jay.archard and
MeanMachine41, and for the benefit of our other forum members and visiting readers, as my esteemed colleague,
Darkbreeze, has already pointed out, Prime95 Small FFTs is a steady-state workload which is ideally suited for testing thermal performance. Without AVX, Small FFTs is a 100% TDP workload. With respect to TDP, which is a separate and controversial topic, TDP is typically only achieved by "Flagship" processors. Lesser variants with lower frequencies or with certain features disabled can't reach or exceed TDP unless overclocked.
However, to be very specific and clear, when any AVX test selections are enabled for mainstream processors, Prime95 Small FFTs will impose up to a 130% workload, which is a
brutal and completely unrealistic workload. Such extreme workloads can adversely affect stability and
severely overload your CPU.
Earlier processors, as well as low-end variants don't have any AVX Instruction Sets, so they're not affected. Mainstream processors which are more contemporary have either one or two AVX Instruction Sets, while high-end variants have a 3rd AVX Instruction Set.
AVX Instruction Sets have been implemented over the past 9 years beginning with the original AVX, followed by AVX2, then more recently, AVX-512. Each implementation is progressively more demanding on a processor's ALU (Arithmetic Logic Unit) or FPU (Floating Point Unit), which are the number crunching elements within the processor's silicon Die. This is why AVX Instruction Sets increase power consumption and Core temperatures.
As the FX-8350 has the original AVX Instruction Set, it is thermally affected. Since
jay.archard uses his rig for rendering, which uses AVX, Core temperatures will accordingly be higher. Comparatively, with the exception of a few utilities and specialized computational apps, the AVX code in real-world apps (rendering / transcoding) and recent games is less intensive than Prime95's Small FFTs workload
without AVX.
There is much misinformation and confusion concerning "stress" testing, because the vast majority of users do not provide the details of their test conditions. Consequently, numbers get flung around like gorilla poo in a cage. Without knowing critical baseline values such as ambient temperature and the precise workload, thermal performance numbers are reduced to apples and oranges thermal fruit salad in a blender.
"Full load" is a popular but
non-specific user term which could mean
anything, so it's important to be
very specific. “Stress” tests vary widely and can be characterized into two categories;
stability tests which are
fluctuating workloads, and
thermal tests which are
steady workloads.
Utilities that don't
overload or
underload your processor will give you a valid thermal baseline.
Darkbreeze provided the scale from my Guide on the previous page, but here it is again; a comparison of utilities grouped as
thermal and
stability tests according to % of TDP, averaged across six processor Generations at stock settings rounded to the nearest 5%:
Although these tests range from
70% to 130% TDP workload, Windows Task Manager interprets every test as
100% CPU Utilization, which is processor resource activity,
not actual workload.
Core temperatures respond directly to Power consumption (Watts), which is driven by workload. Prime95 Small FFTs
(AVX disabled) provides the correct workload for testing thermal performance. If Core temperatures don't exceed
80°C, your CPU should run the most demanding
real-world workloads without overheating.
As is the case for
jay.archard, if you use an overclocked rig for rendering / transcoding, then you may need to dial back your frequency and Vcore to keep Core temperatures in check. In more recent years, motherboards provide AVX Offset adjustments (downclock) in BIOS. When rendering, -3 (300 MHz) or more may be needed to limit Core temperatures. High-end processors with AVX-512 Instruction Sets may require a deeper AVX-512 Offset.
I know of Computronics from Swaziland. A private company with vested interest and opinions.
MeanMachine41, although I've been on the African continent, I've never been to Swaziland, nor am I affiliated with any company or individual using the name "Compu
troni
cs". My username here at Tom's Hardware has instead always been "Compu
Troni
x" since I became a Member in 2006. You've been here since 2014 and we've encountered one another on many occasions, so this should come as no surprise to you.
Moreover, the %TDP values in the above scale were found and verified by running each utility (and many less popular utilities not included in the scale) for numerous iterations using a consistent methodology. A variety of utilities and test equipment were used to meticulously monitor both numerical values and graphical waveform signatures for system performance parameters. This includes ambient temperature, case intake temperature, case internal temperature, case exhaust temperatures, Core temperatures and Vcore, as well as CPU power consumption (Watts) and total computer power consumption at the wall socket.
The results were averaged across six processor generations at stock settings rounded to the nearest 5%. My research and testing has been ongoing since 2006. The scale is intended to illustrate and give perspective as to how dramatically different "stress test" can be; a fact of which many lesser experienced or knowledgeable users and first-time overclockers are often unaware.
The problem with AIDA64 is that it has 4 CPU related stress test selections (CPU, FPU, Cache, Memory) which have
15 possible combinations that yield
15 different workloads and 15 possible sets of Core temperatures. Unless users provide screenshots, they seldom state exactly which test(s) they ran. The individual FPU test is about 115% TDP workload, the CPU/FPU combination is about 90%, all 4 tests combined is about 80% and the individual CPU test is only about 70%. All other AIDA64 test selections are fluctuating workloads which are suitable for
stability testing, but not for
thermal testing.
As
jay.archard ran all 4 test simultaneously which is a fluctuating workload at about 80%, this is why his Core temperatures were higher running Prime95 Small FFTs without AVX due to its steady-state 100% workload, as well as when he was rendering. You are correct that Prime95 can impose torturous workloads. Nevertheless, as
Darkbreeze pointed out, when the correct test and options are selected, Prime95 Small FFTs without AVX / AVX2 / AVX-512 is the correct utility and test for accurately determining thermal performance.
If a user only games or never runs heavy CPU demanding workloads such as for rendering / transcoding, then CPU-Z>Bench>Stress CPU is better suited for testing thermal performance due to its steady-state workload at about 80%, which approximates the heaviest gaming workloads. So if you don't need to be AVX stable, then there's no point in configuring a lower overclock to accommodate it, or test for it.
CT