Marvel's Spider-Man PC - Multi-GPU Nvidia vs AMD Face-Off - PS5/PS4 Pro vs GeForce + Much More!

foreign [Music] s with Horizon zero Dawns Sony is coming into its own with its PC ports of its first party PlayStation titles and Marvel Spider-Man remastered is especially intriguing what you're looking at here is what you might call the footy maxed out experience running at 4K resolution quality settings push Beyond PlayStation 5 Hardware accelerated Ray tracing taken to the next level with increased detail massively improved texture work and full resolution and whether you're gaming on high-end AMD or nvidiographics technology it all plays out beautifully yes one of the biggest takeaways I have from my time with this game is that for all the extra horsepower and silicon Nvidia Hardware dedicates to Hardware raid facing it's actually very performant on AMD 2 optimized settings there are reasons for that but the game's console lineage definitely plays A Part so this video is going to dive deeper into graphics performance on the PC version of this game yes we shall be comparing the high-end 4K experience running on the likes of the RTX 3080 and the RX 6800 XT but we're also going to be taking a look at the current mainstream offerings from AMD and Nvidia RTX 3060 and RX 6600 although in a later case there I really wish I hadn't finally I wanted to revisit the veteran gpus that still somehow dominate the steam Hardware survey I refer of course to nvidia's pascal-based GTX 1066 gigabyte and its AMD equivalent the RX 580 8 gigabyte the Polaris Champion that is still amd's most popular GPU based on valve's metrics we'll also find some time to go back to a couple of specifically chosen gpus from Days Gone By nvidia's GTX 750 Ti was our graphics card of choice for matching PlayStation 4 performance way back in the day while the I 9270x is effectively the PC rendition of the PlayStation 4 GPU but you get some extra frequency a couple more compute units but only half the async compute pipelines but this was the basis for the GPU that dominated an entire console generation so what does it take to run Marvel's Spider-Man effectively on either of these video cards console comparisons well of course you can compare PC to PlayStation 5 But ultimately anybody looking for precision benchmarks between PlayStation 5 and any given graphics card isn't really going to find illumination here as unlike other PlayStation ports there is no original Graphics mode here and plenty of the settings have no exact match between PC and PS5 whether you're talking about the hair quality settings variations in level of detail or more but what you do get in the Nexus Port is plenty of console features that absolutely can deliver excellent results on PC a proper 30fps cap for Less capable hardware for example along with access to one of the best non-ai console upscalers on the market insomniac's excellent temporal injection technique and of course FSR 2.0 is a more PC focused alternative to that so I thought it might be fun to deploy those on the GTX 1060 for a quote unquote Dynamic 4K stacked up against PS4 Pro running the original version of this game first up though there's an elephant in the room to talk about with Marvel Spider-Man remastered and that's the CPU requirement if you're looking to use console equivalent Hardware accelerated raid facing settings or even the heavier very high preset you could find yourself held back by the processor in your PC to make something very clear though no RT you'll be absolutely fine on a range of CPUs but illustrating the issue here's a look at the core i5 11400f paired with 16 gigabytes of 3600 megahertz ddr4 I'm using an RTX 1390 here at 1080p to remove Graphics as the limitation for mates are all CPU limited and this is on a CPU that is significantly faster than the mainstream favorite the ryzen 5 3600 which in turn is a lot faster than Zen one CPUs in the ryzen 1000 and 2000 lines maximum RT takes us beneath 60fps 54 to 55 but it can drop lower in fast reversal as the workload you're seeing here is augmented with environmental streaming and decompression on top of processing raid facing we'll be looking at that shortly dial back RT to hire with object range 8 Alex's optimized settings and a closer match for PlayStation 5 and well you've gained 10 frames per second or 18 percent of performance turn off RT completely though and you're looking at an 85 perf boost over very high and a still impressive 54 over RT optimized there's a reason I've chosen the 11400f here in most scenarios this CPU is more than enough for a gaming page and it's faster than the Zen 2 based processor in PlayStation 5. even with its core deficit but for this use case it's not good enough of course the 12400f is its more recent successor and while I don't have that chip to hand my colleague will Judd has and here's how it performs paired with the same 3600 megahertz ddr4 in addition to the RT CPU cost traversal adds to the processor burden ultimately I think we're looking at a 12 600 class processor or better to get even optimized RT running consistently above 60 so