonable and readable graph and would also be a way better thing to base the fan speeds on rather than Tctl, but unfortunately that's not something we can set in the BIOS. It produces a graph similar to CPU Package on Intel.

 

 

As you see, CPU (Tctl) is producing a constant sawtooth graph, which makes for a really f*cking annoying fan ramping noise when just chilling watching youtube unless you manually tweak away from that ramping range. CPU CCD1 (Tdie) is the hotspot measured at that moment (since it updates so fast, HWinfo "misses" many of the spikes here). CPU Die (average) is an averaged reading of the Tdie. Ryzen Master actually uses the CPU Die (average) reading.

 

The other temperature sensors you mention are indeed, Intel or older AMD ones.

[Guest]

35 minutes ago, ValkyrieStar said:

 

The other temperature sensors you mention are indeed, Intel or older AMD ones.

That was a direct quote from the Author of HWInfo64. I left the link for you to examine. 

You can increase the update rate of HWinfo64 btw.

 

 

 

[ValkyrieStar]

3 minutes ago, ShrimpBrime said:

That was a direct quote from the Author of HWInfo64. I left the link for you to examine. 

You can increase the update rate of HWinfo64 btw.

 

 

 

True, increasing the update rate too much though can cause high cpu usage, 1 second update speed is plenty in my opinion.

[Guest]

4 minutes ago, ValkyrieStar said:

True, increasing the update rate too much though can cause high cpu usage, 1 second update speed is plenty in my opinion.

A big deal with "temp spikes" is strictly transistor density.

 

Your 1950X has 9,600,000,000 (9.6B) 14nm 180w

The 9900Ks is 1.736B 14nm 125w 

 

 

[ValkyrieStar]

13 minutes ago, ShrimpBrime said:

A big deal with "temp spikes" is strictly transistor density.

 

Your 1950X has 9,600,000,000 (9.6B) 14nm 180w

The 9900Ks is 1.736B 14nm 125w 

 

 

I need to update my system specs, i was running an i9-9900KS, now i'm on an R5 3600 waiting till the 4950X arrives someday.

[Guest]

18 minutes ago, ValkyrieStar said:

I need to update my system specs, i was running an i9-9900KS, now i'm on an R5 3600 waiting till the 4950X arrives someday.

You're good. Was just trying to help that's all. Nothing more than to dig up some good useful information. Hopefully you get a little better idea about the program.

I've had a good look at the white papers for Ryzen processor family 17h. The chip is entirely run on algorithms and sensors. It determines everything by the information it receives. It knows if you're sustaining cold temps or not. 

 

To understand SenseMi, is to actually use the term artificial intelligence, or a dumb version of being self aware.

 

To experiment this theory, I used 3 different cooling methods to see where the cpu would give me the max boost. 

 

Air: Ambient - Stock Cooler. All core boost clocks 3.8ghz, temps 87c on average

Water Ambient - All core boost increased to 3.9ghz temps about 78c on average

Water (chilled) Delta 8.8c All Core Boost clocks 4000mhz temps about 67c average

Tec (chilled see avatar <-- ) idle -30c  boost to max XFR 4350mhz all core instead of 1, load 4000mhz max load temp 28c

Power settings did effect clocks. Above is performance mode all core Everything in bios on Automatic including the memory.

 

Here is an example of TEC experimenting. 

The cooler you run them, the more they boost themselves. But you want to be below 70c on a load

 

 

 

[Haro]

2 hours ago, ValkyrieStar said:

spent a few hours yesterday tweaking memory, i managed to get it from the stock 16-18-18-36 @ 3200 to a nice 14-16-16-32 @ 3466 with no additional voltage, performance in benches was unaltered, and i was met with some wierd instability that memtest86, prime95 large ffts, hci memtest all failed to find, this instability wouldn't go away when dropping it to 14-16-16-32 @ 3200 or by increasing voltages, so i bit the bullet and ditched that and just went back to XMP rated timings, which by chance randomly started scoring better than they did before..

