Use Google Translate http://lab501.ro/featured-articles/amd-ryzen-7-1800x-overclocking-ddr4-si-nvme/8
As far as i can tell all are at 14-14-14-28 , with the one exception at 12-12-12-28
Don't have other reviews with reliable data above 3200MHz but the author of this review should know what he is doing http://hwbot.org/user/matose
As for the differences at 3200, my assumtion is that they might be mostly due to the hidden secondary timings.
Notice how bandwidth starts becoming parallel to x axis. That is what I would expect. Obviously latency will decrease with speed as its Cycle/Freq, math there checks out fine.
I suspect most DDR4 kits people are testing are rated/advertised at 2T command rate, and being stuck in 1T isn't helping efforts on achieving stability I suspect.
I have my latest MSI X370 Titanium bios (Official not Beta, 1.30) and running my G.Skill 3200 (16-18-18-38-2N) at 2400 @ 16-18-18-38 with voltage set manually to 1.30V even though modules are rated for 1.35.
The G.Skill RGB kits dying left and right have me waiting for the dust to settle a bit before going anywhere near 3200 @ 1.35V not until the May BIOS updates roll out anyway.
I notice my DRAM voltage frequently fluctuates as high as 0.02V above manually set 1.30, and that is what I can see with the naked eye through software, who knows what the voltage peak is viewed with a scope, but that could be what's bricking G.Skill RGB modules.
Notice how bandwidth starts becoming parallel to x axis. That is what I would expect. Obviously latency will decrease with speed as its Cycle/Freq, math there checks out fine.
I was wrong. Not thinking straight as its late. Apologies. We are not talking about the same thing
They are not changing timing, synthetic benchmark will increase linearly with freq as CL is constant. I agree
That's not the issue. I am saying if you go to 4000 MHz you will HAVE to increase CL to something like 18/19 (because no mater the freq, rise time and fall time of the timing characteristics remains the same). That will counter balance any increase in performance due to freq. Its nothing to do with Intel vs amd. Its just math.
So the table I linked is still valid for amd. Or at least these results don't change the conclusion.
Sorry about the confusion
@imported_jj: those are using 5.8 pre ryzen software. But it shows what I mean
First i want to make a note that we do NOT know if all Ryzen will be able to run memory at 3600MHz or higher so if anyone decides to purchase such kits, it is a gamble today.
Not all mobos will get support for such memory clocks either.
The scaling is similar in terms of ns but in % it's less relevant for Ryzen. Ofc if Ryzen gets to 60ns at high memory clocks when the dust settles, a few extra ns gained with lower latency would matter more than today.
Ryzen scales much better as Intel has issues and with Ryzen the data fabric (call it a bus or an interconnect) scales with the memory clocks.
They get slightly lower than expected results for Ryzen but that side, again much better scaling for Ryzen as the latency of the fabric deceases too.
So i got tired of waiting for my backordered ram... and bought some to get me up and running now, will keep it for the GF's new build this summer as ill replace it in this system when my faster backordered kit shows up(corsair 2x8GB 3600mhz LPX)
And its alive! Just built the following:
Ryzen 1700 w/ Noctua D15
Asus Prime X370-Pro(flashed new 515 bios from yesterday)
Gskill 2x8GB Ripjaws 3000mhz(running at 2133 for now, it is single rank, unknown IC's)
MSI Gaming X RX480 8GB
Asus Xonar STX soundcard
Corsair 650w RMi PSU
Fractal Design Define S Case
Samsung EVO 500GB M2 OS/apps
Crucial MX300 SSD 2TB for games
WD Black 2TB, for media, DSLR archiving
Windows 10 Home, Retail
So far its stock speeds, just finished installing windows and some apps, burn testing tools, etc.
After running prime95 for a hour CPU hit only 49c according to ryzen master, D15 is doing a great job. This is an amazingly efficient CPU at the low stock speeds, my UPS and corsair PSU software said system was drawing around 190w from the wall and feeding 170w to the system running prime95, PSU fan doesnt even come on(88-90% efficiency according to PSU software).
Ryzen master is telling me its only using 1.1V loaded with prime, CPUz reports 1.15v.
Scores 1300 in cinebench cpu test.
Im going to start cranking the cpu up to see what it can do, then ill work on the ram. Wish me luck!
Any one else having major issues with windows 10? I have installed it 3 times now and it all starts out ok but soon as I do windows updates and the computer needs to do a restart, it no longer see's the SSD till I reinstall windows I cant figure out what the issue is.This only happens after I do the Windows updates. before that I can restart a million times without issues booting back into windows.