bear that in mind when considering the GPU tests to follow and yeah those frame times when CPU limited they're pretty gross right all of these tests are running on optimized settings as defined by Alex in the DF tech review and there's a link to those settings in the video description below essentially we're using the high setting as a baseline tweaking downwards where the extra costs in GPU performance does not deliver the requisite increase in quality ultimately what you're looking at is as close to ps5's Performance RT mode as possible given the fact that settings can't be precisely matched you can push texture quality up if you have a circuit of vram available but one area where I'd like to make a tweak to optimize settings is in hair quality ps5's performance RT mode is some kind of hybrid of medium and high by the looks of it but really I'd be considering dropping this down to medium or even low for lower display resolutions and to illustrate check this out same scene the same settings using RTX 3060 at 1440p as the example moving from high to medium gains you anything from around 22 to 35 percent of of extra performance depending on how many characters are on screen at any given point however going forward the benchmarks you'll see are on Alex's optimized settings which do still have the high setting active just bear in mind that performance can go much higher if there's any of that geometrically rich hair in the scene and I'm kicking off with GTX 1060 and RX 580 analysis at 1080p higher end 6 gig and 8 gig models respectively and when I'm saying we benched Mark well we actually did much more than that we played through the entire Kingpin sequence for a spot of cutscene action web swinging Placid effects of each internal sequence then we moved on to the Fisk construction Mission yes home to the original puddle gate debacle remember that one this Mission takes place in the open world and piles on the mission mechanics plus it features lavish set pieces in total about 18 minutes of content edited and compared head to head on top of that uh the info sequence and the initial piece of web swinging were extracted for a locked like for like Benchmark this is rather interesting actually as it shows the Perils of benchmarking from a single scene alone it can never really be representative of the whole experience of Any Given game the sequence here shows 1060 and 580 offering exactly the same performance on Aggregates but the lead between them switching depending on content however many of the other test scenes show a big RX 580 Advantage particularly evident in cutscenes the performance Delta can change radically between the two cars it's not really enough to say that one is X percent faster than the other but the 580s lows are usually higher than the 1060s while the Nvidia card also has more of a tendency to randomly stutter and yeah although we're using High texture settings on both cards here the extra vram on 580 means that you should be able to move up to very high textures this scene highlights why the hair settings should be dialed down to medium not that you'll really be able to tell the difference at 1080p we're looking at frame rates dropping into the 30s here medium or low would be the way forward here for sure in combination with a frame rate cap and the use of dynamic resolution scaling the need for consistency can be felt across the whole clip really especially when it comes to web swinging the erratic frame times would suggest a CPU limitation but that's clearly not the case when you're using a 12 900k paired with high-end ddr5 and especially when the problem reduces and can even Disappear With higher end gpus what you're seeing here is the GPU utilization problem Alex highlighted in the tech review and that Nexus is hopefully still looking into which seems to specifically impact gpus in the lower range the way I'd suggest running this game is to set a frame rate limit 60fps is good and then use DRS Dynamic resolution scaling by setting a performance limit that's below the area where there are these issues you end up with a smoother ride overall both these cars should definitely be able to give you a reasonably good 60 FPS experience but setting that limit definitely Smooths the experience before we move on to the next round of gpus let's have a quick diversion GTX 1060 and 580 were great cars back in the day because fundamentally in combination with a decent CPU you could double the performance of PS4 with the same visual feature set or better but the pro slash enhanced consoles effectively used a similar level of performance to boost resolution instead so Marvel Spider-Man on PS4 Pro typically rendered at around 1584p using insomniac's excellent temporal injection technique to give a really good 4K presentation at 30 FPS and I had to wonder using the 30fps cap the game provides could The Humble G TX 1060 provide a similarly good 4K experience to PS4 Pro obviously content isn't the same because this is the remastered version up against the original so yes this is more for the lulls but even so here's how that checks out essentially