Issue with PHY bus caused by PhenomII x4 975BE mild OC
Issue with PHY bus caused by PhenomII x4 975BE mild OC
Asus AM3+ motherboard
ASUS M5A78L-M Plus/USB3 DDR3 HDMI DVI USB 3.0 760G MicroATX Board
CPU Phenom II x4 975BE
8gb AMD radeon DDR3 1866
Radeon HD3650 PCIE 512mb (OC 750/400) – I know it’s not much, but I’ll stick with base video until I can upgrade to a R9 390X
The issue is that the system bus underclocks to 199.5 MHz X18 multi, CPU runs at 3590 MHz, and all spread spectrum functions are turned off. The BIOS lists the HT link as 200.
A possible fix would be to raise the system bus to 201 MHz still at 18x, but my HS/F isn’t up for any major overclocking, though it performs adequately at stock settings.
This actually leads to a minor overclock since the CPU reports the system bus at 200.5 MHz X18, with a clock speed of 3609. The difference between 3590 and 3609 is negligible, but I still want to maximize performance.
I wonder if this small overclock would really help lower temperatures, and whether the board can handle a slightly higher reference clock (around 0.5 MHz) for longer use. (Once I get a better heatsink and fully optimize it, I’ll consider boosting the multi to 20.)
I’m not new to this. I’ve overclocked AMD chips since the 486DX era, starting with graphics cards from the Radeon X850. I boosted my 3650 by adding RAM sockets, cooling it down, and edited the BIOS with the Radeon Editor to run at 750/400 and 1.3V instead of 1.25V – still a weak card, but noticeable gains.
The only concern I have is whether this board could be affected by such a minor overclock on the 2000 HT link and PCIe bus, which I know can cause problems (though I doubt it with just a slight increase). That’s why I usually just raise the multi, but this is the first board I’ve seen that runs the main reference clock above spec by default without enabling spread spectrum.
The ASUS AM3+ board offers limited information. Some boards are excellent, others are outstanding, while some are quite poor. For optimal overclocking, a 990FX board would be the best choice. A 760G chipset board isn't recommended.
I thought the 760G chipset wouldn't be the best option given the price, so I set it to default speed by underclocking the ref clock. It was just what I could afford. I added some quality copper heatsinks to the VRM to help with the 125W chip and haven’t managed to downclock it beyond a constant .5mhz reduction. My setup has good airflow, and I’ve installed a fan on the board, so I’m hoping this will work for at least 3.8 (x20 multi).
It looks like you're already heading in the right direction. You could improve stability slightly more. The 975s were later versions, so the silicon was a bit more developed than the 965. The main issue with overheating will likely limit how much performance you can achieve. Once you add heat sinks, you might be able to test the touch sensitivity. It's definitely hot, but if it causes burns, you're probably too close.
Also, try overclocking the CPU-NB, which controls the memory for Phenom II. The higher you can push that setting, the better your latency will be. Aiming for 2.8-3.2GHz in air is a solid target. Improving the HT Link won't make much difference unless you upgrade to a stronger graphics card, but even then, your core frequency might still be the main bottleneck.
Thanks
The heatsinks I got on the VRM's seem to be working well. they get warm but not burning hot, and have a case exhaust fan drawing air across them as it evacuates heat from the case (Mid range Delta 92mm 102cfm 25mm thickness)
Power supply is 750w EVGA 80+ Platinum
That's a good tip about upping the cpu-NB clock. I'd read mixed information, some saying push it to 3.2 and others saying anything over 2.6 was unnecessary unless running DDR3 2400.
Speaking of graphics, I planned to get a r9 390x as I know that card has windows 7 drivers, but does anyone know if the newer RX series like a RX 580 has drivers for windows 7 64 bit
On the older C2 revision chips 2.6GHz NB was near the limit. Your chip should be a later C3 revision. E0 even reached higher speeds. You can verify your setup in CPU-Z. You'll notice improvements in L3 and memory latency when using AIDA64. Gaming performance is highly sensitive to latency, so you'll see better frame rates as the pipeline becomes more efficient.
Yeah mine is a C3 revision.
A new heatsink came in today. Running at idle around 35°C with a stock voltage of 3.8, CPU/IO at 2.8 (a slight boost on CPU-IO). (AS Ceramique2 requires about 25 hours for curing, so temperatures might improve. No stability problems in Windows yet, and only reached 54°C while stressing with OCCT for an hour. The real test will be overnight stress once the thermal paste reaches its peak efficiency.)
I sanded the heatsink and CPU IHS to 3000 grit.
At this stage, 3.8 seems like a realistic limit for airflow. Especially with the 760G chipset considering the temperatures.
Maybe we can push it to 3.9 or even 4.0 by using an old technique from the Athlon XP era and swapping the heatsink fan for a 120mm Delta. (The current heatsink uses a standard 120mm fan, so a 120x25mm Delta should fit perfectly for 152cfm.)
Noise levels aren’t a concern since I already have a 92mm delta exhaust in addition to the PSU exhaust, a 40mm delta on the chipset, and a 120mm intake.
Other peripherals include a 750GB drive with a PATA connection to a SATA adapter, a WD Black SATA 6GBPS 1TB SSD (main system disk), and three WD Velociraptor 500GB HDDs (data storage).
overnight burn using prime95 pass @3.8 with 2.8 cpu/nb
Current max load temperature is 52°C
The VRMs are still performing well during peak loads and should support higher MHz without issues.
The copper heatsinks I attached using the silver thermal adhesive are getting warm but not too hot. Both the exhaust fan and PSU fan move air over them, though I could easily add a separate fan to blow directly on them. A 60mm size fits snugly between the case back and the large CPU cooler. I’ll need to shut down and use JB-weld to fasten it, but the VRM should still work for a while longer.