My retail sample parts were provided by AMD, the mother board I used first was an MSI X470 Gaming M7 AC. A nice quality RGB board with a mature bios for this launch, everything seems to work and I did try the OC knob with minimal enthusiasm. The memory from G.Skill, Sniper series and I am sure Samsung B Die based on speed, timing and density. And the CPU is a retail sample complete box with heat sink. Nothing is special about these parts, nor are they engineering or qualification samples.
http://www.xtremesystems.org/fugger/msi.jpg
I spent a week playing with the 2700X at close to 5Ghz on the triple stage cascade at -110c, some single threaded benchmarks I ran at 5150Mhz. The most important detail that I can pinpoint is the IMC is fixed running sub zero temperatures but running high density memory at high speed is not possible, this includes running 4x memory modules will also limit memory frequency overclocking.
Using Samsung B Die and both 8GB and 16GB modules I was able to reach with stability the following speeds without touching voltage or timing with AXMP settings of 3400CL16 (8GB) and XMP 3200C16 (16GB)
2x8GB 3733Mhz
4x8Gb 3333Mhz
2x16GB 3000Mhz
4x16GB 2666Mhz
Previous generations of Ryzen have this density issue but the memory overclocking issue was much worse as it prohibited many from going cold on Ryzen without extensive research. I know the struggle on the early AGESA code and those days I hope are long gone. I test timing at each step and a wrong step will have drastic impact on performance even though common sense tells you it should be better. Locking down timing seems to be best and adjusting your secondary's will produce nice gains and reduce latencies for benching. For those who do not go through that ritual with just set AXMP and seek the highest memory frequency works excellent too, a small bump in memory voltage to 1.5V should get you to 3733Mhz.
Lapping the CPU will gain you a few Mhz, my 2700X needed a leveling of the surface. I lapped my 2700X to 2000 grit and noted a pattern, an "arch" appeared as I removed copper/nickel layer and exposed the copper, I expected a different pattern, this arch pattern shows a high side and a low side.
AMD SenseiMI Precision Boost works great and will give you more performance if your cooling is better so having the best cooling will yield better results from a stock cooler. The stock Wraith cooler is excellent and I am impressed with the stock cooler vs an AIO performance. SenseMI will stop the fan and bump it when needed or until a load is on the CPU from a program similar to what the GPU's do now. Setting up on the Wraith a lot of my results were better than locking all cores to 4200x, locking to 4300x I ran the AIO to manage 4.3Ghz all cores stable for X265 to beat the auto settings. My suggestion to leave it at auto and give it the best cooling you can, SensiMi seems to do a good job in delivering max performance that the cooling will allow. I even tested SenseMI with the cascade and it took advantage of the extra cooling on its own yet stayed in the confines of the preset Mhz limits.
Ryzen master has evolved some, it now shows the best core on each CCX but I am curious on how it determines this best core. I found out through running SuperPi 32M on each core that this "best core" was far from the best core. I turn off SMT and reboot, push max multi to 51.50 and run on each core to find an exact small variation in each core, I re ran several times to find the real best core. Ryzen Master was wrong 4/4 for best core. Back to the cores not being exact. I was watching start times on each core and moving between them to actually see they had a difference, core 0 was supposed to be the second best core in the system and was the worst at -110c and could not complete 32M at 5150Mz and had a start time of 4.88x, core 1 has a start time of 4.72x and was unmarked as a best core, core 2 marked as best core also had a start time of 4.88+, core 3 was good, CCX2 core 4 was good, Core 5 was the best core in the system and was also unmarked, I ran on pi core 4 (Core 0 CCX 2 with downcore set to 1+1 in BIOS).
http://hwbot.org/submission/3838265_...min_0sec_267ms
You can see the LN2 rigs below me at hundreds of Mhz faster, core selection and locking memory down for the win. Maybe the best core is for air?
MSI has a really good board, well built and nicely packaged. A competitive price of $250 is fair the technology and features provided. For overclocking this board worked excellent, recovered quickly, and ran as expected. CPU and memory overclocking went great and on par with capability. I had hoped to see DDR4 past 4Ghz in dual channel but this seems to be the limits of the IMC inside the CPU and not the fault of the board.
The 2700X is a strong move in the right direction but more can be done to improve the process and reduce leakage, one of the things I checked first is the vcore on auto and set the Mhz to 4Ghz, reboot back into bios and see what value it gives the CPU. This value will rise when it gets hotter and drop when it gets cooler exactly in temp window and seems to stop at 0c exactly in BIOS. The MSI BIOS will stop at reading temps at 0c but the Ryzen Master will read to -59c in windows on the 2700X, the 2600X will read to -49c in windows.
http://hwbot.org/hardware/processor/ryzen_7_2700x/
I ran almost all of these results on Windows 7 but Windows 7 has been sabotaged by Microsoft to annoy you to death in the form of a dialog box informing you that your CPU is us supported exactly when you do not want it to pop up and effect the benchmark score. One way is to run with explorer off the other is to do the registry hack, I did see mention of removing the update but that did not work for me. I do feel 3D and desktop experience is better in Windows 10.
The 2700X will be moved into production environment and serve as a Raster Image Processor where I hope it can from the CPU extensions.
The 2600X is now on the cascade with the Gigabyte X470 Gaming 7 Wifi, I should have scores pushed next week The 2600X can run everything at 5Ghz but seems to have the exact thermal limit preventing me from going past 5.2Ghz.
Reply With Quote
Bookmarks