Can we make the Ryzen 7500F perform just as well at half it’s rated power? Let’s find out.
Welcome to Esk Computers, I’m Scott and today on the desk of Esk we’re following up on an idle thought from last week’s build. While testing out that compact PC based around AMD’s Ryzen 5 7500F and 9060XT, the thought occurred to me that while the system was pumping relatively high amounts of power into the CPU, it’s often nowhere near fully utilised. This is because I’m most commonly wanting to play games at higher quality settings and higher resolutions, so while the graphics card is working flat out, the CPU’s cooling its heels waiting to send more data to it.
Now, this calculus should change if you were to put a more powerful graphics card in here, or if you prefer the so-called competitive settings in online shooters like Counter Strike 2, where you’ll be running at lower settings to get the highest framerates possible, which is more CPU dependent.
Also, while the 320-odd watts that this system pulls under full load is already pretty solid, given the price of electricity in the UK its not a bad idea to maximise the efficiency we can get, hopefully without actually hurting the performance very much. So let’s do some investigations, shall we?
Oh, this is going to be quite numbers heavy and a smidgen dry, we should be back poking around hardware next week if that’s more your speed. Still here? Great.
It’s quite easy to head into the BIOS and tweak the settings that the Power Boost Overdrive system uses, essentially telling it how much power it can draw. While the product datasheet for the 7500F says it’s a 65W part, when it’s tuning itself for maximum power draw it can get up to 88W with the standard recommended limits, and if you accept the risk it can go higher, if the motherboard and cooling solution allow.
Now, I’m more concerned here with reducing the power draw than increasing it. From the previous testing, I know this particular example of the 7500F wasn’t the luckiest draw of the silicon lottery, so combined with the compact cooler it’s not going to be pushing any record numbers out. Makes sense, in general if these chips could run at faster speeds they would be binned out as the more expensive 7600x.
WIth all that in mind, let’s see how tweaking the power limits effects the temperatures and the performance in a battery of testing. I’m mostly keeping the curve offset at -15, as per the previous video. I’ve picked a few at the extremes of 4K Ultra and 1080P Low settings to test the boundaries, Cinebench’s Multi core run to see what it’s doing on pure CPU performance, then a bunch at the 1440p High that I think I’d personally be looking to play at.
I’m running these tests at 88W, 75W, 60W, 45W and all the way down to 30W. In an act of mercy I’m not going to talk you in detail through every tab in this spreadsheet, or the seventy four logs of data from the benchmark runs that it summarises, but there’s a few things we can pull out from this.
So let’s get the most surface level, obvious take out of the way first – less power gives less performance. Obviously, although maybe not as much as you’d expect. Cinebench only cares about raw CPU power, so choking it off tanks performance. However, as the CPU lives as part of a system there are limits, and even if we allow the CPU package overall to draw 88W, that doesn’t mean it can feed that into every core and blast them all at infinite speeds, and in fact we get essentially the same performance, power draw, clock speed and temperature results with an 88W limit and a 75W limit. A 60W limit only gives a 4% performance drop, while running 10 degrees cooler. 30W is down at 34% of the PPT limit, but still delivering 61% of the performance, so it’s not a linear drop off.
Depending on the game and the settings you are playing at, 30W might even be enough. Here’s Counter Strike 2 at 4K resolution, Very High Settings. Not how the sweatiest gamers would play of course, but humour me, it’s just to prove the point. As the GPU is running flat out, the CPU is comparatively underutilised, sitting at about 10% utilisation and giving the same performance across the board, at least within margin of error and run to run variance.
It’s at 1080p Low Settings I thought I’d be telling you we’d see a massive difference, but, er, we don’t. It’s overall a bit confusing. There’s a small dip as we go from the 88W limit to the 75W, 60W performance stays the same before recovering at 45W, then finally making sense by dropping again at the 30W limit. This is unexpected, so I re-ran these tests and got the same results. I suppose when you’re above 500 fps it’s pretty academic, and even the 30W result is faster than all bar the fanciest of e-Sports monitors can display. Still weird though.
Cyberpunk 2077 was proving a bit unstable so I’ve dropped the curve optimiser back to a -5 offset, and here at 4K it is much the same story as Counterstrike 2 at 4K. It’s so GPU limited that there’s no difference in overall performance, even at the 30W limit where it’s down on clock speed by a whole gigahertz.
Dropping the settings to 1080p Low, we get closer to the story I’d been expecting to tell. At least, if we exclude the 88W result. There’s no obvious reason I can think of for it to give less performance than the 75W limit result, given that they are drawing the practically the same power and clock speed, barring maybe Windows sticking its oar in. From 75W down, sanity is restored, with drops in power resulting in drops in framerate.
The rest of the games I’d tested were at variants of 1440p High settings, which I think is the sweet spot for the 9060XT. The story I’d expected to tell for these is that at lower power limits, while the overall average fps remains much the same there would be a hit to the 1% and 0.1% low framerates, which would result in distracting instances of stutter. Maybe, at a push you could say that’s what’s happening here in Marvel Rivals, but only at stupidly low power limits.
Pause the video if you want the numbers to back it up, but the benchmarks for Assassin’s Creed Mirage, Black Myth Wukong, Forza Horizion 5, Horizon Zero Dawn, Red Dead Redemption 2 and Shadow of the Tomb Raider all performed essentially the same across all of the power limits tested. Of course, benchmarks aren’t perfect reflections of the games, I’d expect when we figure in the game logic, controlling NPCs and handling physics and all the rest I’d see results more like my expectations of lower performance at lower power limits. But that is not the data I have to present.
So, what have we learned here? Well, I was hoping you wouldn’t ask to be honest. I suppose I have a renewed respect for the 7500F and the Zen 4 architecture. Clearly, it’s highly efficient, even at the lower power limits I’ve been looking at here. For a chip that’s at the lower, cheaper end of the enthusiast segment, and a generation old, it proves that its going to be able to handle graphics cards many times more expensive than it, and I would expect that to hold true into the next generation.
In fact, it’s kinda making me regret pushing the boat out for my personal rig, as it’s driven home that I’m not taking advantage of anything close to the full performance of the 9800X3D, and won’t be able to unless I can steal a RTX 5090 or two from somewhere.
I suppose as with the recent videos on the channel, we’re just backing up the point that while it’s fun to be on the cutting edge of hardware, unless you have some very specific needs, you’ll be able to do everything you need to on lower end or older hardware, probably with overhead to spare. Some solace, as the price of getting into the hobby is heading to the Stratosphere for new parts.
Arguably this has just been two solid days of work just to prove the bleeding obvious, but that’s sometimes just what science must do. Next week ought to be a return to hardware as we look at two handheld PCs, but until then, take care of yourself, and each other.