Monday, February 5th 2024
DDR5-10600 World Record Achieved on Ryzen 7 8700G & ASUS ROG Crosshair X670E GENE
SafeDisk (AKA SoonHo Jeong) has tinkered with an AMD Ryzen 7 8700G APU and ASUS ROG Crosshair X670E GENE motherboard combination—earlier today, the professional overclocker set a validated/verified memory speed of DDR5-10600 world record on Team Red's AM5 platform. This has surpassed a previous result global-leading from late last week—DDR5-10346 was achieved with the same processor, but on a GIGABYTE B650E AORUS Tachyon board. SafeDisk has a distinct advantage in memory overclocking stakes, due to his workplace being the ASUS ROG labs in Taipei, Taiwan—his closest competitor, Benny Lodewijk Nitolo Lase (an Indonesian overclocker) now sits behind him in second place.
SafeDisk's test setup included the aforementioned APU plus motherboard combo, a Thermalright (not ASUS!) AIO, a twin pack of G.Skill Trident Z5 (F5-7800J3646H16G) DIMMs, and an ASUS GeForce GT 730 2 GB Silent graphics card. The ROG Crosshair X670E GENE's default set of VRM heatsinks were removed completely during test conditions. The two memory sticks were set at their default CL36 7800 MT/s speed at 1.45 V—SafeDisk's record breaking DDR5-10600 overclock was achieved with timings of 50-62-62-127-127. Wccftech noted that the: "overclock was able to boot directly into the OS and (SafeDisk) says that the whole OC procedure was very easy."SafeDisk mentions that he did not implement any BCLK adjustments, and brought in some outboard gear—Wccftech clarified this situation: "(it's) important to mention here because apparently, there's a bug within the latest AGESA BIOS firmware that may show a higher DDR5 memory clock but in reality, that's not the case...This was confirmed by using an oscilloscope and adjusting the EX-clock (BCLK) within the BIOS settings. (This) didn't change the memory clock at all so that means that the test result achieved by the GIGABYTE B650E motherboard isn't the real deal whereas the one achieved by SafeDisk is the more factual result. This bug is seen in the AGESA 1.1.0.2b release which was released just a few weeks back and offers optimized support for AMD's Ryzen 8000G APU family."
Sources:
SafeDisk, Wccftech #1, Wccftech #2, Valid X86 Entry
SafeDisk's test setup included the aforementioned APU plus motherboard combo, a Thermalright (not ASUS!) AIO, a twin pack of G.Skill Trident Z5 (F5-7800J3646H16G) DIMMs, and an ASUS GeForce GT 730 2 GB Silent graphics card. The ROG Crosshair X670E GENE's default set of VRM heatsinks were removed completely during test conditions. The two memory sticks were set at their default CL36 7800 MT/s speed at 1.45 V—SafeDisk's record breaking DDR5-10600 overclock was achieved with timings of 50-62-62-127-127. Wccftech noted that the: "overclock was able to boot directly into the OS and (SafeDisk) says that the whole OC procedure was very easy."SafeDisk mentions that he did not implement any BCLK adjustments, and brought in some outboard gear—Wccftech clarified this situation: "(it's) important to mention here because apparently, there's a bug within the latest AGESA BIOS firmware that may show a higher DDR5 memory clock but in reality, that's not the case...This was confirmed by using an oscilloscope and adjusting the EX-clock (BCLK) within the BIOS settings. (This) didn't change the memory clock at all so that means that the test result achieved by the GIGABYTE B650E motherboard isn't the real deal whereas the one achieved by SafeDisk is the more factual result. This bug is seen in the AGESA 1.1.0.2b release which was released just a few weeks back and offers optimized support for AMD's Ryzen 8000G APU family."
20 Comments on DDR5-10600 World Record Achieved on Ryzen 7 8700G & ASUS ROG Crosshair X670E GENE
Although nice too see.
PS2 keyboard and mouse :wtf:
But clock signal is not the only issue, the 128 data wires distort the signals just as much. Lengthy and accurate DDR link training is performed by the IMC in order to synchronise all the bits moving in quick succession, close to physical limit.
I don't know what are they probing, but at that frequency you can't just stick a (differential) probe like that on a production board, moreover without ruining the signal, and above all you won't see anything as clean as displayed.
We should see a DQS pair to have some kind of clock proof but I really would like to see how they did probe it!
Note: a 13 GHz/40GSa oscilloscope should be more than enough to measure a 5 GHz frequency, no ? I'm not talking about jitter analysis.
On close inspection, it appears that some was hitting the oscilloscope with a hammer until it gave in and displayed the desired frequency of 5.25 GHz. Notice the crack above the screen. As the only goal was to set the frequency record, wouldn't there be some chance of going even higher with a single DIMM? There would be a bit less load on the memory controller, and beside that, the two DIMMs probably aren't exactly equal, both can reach 10600 MT/s, but one of them might be able to be pushed farther.
In a different sense, APUs also want fast RAM at all times even if you are using dGPU, anything to help compensate for 16MB L3.
Just had a look see at my Z790 Hero and there ain't no PS2 support. Well check out some of my Apex's later on.
Also, in DDR5 spec, minimum slew rate for both is 2V/ns at 7200MT/s and above, just based on me eyeballing this scope shot, it should be safely within spec even if it's a bit distorted by the slow probe/scope :)