If you want you can maybe play around with the soc voltage 

A little too high soc voltage might result in even worse results but something like 1.1 is totally gonna help

You probably got better results with the xmp profile because of the FCLK 

It was most likely not boosting to what even multipler you set and the interconnect was probably getting really stressed 

2 hours ago, ValkyrieStar said:

The only main gripe i have/had/ish with this board (and it seems with AMD boards in general) is that the POST times can send you to sleep. Default was something like 24 seconds, which was just absolutely painfully slow, memory fast boot does literally nothing (idk why, i tested cold boot, reboot, and warm boot and memory fast boot didn't affect it at all). I eventually got it down to around 16.5 seconds by disabling CSM, disabling the wifi, and disabling audio which i don't use (i have usb external card), so that was a nice improvement. I might email MSI see if they have any comment about why Memory Fast Boot isn't actually doing anything.

 

Long boot ? 

This is really weird I mean if CSM at least half way fixed it (which I'm not surprised that it did lol) then that's good 

But long boot times on AMD ? I wonder what that has to do with it 

Generally fast boot isn't the best and may even result in worse boot times 

2 hours ago, ValkyrieStar said:

i set the chipset up to gen4 and the post time didn't change at all, i then set all the settings back up to gen4 (rather than auto) and it maintained the nice 10.6s boot time, wierd. On a side note, don't enable the "pcie 4gb address mode crypto" rubbish, it increased post times by like 5 seconds and had some other non favourable side effects

Dmi 4 is really nice

2 hours ago, ValkyrieStar said:

I'm just used to pressing the power button on my Z390/9900KS and about 4 seconds later i was at the login screen, post times in that were 2.7s, and windows fast boot bringing up the login screen almost immediately after that.

Oh intel

Even with dmi 3 they still have the little bird and bobs just mastered 

Good thing post time (at least for me ) isn't as important as raw power

[ValkyrieStar]

On 6/6/2020 at 7:09 AM, TofuHaroto said:

If you want you can maybe play around with the soc voltage 

A little too high soc voltage might result in even worse results but something like 1.1 is totally gonna help

You probably got better results with the xmp profile because of the FCLK 

FCLK was set correctly (it automatically followed with RAM)

Both XMP and manually OCing to 3200 or 3466 both got 1.1v SoC automatically

 

On 6/6/2020 at 7:09 AM, TofuHaroto said:

Long boot ? 

This is really weird I mean if CSM at least half way fixed it (which I'm not surprised that it did lol) then that's good 

But long boot times on AMD ? I wonder what that has to do with it 

Generally fast boot isn't the best and may even result in worse boot times 

At first i couldn't find the CSM setting, it's a little oddly placed in MSI bioses, but it was 22-25 seconds boot time with that, and about 16.5s set to UEFI only. Memory Fast Boot is supposed to skip some memory checks, (i'm not on about UEFI Fast Boot, which isn't even a thing on Ryzen - at least for MSI), it's generally safe to use unless memory overclocking.

 

I started getting some wierd issues where boot was taking really long, sometimes 10 seconds, sometimes 16, sometimes 22 or even higher, one time it was even 52 seconds, not sure what was happening there, but it wasn't being consistent at all (i left memory fast boot disabled since it didn't seem to make any difference). Once it started being inconsistent, there was nothing i could do that'd make it stop behaving so wierdly. I even tried reflashing the BIOS and doing all settings from fresh, it just wouldn't have it.

 

I went back to the latest non-beta bios which is A30 currently, A42 was the latest beta, and despite the slightly older AGESA, and slightly slower post (11.2s), it's much more consistent and stable. Even powering on the system after it'd been powered off at the wall resulted in the same post time. 10-11s is acceptable enough for me. (my old threadripper board was around 13-14 seconds, which is why i was so confused how it was being sooo slow to boot since it's got like half of the hardware).

 

I'm not sure entirely what windows scheduler is trying to do, but it keeps putting games on the first CCX, hammering the crap out of them (all those 6 threads), and maybe like one or two threads on the second CCX, like, i know it's a different CCX, but surely 6 actual cores with tiny bit of latency is better than 3 cores in a single CCX, even ffmpeg seems a little scared of the last 2 cores lol, since they're all capable of pretty much 4.4ghz i might try poke it into just ignoring the "better" core thing. The last 2 cores (which happen to be the worst, average over 90% of their time in sleep states).