setting the 1060 to Output 4K then letting DRS take over to deliver as close to 30 FPS as we can get while the image quality looks pretty good pixel count suggests that while PS4 Pro sticks fairly tightly to 1584p to 1060 is far far more variable and often with a lower resolution however insomniac's igti temporal injection really goes the extra mile in making the presentation stand up on the Nvidia side the image quality doesn't really bother me as much as the performance which isn't quite as solid at the 30 FPS Mark as the PlayStation 4 pro version despite the frame rate limit and the dynamic resolution scalar in effect we don't quite have the same locked 30 FPS as enjoyed on the console version of the game so at this lower frame rate I'm still wondering whether we're seeing that issue that we were talking about earlier with lower spec gpus so yeah a fun exercise ultimately but 1060 is at its best native 1080p here with Dynamic resolution scaling and a frame rate cap engaged as Ray facing is off the table the CPU burden isn't a problem for most modern processors either next up are at a glance look at the second tier of GPU I chose to look at here we're comparing the RTX 3060 which you can consider as similar to a 2070 but with 12 gigs of vram versus the 2070s eight and that's compared against the RX 6600 from AMD rdna2 raid facing capable 8 gigs of frame buffer memory both of these cards had a 330 dollar MSRP but I do feel in retrospect I made a mistake in using the 6600 here it's way off Pacer optimized settings at 1440p resolution to the point where turning RT off all together on the 6600 only gives it something like a 12 and a half percent performance advantage over the 3060 with rt the on in this limited excerpt of the game at least while both Nvidia and AMD tout these cards for higher end 1080p gaming 1440p is fine actually but far more so with Nvidia which sees its performance lead vary drastically according to content you can see that here in just one scene when the camera flips to a full view of Spider-Man in costume the 6600 just loses a massive chunk of performance the Radeon card also seems to have profound issues with bandwidth intensive scenes with Spider-Man standing in front of the fire here once again the 6600 takes a nosedive in performance with web swinging in particular we're almost looking at two entirely different classes of card here even if the footage is not synchronized the Gap is just gigantic it's not just here either looking at The Swinging Pursuit section in the Fisk construction Mission the Gap there is just as profound this seems to be a case of both the rasterization performance Advantage for NVIDIA being augmented by a ray tracing Advantage too while the 3060 gives you a decent enough turnout 14 OTP I'd be more inclined to pushing down the 6600 to a lower resolution or alternatively keeping 1440p but disabling RT even so there's still evidence of erratic frame times if not quite as pronounced as in the RX 580 and 1060 stats it seems that some kind of artificial ceiling is still there but the more GPU overhead you have the less pronounced the frame time stutter during City swinging so yeah pleasantly surprised by the 3060s turnout here so here's a bit of a bonus comparison comparing the RTX card at 1440p with rt versus the 1060 and RX 580 at 1080p without dislike for like sequence actually gives the 30 60 an 11 lead over those two generally which actually represents a pretty impressive upgrade path bearing in mind that you're moving from 1080p to 1440p and you're adding Hardware accelerated Ray tracing on top of that and that's without factoring in dlss at all another feature that the 1060 won't have 3060 TI always struck me as the sweet spot for the Ampere architecture in terms of price versus performance but even without factoring in dlss which I think you should use you're clearly getting a big chunk of extra performance and features up against the GPU that is still the number one in the steam Hardware survey the 6600 as I said I'm regretting including it to be honest I'd have liked to have tested the 6600 XT instead because once we look at the higher end 6800 XT the comparison against RTX 3080 is a lot more favorable this is a game where AMD RT is actually fully viable as you would hope from a game that originally targeted the rdna to RT Hardware gonna use the 3060 here to tackle the PC versus console comparison this is the 3060 on as close to PS5 settings as possible which pushes a little higher than optimized but but I'm using 4K Dynamic resolution scaling with dlss doing the upscaling pass benchmarks are kind of pointless here to be honest as you are not comparing like with like but on a more General level both games are piling on the effects in terms of rasterization and raid facing then using reconstruction to present nicely on a 4K screen we know the DRS range on PS5 is between 1080P and 1440p depending on content with PC we don't have that visibility short of doing extremely tricky pixel counts when it comes to dlss but the ballpark seems to be similar if not a touch higher bearing in mind how