What RAM are you getting/bought for Ryzen

Permabanned
Joined
15 Oct 2011
Posts
6,311
Location
Nottingham Carlton


Same settings 4000% plus. £600 CL17


Nice DDRs shame they not came out when i picked up 3766 Dark Pros. Thing is 3600 cl18 is still slower than super tight 3333mhzcl 14 :D And slower than 3200cl12 but that needs some srs volts. Not to mention going pass 3466 usually tends to cut down maximum cpu stable OC.

How Low on timings on 3200/3333 can Your kits go ??


btw i had somewhere 3672mhz cl16 1000% pass on my system still slower than 3333 :D


Damn id had a play with Your kit my opc been runing memtests 24/7 since start of June when i gotten 3766 kit :D
 
Permabanned
Joined
15 Oct 2011
Posts
6,311
Location
Nottingham Carlton
Originally Posted by The Stilt

Half of the 3DMark Sky Diver Combined Test score comes from the physics score. The physics engine used in 3DMark is Bullet, which is extremely sensitive to latency.

3200MHz 14-14-14-75-350ns-BGSE = 116.85fps
3200MHz 14-14-14-75-350ns-BGSD = 119.51fps
3520MHz 14-14-14-56-177ns-BGSD = 125.07fps
3200MHz 12-12-12-54-140ns-BGSD = 127.07fps

These are with R7 clocked at 3.575GHz.

Those are five run averages, since the score varies slightly between the runs.
3DMark SD CT isn't the best benchmark to evaluate the real world memory performance, however it is quite decent and can be run very quickly.
Quote:
Originally Posted by The Stilt

Higher DFICLK itself doesn't improve the performance. Increasing the DFICLK (by increasing the MEMCLK) only helps if the DFI is the bottle neck.
1600MHz DFICLK (i.e. 3200MHz MEMCLK) is sufficient for > R7 1800X and the returns from higher speeds are diminishing. At higher CPU speeds (which are unachievable on current gen. Ryzen CPUs) the required DFICLK for optimal performance will increase as well.

https://m.imgur.com/68n8LJN
 
Permabanned
Joined
15 Oct 2011
Posts
6,311
Location
Nottingham Carlton
8 pack Could You test Your kit at those settings ??
3466mhz
DRAM CAS# Latency [14]
DRAM RAS# to CAS# Read Delay [14]
DRAM RAS# to CAS# Write Delay [14]
DRAM RAS# PRE Time [14]
DRAM RAS# ACT Time [28]
Trc_SM [54]
TrrdS_SM [6]
TrrdL_SM [9]
Tfaw_SM [36]
TwtrS_SM [4]
TwtrL_SM [12]
Twr_SM [12]
Trcpage_SM [Auto]
TrdrdScl_SM [2]
TwrwrScl_SM [2]
Trfc_SM [333]
Trfc2_SM [Auto]
Trfc4_SM [Auto]
Tcwl_SM [14]
Trtp_SM [8]
Trdwr_SM [6]
Twrrd_SM [3]
TwrwrSc_SM [1]
TwrwrSd_SM [7]
TwrwrDd_SM [7]
TrdrdSc_SM [1]
TrdrdSd_SM [5]
TrdrdDd_SM [5]
Tcke_SM [6]
ProcODT_SM [60 ohm]
Cmd2T [1T]
Gear Down Mode [Disabled]

AMD CBS > UMC Common Options > DDR Memory Mapping > BankGroupSwap [Disabled] (Note: As kit was 1 DPC SR)

VBOOT/VDIMM: 1.365V



This is Fastest option for ryzen on 1.0.0.6a atm without cutting down max CPU OC by more than 25mhz :) Thats fastest option we found so far.


Very interested :D love messing around with memory timings hhahahha
 
Associate
Joined
29 Jun 2016
Posts
825
@Lord_ZED Do you think Bankgroupswap disabled is an improvement? I've just run a bunch of benches and noticed big improvements with BGS enabled.

In some benches I've been seeing around a 7% improvement enabled with 3200C14 samsung b die
 
Soldato
Joined
17 Nov 2007
Posts
3,164
@Lord_ZED Do you think Bankgroupswap disabled is an improvement? I've just run a bunch of benches and noticed big improvements with BGS enabled.

