Thursday, December 02, 2021

Heldscalla system cooling, heat, benchmarks

This article will benchmark my current Heldscalla system against my previous VALIS computer. I will be comparing specifications and getting results from both Passmark and Cinebench R23. I'll explain that I turned off the overclocking built into the motherboard settings. And I will describe the physical air flow and choice of CPU cooler, as a prelude to measuring operating temperatures and noise levels. 

As usual, my goal is an article that doesn't avoid discussion of necessary technical considerations, while remaining comprehensible to readers not steeped in jargon. YMMV.

I previously outlined how my VALIS system built over eleven years, using the same case and PSU. Until recently I was using this lowly Core i3 chip and a graphics card that was three generations old. Though this computer would generally be considered woefully underpowered, it could play every game I tried, and allowed me to edit 4K video clips to an HD timeline. It wasn't fast, but it worked. 

That's a point I really need to stress. There's an entire industry telling you that the latest, greatest piece of hardware is required for your life to be complete. But for the vast majority of people, this is simply untrue. Computers were already plenty fast a decade ago. 

Nonetheless my post-production and media editing requirements have finally outstripped VALIS. And so I built a new system (full details). Here's a comparison of the three key components.

CPUIntel Core i3-7350K
60 W, 4.2 GHz
Ryzen 9 5900X
105 W, 4.8 GHz
RTX 3060 Ti
64 GB

Benchmark tests

I benchmarked these two systems using standard tests: Passmark and Cinebench R23. Here are the results.

Passmark VALIS Heldscallafactor
overall 4,389 8,237 1.88
CPU 5,051 35,901 7.11
2D graphics925 992 1.07
3D graphics6,247 21,350 3.42
memory 2,875 3,160 1.10
disk 15,68728,399 1.81
percentile 43 95
Cinebench R23VALIS Heldscallafactor
multicore 2,475 17,942 7.25
single core 1,112 1,376 1.24

This makes clear a few facts. First, my old CPU had very good single core performance: that's why I bought it! A Core-i3 is not necessarily slower than an i5 or i7.... despite its branding. It depends a lot on the task. At the time of purchase, most of my key software applications did not make use of multiple cores.

However, it's hard to deny that the overall CPU performance has gone up by a huge factor (the two tests agree). This is because there are twelve cores instead of four, and each core is both faster and more efficient. 

2D graphics performance remains roughly the same, despite moving up three generations in graphics cards. Hence the basic Windows UI won't see any speed improvements... not that any are necessary. However, demanding 3D graphics will be much faster. So will any calculations that are allocated to the GPU instead of the CPU. This will help with VFX in DaVinci Resolve, one of the key factors in my upgrade. (I will dive into this in a future article.)

I am surprised that the memory upgrade does not result in a more significant benchmark change. This demonstrates that memory speed is not the bottleneck in these tasks. It could be a greater factor in CPU-bound VFX, where video frames need to be copied repeatedly. We shall see. 

Though I had two SSDs in the previous system, the two new solid state drives are much faster models. Indeed, one is potentially faster than the bus it is on.

Though it's not a factor in the benchmark tests, the 8 GB of dedicated VRAM will enable video editing at higher resolutions, compared with the 4 GB on the older card.

Performance tuning

There are many ways to tune performance. Some builders will overclock their components, pushing memory timings and CPU speeds higher than factory preferences. I am not a believer in this practice. At best you get negligible increases that won't be noticed in daily tasks. 

I discovered that my motherboard (ASUS ProArt X570 CREATOR WiFi) came overclocked out of the box, so that users can be impressed by its performance. That makes sense, since the main consumers of high-end components are gamers who wish to flex. And why not eek out a bit more horsepower? Three reasons: increased power consumption, heat generation, and noise. 

My approach was more conservative. Venturing into the incredibly-complex BIOS, I disabled three features: "Core Performance Boost", "PBO Fmax Enhancer", and "Precision Boost Overdrive". By the way, I did this before the benchmarks were run. Since the standings nonetheless place Heldscalla in the top 5% of all measured computers, I think I am justified in leaving those settings off. At a later date might experiment with underclocking, in order to reduce power consumption further.

Heat and noise

Heat and noise are related factors. As components get hotter, fans must spin faster to move air. This produces noise from the air turbulence itself, the fan motors, and supplementary sources including the bearings. 

For VALIS, I'd prioritised silence by building inside a dampened case. I made that decision because the components were low power and hence wouldn't let off much heat. But Heldscalla operates in a different power regime. The 4864 GPU cores and 12 CPU cores are going to produce substantial heat when under load. Hence I built inside a mesh case, allowing air easier ingress and egress. Of course this also allows sound to travel out from the components to my waiting ears.