[Haro]

12 minutes ago, ValkyrieStar said:

but surely 6 actual cores with tiny bit of latency is better than 3 cores in a single CCX

Interms of performance it is but you might encounter worse 1% lows 

 

13 minutes ago, ValkyrieStar said:

I'm not sure entirely what windows scheduler is trying to do, but it keeps putting games on the first CCX, hammering the crap out of them (

Oh let's not even begin with windows scheduler

It's been really bad since zen 1 and while it has been improved it still has some room left to get alot better

[ValkyrieStar]

10 minutes ago, TofuHaroto said:

Interms of performance it is but you might encounter worse 1% lows 

 

Oh let's not even begin with windows scheduler

It's been really bad since zen 1 and while it has been improved it still has some room left to get alot better

Well I cured it of being scared of cores by disabling SMT, generally i run gsync so i cant tell if its improved anything, but stuff is being spread out across all cores now (its still barely tickling the last two cores unless i force it to though)

 

Because of how my encoding work is mostly 1080p, it doesn't scale much beyond 16 threads, and it performs much better with 16 cores vs 8core with smt, so my plan *was* to get the 4950X and disable SMT, but that has the wierd side effect of making sleep mode disappear. Wonder if that's something that'll ever get fixed.

 

I do know Ryzens SMT does scale better than Intels HT due to differing implementations and bigger L3 cache, but windows scheduler just seems to forget that SMT exists and uses both threads of the "faster" cores before using the "slower" cores. My 9900KS would spread stuff out fairly evenly between all the main cores and only after that would it start using the second thread of each core.

 

*edit while still writing*
 

I disabled CPPC Preferred Cores in the BIOS, now it's distributing stuff pretty evenly, it's even preferring to only load up one thread per core which is preferable too. I'll go as far as saying its actually running cooler and using less power (the whole cpu is sitting at about 3ghz in valorant, rather than 3 cores being pegged at 4.3-4.4ghz constantly), and i have all 12 threads again, nice! I guess i'll just be running two encodes at the same time to fully utilise the 4950X.. damn that thing gonna rip!

 

The preferred cores thing supposed to hint to windows to *prefer* using certain "faster" cores first, but it seems windows just dumps everything it can onto those cores and is scared to touch the other horribly slow crippled weaker cores that can only hit 4.375ghz instead of 4.400ghz

[The Torrent]

If you can touch the silicon for 3 seconds then the silicon will survive.

[Haro]

15 minutes ago, ValkyrieStar said:

disabled CPPC Preferred Cores in the BIOS, now it's distributing stuff pretty evenly, it's even preferring to only load up one thread per core which is preferable too. I'll go as far as saying its actually running cooler and using less power (the whole cpu is sitting at about 3ghz in valorant, rather than 3 cores being pegged at 4.3-4.4ghz constantly), and i have all 12 threads again, nice! I guess i'll just be running two encodes at the same time to fully utilise the 4950X.. damn that thing gonna rip!

 

Well good thing

The 4950x and even zen 3 as a whole is expected to be the killer of Intel 

Apparently it will have a 20% ipc improvement while still being 7nm 

Honestly I just want a better imc and FCLK out of it 

It's at a state where is it good enough now but no where near Intel's imc 

I mean I have seen Intel imc get to 4800 with minimal VCCIO and VCCSA bumps and it was going smooth 

With Ryzen's soc and imc there is no way in hell you'll be able to get to that especially with the interconnect and the FCLK 

 

[Haro]

16 minutes ago, The Torrent said:

If you can touch the silicon for 3 seconds then the silicon will survive.

?

[The Torrent]

Just now, TofuHaroto said:

?

all i remember from electronics is if you can touch a transistor for 3 seconds without burning yourself it will take the heat.

im sure it somehow applies to vrm's and idk if i could touch 100 and something c for 3 seconds.

[ValkyrieStar]

11 minutes ago, The Torrent said:

If you can touch the silicon for 3 seconds then the silicon will survive.

