But It still doesn't explain the root Issue. FMA3 we knew, why FMA3 is still not certain, from what I last heard it only affected CPU's at stock voltages and overvolted CPU's where safe since the FMA3 section got enough power.
The whole :) a patch is coming. It was my understanding that under certain conditions the FMA3 was literally a starving for power. As to what in that specific workload causes it ?
After more testing, Not as stable at 3200 as I first thought. Won't boot from cold. Have to boot at 2666 into windows first then switch over every day. But proof it works. Just the trimming on post when cold that has issues. Have yet to try LN2 mode.
Also does not like to run with overclock either. Can manage 4.1GHz at 1.425v stable. Not running 24/7 just testing. The new bios should be out soon that will fix the cold boot issue with 3200MHz and hopefully allow for overclock with that as well.
But yea, G.Skill with the C6H is the only board I myself have managed 4x16 and I have seen others manage 3200 with 2x16 and 4x16. Not many but not a common config.
But when buying G.Skill and going 3200, Don't go 3200 C16. It isn't b-die and so you will at best see 2666 if posting at all.
You need b-die if you want any chance of running dual rank memory.
Also anyone buying the C6H, First boot, set the SOC voltage to 0.95v and flash the 1001 or 0032 bios from overclockers.net discussion. It will remove any chance of killing your board through a bug from launch day.
ok i'ma go do my taxes then finally order ryzen, mobo, mem...
A tidbit about RAM timings.
I was having a stable overclock at 3.9 GHz I thought, until I started playing Battlefield 1 that kept crashing at least every other round. So I upped the core voltage, then lowered the clock a lot. Didn't help. Ran a couple of passes of RAM testing that came out clean. I began to doubt the fabric of reality.
Then I remembered a passing note of how Ryzen might more or less not "like" running with odd numbered CAS timings. The RAM was running at 15-15-15-36 as per its 2133 MHz specs, but I ticked them up to 16-16-16-36 instead and it's been stable since.
I guess it's possible that the dedicated RAM testing didn't pick it up because the common RAM tests don't also load the the cores much at the same time, while Battlefield 1 can. I don't know what else, but am happy it's stable for now.
4 Likes
Managed to push my chip to a safe limit
Absolutly stoked.
Won't run this 24/7. Will change it to P state overclock down the track and waiting for a better bios but for now, This is one of the best chips I have seen. Seen one person hit over 1900 at 4.1 with 3200 memory, slightly different timings but jesus ryzen is powerful
Validation
2 Likes
That's why the blend test in Great Internet Mersenne Prime search, aka Prime95, is useful, load RAM, IMC and CPU at the same time. If there is a problem in any part or in the underlying fabrics/buses, it will show.
I haven't seen much testing and conclusions on running Prime95 on Ryzen. Latest version uses FMA however, so it might crash until that microcode update shows up.
I'm running the g.skill at 3200 on a gaming 3 b350m. No issues at all.
Time for a new rumor.
Apparently Asrock mentions the ryzen 3 in a cpu support list. The ryzen 3 1200 is a 3.1 ghz quad-core with a tdp of 65 watt. Looking forward to pricing getting public.
source: https://www.computerbase.de/2017-04/amd-ryzen-3-1200-spezifikationen/.
2MB cache looks realistic for low end part.
It looks like OC3D have discovered something on the x99 platform that is similar to what we have seen with Ryzen gaming performance. The Intel limitation just seems to sit a bit higher up the performance range
Amazingly enough it could only be the Nvidia Drivers and have nothing to do with the Intel CPU
1 Like
The DX12 API will easily drown the software base scheduler on Nvidia's chip, I wouldn't be surprised if the OSD showed 100% usage on all threads which is a tell-tell sign of big CPU overhead. But even on DX11, the CPU usage and overhead could be considerably higher for Nvida cards. Their software based scheduler could potentially give them more drawcalls but at the cost of higher CPU overhead, thus, limitations in performance scaling.
So Biostar have a ITX x370 board coming sometime, but they said only supports DDR4-2666mhz, does anyone know much about which brands are supporting or working with 3600 memory timings? I think it was ASUS with a bios update?
I was being facetious. If it was a CPU processing bottleneck, then none of the single cards would be out performing the Ti SLI - the CPU doesn't know if there is a single or dual GPU demanding frames, it is just having to process more requests for frame data and deal with all the associated interrupts etc. Just like a CPU cannot tell the difference between a cinebench thread and a Tomb raider thread.
He has discovered exactly the same behavior as what is being seen with Ryzen, albeit at a lower level on the Ryzen CPU. His testbench was a 6850K. The intel chiothat does have a monoliticic design with higher bandwidth interconnects to the quad channel memory controller than Ryzen, which has a 32Byte per cycle interconnect to a dual channel memory controller via the data fabric on the AMD.
Ryzen chips don't officially support memory speeds higher than 2667Mhz.
The current bioses from all manufacturers have dividers that will support up to 3200 without having to use refclk to overclock it higher. Currently the only boards that can do refclk adjustments that I am aware of, are the asus crosshair, asrock gaming professional and taichi, and the gigabyte K7
There is a promise from AMD that there will be potential for the vendors to support higher memory dividers in a microcode/core bios release in May
I don't think that is true as it was advertised as supporting higher in their bullet point press release before release. I can't remember which one but it said above 3400 etc.. or something similar
It's a driver fail with Nvidia. Their software scheduler is known for this problem
Nope. official amd support is definitely 2667Mhz. Manufacturers are currently selling baords that have dividers up to 3200Mhz
Could you not provide a reference to AMD stating that? Because companies like Biostar and some others are advertising support for DDR4-3600 and like I said AMD said it in a slide way back before or on launch, but I cannot find on youtube atm as it would take me a few hours of watching it all over again to get the exact slide timeframe.
AMD said that it supports up to 2666 mhz by default. Beyond that is motherboard manufacturers job to make work.