Benchmarking CPUs for SC2

Two reasons tests like that are very misleading.

  1. It’s a single data point.
  2. A common issue with CPU performance is over heating. The x3d chips run cooler, meaning people who have issues with overheating will see a performance increase. Without a mechanism to control for the overheating issue in particular, these tests are useless.

Upatree thermally throttled his own CPU on stream, demonstrating he has overheating issues.

So where’s your data point? The person swapped his chip out and had a 50% increase in 1% lows, that’s pretty relevant when it drops down to 50 FPS. You are obviously biased because you bought a non-x3d chip.

Passmark alone has 9047 data points.

Passmark is not SC2. Re-read the title of this thread.

Yes, yes, the CPU performs universally worse across 658 different applications, but SC2 is the one application where the performance trends are inverted and magnified. A simpler explanation is that gamers tend to be bad at building PCs and have overheating issues; these people benefit from chips that use less power because the lower power chips don’t provoke the overheating issue as badly.

This is where my test comes in. Passmark says the 5900x3d is supposed to have slightly better single core performance than the 7950x. My 7950x outperforms UpATree’s by about 6% according to benchmarks from his stream. Combine that with UpATree showing his CPU maxes out at 95c is a very strong indication heat is what is driving this trend, and not the CPU’s actually speed.

It’s not dude, alot of games in particular benefit from 3d cache. We are only arguing about the performance of SC2 here, not 658 other applications.

Then why does the 7950x outperform the 5900x3d in SC2. Answer: heat management.

It’s a newer chip, and i’m not even 100% sure if it does.
If you can cool a 5800x3d sufficiently, that’s different.

Lmao the 5800 uses 105 watts while the 7950x uses 170 watts. It’s definitely a skill issue that a cpu that uses less heat is being thermally throttled.

Yeah, you’re right, however if you can cool those 105 watts, which can be done with a big air cooler, then it reaches max performance.

How cool the chip runs is probably why people get better results with it. They throw in a junk cooler with some BS marketing about “heat pipes” and think it’s good to go. You have to max out the cpu with something like prime95 and see if you hit the max temp. If you do, you are thermally throttled and the cooler isn’t sufficient. Upgrading the CPU won’t accomplish much unless you are buying better cooler and or a lower TDP cpu. I know the CPU tests that youtubers use don’t come close to testing a cpu’s limits when prime95 will make it impossible to even move your mouse. Run openhardware monitor in the background. If you hit 95c or don’t reach the full rated clock rating on all cores, your cooler needs to be upgraded.

Example.. 7950x is rated at a max temp of 95c and a max all-core boost rate of 5.1ghz. All 16 cores are reaching ~5.5 ghz simultaneously and yet the CPU’s max temp is 69c.

1 Like

Yeah, it’s pretty unfortunate that just now they have made a perfect cpu. They fixed the problem with the temps on the 3d chips. I don’t feel like spending $600+ on a chip though.

If you’re loaded, 2025 is looking pretty good for an overkill PC

Zen6 is supposed to be the last AM5 series. That means the CPUs will probably be drop in compatible with existing motherboards. It’s also expected to leap in performance by ~30%. Reports are saying between 8 and 12 billion resistors. The previous increase was 6 → 8 and if extrapolated it means ~11 billion. There is also talk that a 24 or 32 core variant might reach the desktop cpu market. Shifting to 12 billion resistors would extrapolate kinda OK with an increase from 16 → 24 cores.

1 Like

I’m actually a bit envious of the am5 chips, they get double the performance of am4 in certain tasks!
(avx-512)
I’ve seen somebody renting out their 7950x on a cloud server for chess analysis lol. He charges like a dollar.

AVX-512 is good for scientific computing but it’s not terribly great for video games. To do video game logic, you need 3 floating point values which you can easily pack into the AVX-128 standard. AVX-256 is nice because you can do vector operations on doubles. So really the 512 variant benefits matrix multiplication mostly. 4x4 matrix operations can be sped up using the 512 standard. It’s still pretty niche. Most of the operations for a game like SC2 would be adding and subtracting vectors.

My personal favorite for speeding up these computations is to just use integer math. You don’t need accuracy in video game logic, so you just make a class that does the vector and matrix transformations as integers with 2 decimal places of precision and cast it back to a float for rendering.

1 Like

You’re a genius honestly.

BTW, did you watch Nvidia talk at the Las Vegas CES gathering? It was really interesting.

I haven’t heard about that specifically, just some general news about their insane stock price as well as how the 5000 series gpus are supposed to be significantly better than the 4000 series. I went with the 4070 ti as soon as the prices were close to msrp. There is a huge jump in cost to the 4080’s and 4090’s that just didn’t justify the increase in computational power, so you are paying mostly for the memory. The 4070ti was the only sane option, and even at that a double 4060 ti would’ve been a good option. The problem is doubling the water block cost which doesn’t make sense with graphics cards that are good because they are a good value. I was going to do a custom GPU water block but ended up buying one because work got busy and whatnot. The only game I really play is sc2 so the only reason to buy this thing was to speed up physics simulations.

If you are looking to get ahead of the tech news, Asianometry is pretty decent. He’s been talking about the challenges of the stacked transistor method for awhile (the one used to make the 3D v-cache found in the x3d chips).

I believe it was this video that mentioned it. Basically the way they layer together the slices of the chips is by making them super smooth then combining them in a vacuum and they magically stick together with a permanent bond.

But his channel isn’t for everyone. It’s super technical. My father was an electrical engineer at philips so I am a weirdo and am into these kinds of things.

1 Like

Yeah, the talk was more interesting to me when they got to the robots :stuck_out_tongue: . It’s worth a watch.

Sorry, it was this one where he mentions the 3d transistors.