As a composer and audio engineer, I am particularly sensitive to sound. My studio is relatively quiet, more so than most domestic environments, even though it has not been treated. My goal is to have the computer running nearby, without any sound being picked up by a vocal microphone. I wish to create voice-overs and teach remotely without distraction. 

Fractal Design equips the Meshify 2 Compact with one rear fan and two 140mm front fans. I added two Noctua fans, the most efficient and quietest in the industry. The Scythe Fuma CPU cooler has two 140mm fans and the GPU has three of its own. 

That's ten fans in total. Wow! But the more fans and the larger their diameter, the less work each has to perform. This keeps rotation speeds low. 

The air flow is as follows. The two intake fans on the mesh front bring air directly to the CPU cooler, which works to push air to the mesh rear of the case for evacuation from the rear fan. Meanwhile, the GPU fans bring air from the bottom of the case (mesh) up to where the CPU cooler sits. That's why I have two out-take fans on the mesh upper panel. They immediately help suck hot air from the CPU but also assist in the GPU cooling. 

In case you're wondering, the two side panels are solid. Indeed, they are the only solid panels of the Meshify 2. (I will review this case in a later article.)

Because the case is relatively compact, there's a lot of air moving even at low fan speeds. The fans are tightly coupled to the air volumes they need to move, so I imagine that there's little turbulence. 

CPU cooler

I should make special mention of the Scythe Fuma 2. This CPU cooler has a lower profile than others, with a special cut-out on the underside, so that it can rise above your RAM sticks. The asymmetrical build is easier to fit inside any case, unlike much larger Noctua coolers. 

The package includes everything you need: a mounting kit for both major platforms, two fans with clips, thermal paste, and even a heavy-duty screwdriver! This is magnetic and has an especially long barrel. You'll need this to reach through the mounting screws from the top of the assembly. 

This screwdriver is so nice that I used it for the entire build.

I found it much easier to attach the two fans before installation, even though this leaves less room to attach the assemblage itself to the motherboard. This is because the fan clips are particularly fiddly. I am not sure who came up with this system, but it's a chore for the geometrically challenged (e.g. me). We can't blame Scythe, because it's use is widespread. 

One of the fans acts as an intake, while the other sits between the two fin groupings, in the middle of the cooler. Hence a continuous air flow is guaranteed. Apparently Scythe uses a system where the fans rotate in opposite directions, increasing the static pressure. 

I don't normally rave about CPU coolers, but this one is a winner. Literally so, as Gamer's Nexus just granted it their cooler of the year award. (Yes, some people really are that geeky.)

Some of you might be asking why I didn't use a water-cooling system. First, there's no need. Second, these don't have the longevity of simple fans. Water cooling is very popular with online sites, for the simple reason that they look flash. None of those builders use a system for long before tearing it down and starting over. They never test a water system over the long term. 

VALIS lasted me a decade. By that time, water pumps will fail or become noisy. And a water pump failure is far more catastrophic than a fan that stops working. Not that I have ever had a fan stop working.

Testing temperatures

This can be tested by measuring temperature. I used two software packages: Ryzen Master monitors the CPU while GeForce Experience writes a log of GPU activity. 

The working temperature when performing normal desktop tasks (writing and web browsing while listening to music) is 40°C on the CPU and 39°C on the GPU, with GPU utilisation at only 1%. The fans are nice and quiet; you need to get within 1 meter to hear a constant whirr. Constant sounds are far less annoying than those that cycle or have a rhythm. Indeed, the hard drive is far more obnoxious in this respect (as I wrote in a previous article).

The fans can be tuned in five groups (GPU, CPU, front case, top case, rear case). With a temperature ceiling of 90°C, I have headroom to reduce the speeds, but I've not looked into this yet. The computer is not loud enough for me to care. 

Now for a stress test. I ran Titanfall 2 at WQHD resolution (2560 x 1440) with most settings maxed out (only film grain was turned off). In-game the CPU ramps up to 92% optimisation and GPU peaks at 97%. The temperatures are 52°C CPU (with some rare spikes into the higher fifties) and a maximum of 51°C on the GPU. 

A rise of 12 degrees from baseline to full load is clearly a great result. I am pleased that the CPU and GPU have consistent temperatures, indicating no serious hot spots. The top-rear and the rear of the case get appreciably warm, but not hot. This confirms that the air flow is working as designed. If these areas were not warm, then this would indicate that heat was being trapped inside. 

And the sound? It rises from a whirr to a purr. It's not at all like those wind tunnel effects I've heard with other computers. I'm sure that an open microphone on my voice would pick up no extra background noise but I will test this later. 

Success, then. And with no special tweaking (motherboard BIOS settings aside), only a little thought to the fan placement. 


No comments:

Post a Comment