I could not touch the heatsink, it would burn my fingers almost instantly, it made my finger feel tingly fuzzy for a good while after lol

 

Got a new board and CPU, all is good now, re-learning how to beat Ryzen into submission to make it do what its supposed to

 

Mosfets themselves are rated to be able to run at up to 125c and will do so generally quite happily (though at a reduced max current output). It's more what damage it's doing to other nearby components, capacitors, inductors, usb controllers, wifi etc. My USB ports would sometimes get flakey and my mouse would flip out when the vrm had been really hot for some time.

[Haro]

3 minutes ago, ValkyrieStar said:

Got a new board and CPU, all is good now, re-learning how to beat Ryzen into submission to make it do what its supposed to

 

Oh this thread has been going for so long

[ValkyrieStar]

1 hour ago, TofuHaroto said:

Oh this thread has been going for so long

True lmao, disabling the CPPC feature really help spread the load on Destiny 2 and CoD MW, before they were completely choking up cores 0,1,2 and 3,4,5 were basically idle doing nothing, now all cores are averaging about 4.3ghz and im no longer getting any stutter from it choking up the CPU cores. GSync is doing a great job of smoothing any microstutter (if there is any) from it being across two CCX's.

 

 

1 hour ago, TofuHaroto said:

Honestly I just want a better imc and FCLK out of it 

It's at a state where is it good enough now but no where near Intel's imc 

I mean I have seen Intel imc get to 4800 with minimal VCCIO and VCCSA bumps and it was going smooth 

With Ryzen's soc and imc there is no way in hell you'll be able to get to that especially with the interconnect and the FCLK 

Honestly, for me i have some 32GB of 3200 mhz, it'd be more hassle than it's worth (and more expense) to get more ram. I'll keep a lookout for any offers on some good low CL high mhz kits (maybe 3600 C15 or something). I'd just run them at XMP and call it a day.

[Haro]

1 minute ago, ValkyrieStar said:

disabling the CPPC feature really help spread the load on Destiny 2 and CoD MW, before they were completely choking up cores 0,1,2 and 3,4,5 were basically idle doing nothing, now all cores are averaging about 4.3ghz and im no longer getting any stutter from it choking up the CPU cores. GSync is doing a great job of smoothing any microstutter (if there is any) from it being across two CCX's.

daaang i thought enabling CPPC would make things better for people holy 

i mean i am still running an old and aging kaby lake i7 but it does the job for me 

ipc count is still decent and even tho i dont do any AVX etc the fup is still pretty good 

rendering a 4k video doesnt take more than 3-5 minutes 

[ValkyrieStar]

On 6/7/2020 at 2:24 PM, TofuHaroto said:

daaang i thought enabling CPPC would make things better for people holy 

i mean i am still running an old and aging kaby lake i7 but it does the job for me 

ipc count is still decent and even tho i dont do any AVX etc the fup is still pretty good 

rendering a 4k video doesnt take more than 3-5 minutes 

I don't know, it probably helps when only doing 1 thing at a time, but as soon as you're doing more than 1 thing, those 2 or 3 fastest cores start capping out and stuff starts choking up and stuttering. Disabling CPPC stopped my Destiny 2 and CoD MW stuttering completely. With it enabled i'd also get stuttery youtube while running encode or simulations, but with CPPC disabled it's possible to watch youtube without stutters and the whole system feels more responsive in general. Wierd.

 

I've been eyeing up some 3600 @ 15-15-15-35 G.Skill ram, it's 50% more expensive than my current kit. But i guess it'd be a decent investment vs 3200 @ 16-18-18-36. DDR4 isn't going anywhere for a couple years at least, and once i've got everything in this system i won't be upgrading for some time except for GPUs. Plus i've a couple friends that could use my current RAM. Not sure though, opinions?

[Haro]

1 hour ago, ValkyrieStar said:

I've been eyeing up some 3600 @ 15-15-15-35 G.Skill ram, it's 50% more expensive than my current kit. But i guess it'd be a decent investment vs 3200 @ 16-18-18-36

I guess in this kit your investing in a b die kit more than the actual rating 

The imc should OC way more 

But like this kit at 15-15-15 is b die binned

It should be way better 

Is it worth paying extra 50% in price

Imo not really but it comes down to your decision

1 hour ago, ValkyrieStar said:

Disabling CPPC stopped my Destiny 2 and CoD MW stuttering completely.