much of the Benchmark sequence was running in 50 to 70 FPS range at a locked 1440p so you can't really consider this as a benchmark more of an at a glance confirmation that the 3060 can produce a PS5 like experience even if you are using different settings and an entirely different upscaling formula I was surprised to see that this is actually a viable 4K experience delivered by the 3060 here and with that second console based diversion out of the way it's time to look at the high-end juggernauts well not quite the best of the best but if it were my money I was spending I'd choose a 3080 over a 30 90 or indeed a 3080 TI whilst I'd take the 6800 XT over the 6900 XT simply in terms of bang for the buck and it's either on optimized settings with both gpus locked to a native 4K 3840 by 2160 we get a fairly closely matched experience far closer in performance terms when usually comparing Nvidia and AMD we can only theorize hearing that the kind of RT implementation here is cleverly matched to the capabilities of the Radeon implementation it optimized settings here the high RT preset reduces geometry in Reflections significantly while the nature of the reflections is more mirror-like with fewer dull less reflective surface is where Reflections are actually more challenging to calculate in fact optimized settings High RT with object range 8 is very close to where PS5 sits by comparison how close are these two cards really an isolated Benchmark shows a six percent lead for NVIDIA over AMD in both Arty and non-rt modes optimized settings which is intriguing across the whole gamut of tests across the introduction cut scene the entire first mission and the Fisk construction Mission later on in the game everything plays out on exactly the same lines with very similar performance differentials from start to finish and this is somewhat at odds with a benchmark that basically ensures that RT Reflections dominate the entire camera view here the 10 gig RTX 3080 effectively renders very high RT reflections with extended range at the same frame rate as the RX 6800xt on optimized settings and yes this is at 4K resolution no Dynamic resolution scaling no upscaling at all in fact I guess this is a worst case scenario for AMD here as not only is RT dominating The View it's also a shot taken from Times Square One of the most geometrically rich areas in the game the more geography the higher impact the RT calculations ultimately though in gameplay at least on either one of these cars you can engage vsync or cap to 60 in another way if you have a vrr display ramp up everything to Max and I mean everything and then allow Dynamic resolution scaling to take the strain and it seems to work rather well remember that the upscaling here isn't basic as such it's augmented either via dlss FSR 2.0 or insomniac's own console orientated igti and it looks phenomenal but having scaled all the way up what about going below minimum specs we know the game can do it firstly because steam deck is below that minimum desktop spec and secondly because developer Nexus spoke to Alex about this in a tech interview we've run on Euro gamer.net and as you can see here I'm not really benchmarking as such I'm using a variant of my steam deck settings with the game's 30fps cap enabled and with resolution at a dynamic 900p there are problems clearly most noticeable in cutscenes actually but the game kind of holds up pretty well I think to be fair these are the Unicorn four gigabyte versions of these relatively ancient cards most 270x's and 750 tis shipped with two gigs of RAM not four but so we don't have to run textures at ultra low settings or anything like that it actually looks alright and yeah actually I bought this 4 gig 270x originally to compare this GPU that's very like PlayStation 4s to its PC counterpart for death stranding but funnily enough I couldn't even get the game to boot because the 270x doesn't feature the required dx12 feature set compatibility but here no problem whatsoever so kudos to Nexus for that an interesting way to end this video but that's where I will be finishing up for now I'm planning to release most of the data in unedited form on the DF supporter program but in the meantime I hope you enjoyed this extended look at Spider-Man PC performance it's been fun doing this but there's still the sense that Nexus continues to work away in the background on improving this game probably a good idea with the port of Miles Morales coming hopefully later this year but for now please do like subscribe share if you enjoyed the content and ring the bell for those notionally instant notifications and yes please do consider the DF supports program for high quality downloads a ton of bonus material Early Access opportunities and much much more but that's all for me for now thanks for watching and supporting digital Foundry I guess this is the plan come on Pete you got this you got this you got this you got this [Music] please don't screw this up foreign

Channel: Digital Foundry

Views: 105,492

Keywords:

Id: deUAqRMSXgM

Length: 22min 12sec (1332 seconds)

Published: Tue Aug 30 2022