In some benches I've been seeing around a 7% improvement enabled with 3200C14 samsung b die

It was noted on OC.NET that BGS showed reductions of up to 10% in benchmarks but improved overall performance, was posted by The Stilt.
 
Associate
Joined
29 Jun 2016
Posts
825
It was noted on OC.NET that BGS showed reductions of up to 10% in benchmarks but improved overall performance, was posted by The Stilt.

Enabled or disabled? I've run a bunch of stuff through CPUZ, Cinebench, Realbench, Firestrike, Timespy and Super Position. Every single one showed pretty significant gains. Surely the likes of FS, TS and SP are representative or actual gaming performance?
 
Associate
Joined
29 Jun 2016
Posts
825
Disabled, apparently not needed if you are only running two dimms.
Interesting. Well as far as 'not needed if only running two dimms', i'm seeing fairly dramatic improvements from enabling compared to disabling. Do you have any suggestions for tests I can run that'll be more representative of 'overall performance increase' than the benches I've been using?
 
Soldato
Joined
17 Nov 2007
Posts
3,164
Interesting. Well as far as 'not needed if only running two dimms', i'm seeing fairly dramatic improvements from enabling compared to disabling. Do you have any suggestions for tests I can run that'll be more representative of 'overall performance increase' than the benches I've been using?

I will see if I can find the post from Stilt on OC.NET.
 
Soldato
Joined
17 Nov 2007
Posts
3,164
You also seem to have BankGroupSwap enabled. A no no on 1 DPC SR configuration.

Yes, disabling BankGroupSwap will improve the real world performance by couple percents.

If you are using dual rank or 2 DPC single rank modules, you don't touch these options at all. This results in BankGroupSwap = Enabled and BankGroupSwapAlternative = Disabled (the configuration I was using).

With 1 DPC single rank configuration you should either disable both of them, or set BankGroupSwapAlternative = Enabled.

Enabling BankGroupSwapAlternative has the same positive effect on the real world performance as disabling BankGroupSwap does, however it doesn't have it's down sides (i.e. the lower reported bandwidth figures).

The default configuration is BankGroupSwap = Enabled, BankGroupSwapAlternative = Disabled.
These two options are mutually exclusive, meaning they can both be disabled but they cannot be enabled simultaneously.

Disabling BankGroupSwap will improve the real world performance (by couple percent), however the reported bandwidth (e.g. AIDA) decreases by < 6%. Enabling BankGroupSwapAlternative has nearly the same positive effect on the real world performance, while the reported bandwidth remains at the same level with BankGroupSwap = Enabled.

Neither of the BankGroupSwap options should be touched, unless 1 DPC SR modules are used.

AIDA64 can't always be trusted.

For example disabling BankGroupSwap, which is the right thing to do with 1DPC SR configuration reduces the bandwidths reported AIDA by ~2GB/s each. Meanwhile the performance improvement from disabling the BGS is < 4% in latency sensitive scenarios (e.g. HITMAN DX12 w/ CPU being a bottle neck).

Here are some of The Stilts comments on BGS, I am running single rank memory so disabled it :D
 
Soldato
Joined
17 Nov 2007
Posts
3,164
Have you done any testing for yourself though? As I also have single rank memory, but testing is clearly showing a decent improvement when enabled.

As per Stilts comments I did see a slight decrease in bandwidth in AIDA64 but machine has been stable and performing well so left it disabled.
 
Associate
Joined
10 Apr 2009
Posts
10
Location
Dorset, UK
Sorry guys, I know this is a Overclocking forum, but could anyone suggest a pair of 8GB DDR4 (16gb total) for a Ryzen 1600 with a B350M Mortar Artctic MSI motherboard that works out of the box without needing to overclock ? I just really want the fastest compatible memory, I am not in to overclocking. I keep hearing conflicting things that 2400 is best but then some saying they have 3200 and so on, arrgggh, brain hurts. Youtube only seems to deal in overclocking. My PC would be for video encoding so just need something I can shove on the M/B and it works. I don't even understand why the letters "OC" are in brackets on the MSI website beside the compatible memory, does this mean to reach that speed it needs to be overclocked or does that mean it is OC'd out of the box. I am new to Ryzen and building my own stuff. Looking to spend around £130. Thanks all !!
 
Back
Top Bottom