@Armageus Great idea to have a simple table to aid customers in selecting a GPU. I suggest you add two columns for AMD, rasterisation and raytracing, as at the moment it is very misleading for new customers looking to buy a new GPU for up to date gaming, an area where the 6900X does appear to be more on par with a 3070.
-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
*** GPU Hierarchy ***
- Thread starter Armageus
- Start date
I did think about adding * disclaimers next to certain cards and having them appear in e.g. for Raytraced vs Rasterised performance, but it's a slippery slope that I don't want to go down. (Otherwise we end up with columns to specifically benefit low vram cards, cards that are specifically good at certain settings, or certain games etc.)
This should be an "at a glance" table to give a rough idea as to how different models/generations stack up overall. I think the additional tier for the 6090XT was a fair compromise - it implies that it's generally faster than the 3070ti/3070/2080ti etc, a good match compared to the 3080Ti/3080 but perhaps not quite as good overall as the 3090Ti/3090.
Caporegime
That's actually a myth, as that workload is not representative over a wide selection of games. For example, Computerbase just published an article which tested a whole suite of rasterization and ray tracing games and separated the results into rasterization results and ray tracing results. The 3070 is not in the same league as the 6900 XT, even with RT games. Infact at 4K, the 6900 XT came out ahead of the 3080 in RT games. So I think @Armageus moving the 6900 XT up a tier was the right thing to do at the very least. If you took pure averages overall with both, it would be up there alongside the 3090, especially given the majority of the forum admittedly does not care for Ray Tracing at this moment in time.
1080P
1440P
2160P
Caporegime
Yeah, this generation of GPU's is pretty complicated in the sense that they all have strengths and weakness, the only card that sits above all that is the 3090/TI.
Decent article that, however, they are missing a lot of RT games where nvidia does see more of a gain over amd, just to name a few:That's actually a myth, as that workload is not representative over a wide selection of games. For example, Computerbase just published an article which tested a whole suite of rasterization and ray tracing games and separated the results into rasterization results and ray tracing results. The 3070 is not in the same league as the 6900 XT, even with RT games. Infact at 4K, the 6900 XT came out ahead of the 3080 in RT games. So I think @Armageus moving the 6900 XT up a tier was the right thing to do at the very least. If you took pure averages overall with both, it would be up there alongside the 3090, especially given the majority of the forum admittedly does not care for Ray Tracing at this moment in time.
1080P
1440P
2160P
- chernobylite
- dl 2
- cp 2077
- the ascent
- control
- watch dogs legion
- ghostrunner
- bright memory infinite
- sword and fairy 7
And there are a whole bunch of other ones you could add, not to mention if you add the path tracing classics like quake, doom, mario, serious sam.
Their list:
But I think if you look at the big picture where people will also be using FSR or DLSS, the 3070 will still come out quite a bit better for 2 reasons:
- DLSS gains more performance than FSR 1 in RT scenarios as we have seen in a few comparisons now, bang4buck showed this of well, time will tell if FSR 2 can achieve the same perf gain as dlss
- DLSS is available in certain games where FSR is not
Caporegime
Sure if you only test examples using certain titles then it will favour one brand over the other. For example if you only tested a few specific RT titles, my 6900 XT is faster than my 3090. That’s why it’s important to take a balanced view and test over a range of titles. That Computerbase article is quite revealing I think.
Caporegime
Someone should do an average of all RT games currently available.....
RT performance on Ampere IS better than it is on RDNA2, but its also not as bad as many seem to think.
The problem is early on in RT games when these GPU's were reviewed all we had was a few Nvidia sponsored games that were and still are completely broken on the RDNA2 GPU's.
Later RT games its a much closer run thing but yes Ampere still comes out on top.
Adding downsampling tech to the mix really overcomplicates things, you have DLSS1 and RIS where the latter is actually better than DLSS1, DLSS2 and FSR1 where DLSS2 is better but now we also have FSR2 which is also very good.
Then you have the VRam capacity debate, there are some games with HD Texture packs that some Ampere cards just can't run because they don't have enough VRam, that matters too.
Once you start breaking all this down it becomes far too complex, hopefully the next generation is far more clear cut.
Caporegime
Get outta here with that logical rationale!