Even I'm games ?!

I thought if would actually make the 1% better because of reduced latency 

Wow I'm surprised 

[ValkyrieStar]

3 minutes ago, TofuHaroto said:

I guess in this kit your investing in a b die kit more than the actual rating 

The imc should OC way more 

But like this kit at 15-15-15 is b die binned

It should be way better 

Is it worth paying extra 50% in price

Imo not really but it comes down to your decision

It's listed in the QVL for my motherboard for use with 4 modules too, the kit only comes with 2 modules, but 2 kits.. 4 modules.. spicy!

 

For the IMC 3600 with 4 single rank modules should be easier to run than 3200 with 4 dual rank modules on the memory controller. I read dual rank modules are slightly faster than single rank when comparing a single module, but i wonder if that changes when there's two modules per channel rather than one. It's also only gonna push the infinity fabric up to 1800 which most/all chips should be capable of. I don't wanna both with the hassle of trying to get 1900 IF working with a 3800 kit, and i don't really wanna bother messing with timings, so i think this 3600 @ 15-15-15-35 kit should be a perfect match, it's already real tight and the IF should have no problem keeping up.

 

8 minutes ago, TofuHaroto said:

Even I'm games ?!

I thought if would actually make the 1% better because of reduced latency 

Wow I'm surprised 

Yep, was sat in a discord call on destiny 2 with a few friends, screensharing the live event for our friend who doesn't have access, at first it was choking up cores 1-3 literally 100% usage across all of them, stuttering the game quite badly, while the other 3 cores pretty much just sat idle. Figured i'd just go brb 2 minutes to change a single bios setting, came back and stuttering was gone both in destiny 2 and on discord, whole pc was more responsive, all cores saw roughly equal load and i even was watching a youtube video on my second monitor whilst the event was taking its jolly time to play out with no issue whatso ever. It feels like i'm on an intel system if i'm honest. Even my ThreadRipper with 16 cores at 4ghz didn't cope this well.

[Haro]

1 minute ago, ValkyrieStar said:

For the IMC 3600 with 4 single rank modules should be easier to run than 3200 with 4 dual rank modules on the memory controller

Oh it's so single rank 

I thought both kits were single and so technically would have been just a slight difference but NVM that's worth it 

2 minutes ago, ValkyrieStar said:

and i don't really wanna bother messing with timings, so i think this 3600 @ 15-15-15-35 kit should be a perfect match, it's already real tight and the IF should have no problem keeping up.

Yea I agree 

Working with ram and the imc and the interconnect with ryzen is a pain

I personally find it fun and interesting just trying to figure out which corner is bottlenecking the other but I can see how it can be annoying

3 minutes ago, ValkyrieStar said:

Yep, was sat in a discord call on destiny 2 with a few friends, screensharing the live event for our friend who doesn't have access, at first it was choking up cores 1-3 literally 100% usage across all of them, stuttering the game quite badly, while the other 3 cores pretty much just sat idle. Figured i'd just go brb 2 minutes to change a single bios setting, came back and stuttering was gone both in destiny 2 and on discord, whole pc was more responsive, all cores saw roughly equal load and i even was watching a youtube video on my second monitor whilst the event was taking its jolly time to play out with no issue whatso ever. It feels like i'm on an intel system if i'm honest. Even my ThreadRipper with 16 cores at 4ghz didn't cope this well.

Yea that I just found this out 

I remember when zen 2 came out there were some complaints about ccxs and core arrangements and one of the fixes was to enable CCPC

And it worked for the most part no stuttery 1% lows no nothing

But maybe once the patches got better the better cores got arranged more equally through the ccxs  or something

Or it straight up just was the case with more multi tasking work loads like streaming a video on discord

I wonder if it is better to enable CPPC on like a 3900x where the single ccxs hold way more cores than let's say a 3600