Home
Artists
Search Recent Random
Posts
Search DMs Popular Hash Lookup Tags Random
Importer
Import FAQ
Register
Partychan Telegram ThePornDude
Home Artists Posts Import Register

Patreon Exclusive: First Thoughts on PS5 CFI-1100 Testing (Patreon)

Published:
2021-09-08 12:19:12
Imported:
2023-01

Content

Update: Hi everyone, so the full review will finally be going live tomorrow and I think you'll like it. I do now have a working explanation for the performance differences here, and essentially, if I remove the beta firmware from contention, the new PS5 runs just like an old PS5. In the video I pointed to Resi as negating this hypothesis as it runs slower on the new firmware. However, I did multi-run testing and found that some titles - including Resi - can vary from one run to the next, which explains that. Some other titles such as Godfall show no difference at all on the new firmware.

The beta firmware is now the official firmware, however. So I updated both of my PS5s and found that any performance difference is now gone. So it seems that possibly in some cases, the new firmware may add a percentage point or two to in-game performance. At least now we have some answers but maybe I should dig a little more into his.

Original Story: Hey everyone, it's Rich here and last night I recorded a short video talking about initial tests on the PlayStation 5 CFI-1100 model. We've revealed that we're collaborating with Gamers Nexus on this content, which means it's going to take some time, so why not share an early update with a first look at our data - and it presents a mystery.

The idea of a console is that every unit behaves in the same way, but comparing the CFI-1100 to my launch 1000 model (both disc versions), I found that rather than getting identical performance, in several unlocked frame-rate tests I found my launch model to have a small, but measurable 2-3% performance uplift compared to the new machine. I thought this may be because I was running beta firmware on the unit, so I also tested a PS5 CFI-1000 Digital unit on the latest retail firmware, which presented results pretty much identical to the CFI-1100 - as things should be, right?

New firmware, better performance? Nice! Except... in Resident Evil 5, the CFI-1100 seemed to be a touch faster than my launch model in identical content. So I'm wondering... are we looking at per-run variances with dynamic changes in scene content causing the difference? If so, I'd expect perf differentials of 1% or lower. Or maybe the boost clock algorithm acts in slightly different ways on a per-run basis.

As the video also confirms, any variance IS NOT DOWN TO TEMPERATURE. Let's just make that clear. The variance may affect all machines on a per-run basis to a certain extent, as the thermal testing demonstrates. But maybe the Boost Clock does make very slight changes on a per-run basis? I guess the only way to be sure - as I say in the video - is to perform the Control, DMC5 and Resident Evil Village tests multiple times on a Series X. Multiple runs on a specific PS5 may also be a good data point just to see if we get different results from the same machine.

Ultimately, a 2-3% performance difference isn't a big deal except that I'm not sure we should be seeing it at all, so I may well dig deeper into the results and would welcome any ideas. Beyond that, I hope our Discord community appreciate the brutal test conditions of the IKEA Besta media cabinet! 

On another note, this video is not scripted, it's more conversational and just represents my thought processes as opposed to final testing. But the formula for making this kind of content is pretty simple, the editing job is not arduous and I'm wondering if the approach could work on main YouTube videos occasionally too.

Downloads here: https://www.digitalfoundry.net/2021-09-08-patreon-exclusive-playstation-5-cfi-1100-the-story-so-far


Files

Patreon Exclusive: PS5 CFI-1100 - The Story So Far

An update video on PlayStation 5 CFI-1100 testing, exclusive to our Patreon supporters.

Comments

Ben Fisher

Can someone carry out some statistical analysis on the dataset to see whether more data is required to validate the 2-3% variance? Or whether the difference is swallowed up in confidence limits. Will help eliminate the run to run dynamic theory.

digitalfoundry
>>67421524

I can say with some certainty from years of PC benchmarking that you should not be seeing anything more than 1% run-to-run. I agree that the data is incomplete at the moment. I think that Series X testing on the multi-plat games with multiple runs should be compared with the same tests being run on a PS5 with multiple runs. As we have different data coming from the 1100 on the same titles, I suspect that it's just something PS5 does. I'm very curious to see if Xbox Series X does the same in the same conditions.

Ben Fisher

Fair play. Thanks for the update. Yeah that would be interesting

Anonymous

Could it still be the beta f/w somehow? Not knowing what they changed it could be possible that one title is negatively affected...

Anonymous

You said there's naturally some run-to-run variance. Can you quantify that so we have some context for this 2%-3% difference being 'very irregular'...?

Anonymous

Re: charbroiled 1100 performance advantage over the oven-ready 1100. Maybe the charbroiled 1100 has cached some stuff to give it that tiny, tiny advantage?

digitalfoundry
>>67437811

I've just re-run the Devil May Cry and Control tests on an Xbox Series X - no quick resume, reboot after each test cycle. The max variance is 1.5% and the lines are essentially overlaid over one another. EDIT: That's comparing three test runs btw.

digitalfoundry
>>67438850

Unlikely as the 'cool' test was done at my workstation next to an open window, while the charbroiled test was in my lounge. Hours apart with a power down cycle. The charbroiled perf is more similar to the CFI-1000 model with the firmware beta.

Anonymous

The first power part is interesting. I believe on PC GPU if you have a fix core/mem clock, the total GPU board power will slightly increase when the temperature increases. Can we assume these two PS5 have very similar frequency at the same activity? It is mentioned in The Road to PS5 video at 36:45, that PS5 boost frequency determined by the render workload rather than temperature so the frequency is deterministic and repeatable. The later cabinet vs normal part, I wonder if that is b/c fan speed is different. If for example, in cabinet some temperature detection points like VRM/VRAM/board are higher than normal and leads to higher fan speed, then the overall temperature is lower which slightly increases the total board power budget if my first power<->temperature POV is reasonable. Pure speculation though, thank you for the exclusive video here and can't wait for Steve temperature data :)

Anonymous

Will be interesting to see if over a bigger sample size of PS5s the performance differential is consistently in favor of older hardware or if it just comes down to lottery. Appreciate the early peek!