Any one else having major issues with windows 10? I have installed it 3 times now and it all starts out ok but soon as I do windows updates and the computer needs to do a restart, it no longer see's the SSD till I reinstall windows I cant figure out what the issue is.This only happens after I do the Windows updates. before that I can restart a million times without issues booting back into windows.
My windows 10 install went smoothly, when i installed it i only had the M2 drive installed, then after windows was installed i added the Crucial SSD and the spinner and it saw them both fine. So far so good.
My windows 10 install went smoothly, when i installed it i only had the M2 drive installed, then after windows was installed i added the Crucial SSD and the spinner and it saw them both fine. So far so good.
Never mind guys I am such a dork. I had the SSD drive and DVD burner drive plugged into Asata and it was causing lockups. went to the user manual and seen the ports named Asata. I quickly changed the cables to the standard sata 6 and wammo im in windows without another reinstall.Its running super smooth now and I am able to open multiple programs at once where before it was stuttering and freezing opening only 2 programs.Boot up is about 7 seconds flat now very happy guys!! First time I ever heard of a port called Asata??
Never mind guys I am such a dork. I had the SSD drive and DVD burner drive plugged into Asata and it was causing lockups. went to the user manual and seen the ports named Asata. I quickly changed the cables to the standard sata 6 and wammo im in windows without another reinstall.Its running super smooth now and I am able to open multiple programs at once where before it was stuttering and freezing opening only 2 programs.Boot up is about 7 seconds flat now very happy guys!! First time I ever heard of a port called Asata??
Well im about to go to bed, got the ram running @2666mhz now, ill leave it on prime95 blend overnight to test if its stable. Then ill give 3000 a shot.
Does anyone know if its possible to drop the 1T command rate down to 2? or is this the problem and why no one can get there ram running fast.
Exact opposite. Well not quite exact opposite but pretty close. The only drive tech to be connected to the CPU is main M.2 connection (PCIe 3.0 4x). All of the SATA ports are handled by the X370/B350 chipset. The chipset supports I believe 6 sata.
Erm NVM. It gets more confusing. So if this slide is right
Then there are two sata ports fed from the SoC. Almost every board offered forgoes Sata-express for M.2 PCIe. The two supported Sata-e connections use PCIe 3.0 2x. We know that the M.2 connections use PCIe 3.0 x4 (for the primary on boards with 2) and that those are fed right from the SoC, implying that the SATA connection for those SATAe are also from the SoC. So when board makers salvage those SATA connections as just SATA only, then they have to be from the SoC.
But those aren't the SATA connections that were connected. A.) ASMedia has nothing to do with the SoC SATA. Those would be AMD if anything. B.) ASMedia actually are responsible for the Chipset as AMD spun off and sold their chipset business to Asmedia. C.) This is the important part and my reason for correcting this, on my board (Asrock Taichi) there are 8 SATA ports that don't have the ASata tag (which is 2 more than the above diagram says I should have including the 2 Sata-e ports and 4 more not counting them), and 2 Asata ports. Those two Asata ports aren't from the SoC its from a extra on board controller (2 x SATA3 6.0 Gb/s Connectors by ASMedia ASM1061, support NCQ, AHCI and Hot Plug).
Backed down my 1700 to 3.6GHz 1.175vcore for now because when I was running handbrake my system would just instantly crash and reboot with the CPU at 3.7GHz 1.25vcore. I'll have to dial in the voltage more but I like the temps so far at that speed.
Yeah luckily I had an opportunity for about 3 weeks of waiting for my board to read up on it and skim the manual. I knew there was an Asmedia controller added but even then I wasn't ready for power up and finding two of my drives connected to A_Sata when they were detected. One of those being my optical drive. I instantly powered off my machine, double checked the port configuration and rewired. Absolutely no reason to use a add-on SATA connection if you don't need to, and unless I was using some workstation Scsi/SAS setup, I wouldn't ever want my boot drive connected to one.
My G.Skill FlareX 3200 is on the QVL for the Gigabyte Gaming 3 (F5 BIOS), but given that the RAM was released after the board/latest BIOS and the fact that it isn't stable on it at any speed above 2400... I'm going with BIOS being the issue. Because it works 100% on 1-click A-XMP on my MSI motherboard.
I'll have to see how it does on the ASRock X370 Taichi and the Gigabyte AX370 Gaming K7.
I finally got an ASRock Taichi delivered when they were momentarily in stock a couple of days ago, and then yesterday I had the chance to put it together.
The GPU (Sapphire Tri-X Fury) and everything else was reused. The RAM is not on the QVL, so anyone wanting ECC might be interested to know that it works fine.
A few thoughts on the motherboard:
It looks really well built, with a thoughtful layout.
Two NVMe slots means I don't need to use any SATA ports at all (this is why I chose this board).
The two screws for the NVMe slots come in individual bags. And I thought wasteful packaging had already reached its peak.
Not enough USB ports (although the built-in Bluetooth frees up one USB port that would have been taken up with an adapter for my headphones).
As of BIOS 1.94A, there are an overwhelming number of options for tweaking.
The southbridge feels pretty hot. They probably should have designed the heatsink to sink heat effectively instead of to look pretty.
1. Ryzen scales much better as Intel has issues and with Ryzen the data fabric (call it a bus or an interconnect) scales with the memory clocks.
2. They get slightly lower than expected results for Ryzen but that side, again much better scaling for Ryzen as the latency of the fabric deceases too.
CL: CAS Latency. The time it takes between a command having been sent to the memory and when it begins to reply to it. It is the time it takes between the processor asking for some data from the memory and then returning it.
tRCD: RAS to CAS Delay. The time it takes between the activation of the line (RAS) and the column (CAS) where the data are stored in the matrix.
tRP: RAS Precharge. The time it takes between disabling the access to a line of data and the beginning of the access to another line of data.
tRAS: Active to Precharge Delay. How long the memory has to wait until the next access to the memory can be initiated.
CMD: Command Rate. The time it takes between the memory chip having been activated and when the first command may be sent to the memory. Sometimes this value is not announced. It usually is T1 (1 clock cycle) or T2 (2 clock cycles).
Theoretically, the time in cycle elapsed between issuing the request and actually getting the data will be tRP + tRCD + CL, which is about 42 cycles for a 3000C14 RAM (14-14-14-34). This delay increases to a delay of 42/3G = 42* .33 ns = 14 n;s for a 3200C15(15-15-15) ; it is 14.1 ns and 3400C16 is 14.1 (16-16-16) , 4000C18 (18-19-19) is also 14 ns.
Hence, to answer your points:
1. that just shows that Ryzen IMC has better memory utilization. K17 can use upto 98% of the theoretical limit while intel can do 85%, IIRC from leaks in January. This lower utilization factor means Intel will scale worse. Itel manages to make up for it by having a better IMC related delay. The two implementations are different and the percentage utilization comes down to the engineering decisions made during coding
2. BUT only with relation to the maximum allowed theoretical capacity. Those graphs you shown has the same timings with all tested Freq. I am yet to see a lower than C16 out of the box timing on any 3400 (14 ns), C15 on 3600 (13.1 ns); and C18 on 4000 (14 ns) and C19 on 4266 (13.4 ns). That's an almost non-existent 0.6 ns save in delay. so performance will remain almost the same between 3000C14 and higher.
The reason for not seeing tests done at higher freq with the same timing on AMD is:
1. AMD does not yet support it <== will be fixed
2. RAM with C12 on 2133 will not do a C12 on 4000, <== this cannot be fixed, because every clock has a fixed rise time, hold time and fall time. the 1/freq must be > t_r +T_h + T_f. as you increase freq, 1/freq will fall and inorder to ensure that 1/freq > t_r +T_h + T_f holds true, we have to increase delay by increasing the CL, tRP and tRCD. ref: check background section. its not a ram based demonstration, but the principal is the same: ftp://ftp.altera.com/up/pub/Altera_Material/10.1/Tutorials/Timequest.pdf
BUT, if you test the entire line from 2133 to 4000Mhz with C19, I am sure we will see the scaling that you are referring to. but this is only an academic test and have no significance in real life.
Backed down my 1700 to 3.6GHz 1.175vcore for now because when I was running handbrake my system would just instantly crash and reboot with the CPU at 3.7GHz 1.25vcore. I'll have to dial in the voltage more but I like the temps so far at that speed.
Well we are using the same motherboard right now. Its the Gigabyte Ab350 Gaming 3. I was really starting to think I got another bad part either the cpu or motherboard and was getting ready to rip it all apart and send it all back to Newegg again. But out of the blue I decided to read the manual for what ever reason. Glad I did. It just doesn't make sense to me that there would be ports included that cant be used?Well actually thats not true, those two ports just cant be used together. If I plug in the SSD drive only to Asata and the DVD drive to the Standard sata 6 then it works. So I am wondering if they are made only for some special kind of Raid set up?Anyways for now I will keep everything plugged into the Sata 6 port connections just to be safe, well that and so it I boot into windows LOL. F5 bios is pretty good so far for me. it lets me run my double sided Crucial ballistix ddr4 2400 no problem.
This site uses cookies to help personalise content, tailor your experience and to keep you logged in if you register.
By continuing to use this site, you are consenting to our use of cookies.