Well yeah that's true but you can't say it is a "wide" range of games, let alone a balanced/fair representation when they are missing a lot of RT titles of which mostly will perform better on nvidia because of them having more RT effects. Of their list, which they tested, they might as well have not bothered due to how limited the RT is in bf 2042, f1 and then you have 3 titles, which are amd sponsored and pretty limited in RT too (although the RT whilst limited in riftbreaker is great).
Define broken? If you mean the issue with RT reflections not fully working, given that it is happening in several games from the looks of it, I would say that is something more on amds side.
Why are we still talking about dlss 1? It is no longer used now, we're now on dlss 2.4, which has come a very long way since dlss 1, it also shouldn't be compared to FSR 1 or RIS or NIS given how different they are, amd have even said this themselves.... FSR 2 is the competitor to dlss.
Is FSR 2 out now then? Any results from tech press? Other than what amd are providing them with and not just from one game, deathloop.... If not, then the jury is still out on that, it is certainly a lot better looking than fsr 1 though.
PS. the performance difference I was referring to with FSR 1 and dlss in RT workloads:
Don't disagree with the last 2 points, I'm just pointing out the flaw with basing RT perf. on one article with a very limited RT selection of games and the other issues with it when you take into consideration, the strength dlss holds. Although I think we can all agree that RT + DLSS far outweighs the benefit 1 amd sponsored game sees with higher vram and only when using max settings with RT @ 4k with an optional HD texture pack and with no FSR....
Caporegime
@Nexus18 If going on sponsorship, at least four of those tested titles are Nvidia sponsored, BF2042, Doom Eternal, Guardians, and Metro Exodus. So it seems reasonably fair to me on that front with 3 for AMD.
Caporegime
@Nexus18 by broken i mean broken, watch the Jay2Cents 6900XT review.
The point was more in reference to amd sponsored titles being limited in RT too i.e. pretty much all of the titles listed are limited in RT except for metro ee.
19 minute long video and from Dec 2020, tldw summary? Or timepoint for the point you're referring to please.
Caporegime
tldr is early Nvidia RT games would barely even run on RDNA2 with RT enabled, hanging, stuttering, crashing...
Soldato
This observation seems to mirror the debates on game threads months back. Whilst we all want games to work from the get go the sad reality is most are in beta for ages, it would be interesting to get a RT update review on none cherry picked samples to see if the RDNA2 performance is as shocking as its made out.
Caporegime
Right, if AMD are keyed in at the development process of the game they can't optimise their drivers for it, or at least that should come later.
From a hardware and with that code perspective AMD and Nvidia RT are quite different, so one does not transfer to the other, when you have games developed as a showcase for RTX AMD have to rely on Nvidia to optimise for AMD GPU's, which is never going to happen, not least because AMD also don't want Nvidia snooping around AMD's code.
Post RDNA2 the game developer should have the tools they need for optimisation, there in no involvement from the competing hardware vendor.
This is also why i'm not sold on the idea that Control is the yard stick for RT performance between these two GPU's brands.
I think we're straying back into general RT vs non-RT discussion 
I often recommend OCuk, kids birthday etc., due to a combination of the forums being helpful and not wanting to get involved myself. Such people will think they can save a few pounds when looking at the tiers, only to find out the truth later on when trying the latest and greatest e.g. CP2077, Dying Light 2, etc. We are not talking about a small difference here. Of course by that time they have passed the simple returns period thus ending up not happy with both OCuk and, more importantly, myself for the recommendation.
Of course you can debate how important this is as many games still offer very poor raytracing support e.g. Dirt5, SOTTR, etc., but as I've already asked on these forums, if you are willing to use 3DMark for rasterisation performance, why not use it also for raytacing also, especially as the number of more advanced titles grow as we can all agree they will.
Edit - This isn't meant to be RT vs other, but simply to aid/guide customer expectations.
Caporegime
SorryI think we're straying back into general RT vs non-RT discussion
Define cherry picked? Also, there are a good amount of comparisons on youtube showing RT performance comparisons, including from digital foundry when they do their RT in depth analysis videos. Essentially if it is just limited RT effects, amd holds up quite well but when you get various effects or/and higher resolution/more complex RT effects thrown in then amd buckles by quite a bit. Think this article from pcgamershardware (a fan favourite on this forum 
) summed it up quite well:
www.pcgameshardware.de
Although given that in games, turing seems to still match rdna 2, not sure how accurate that really is for "real world usage"
I can't really recall of people mentioning crashing etc., I just recall of cp 2077 not having RT enabled for months on AMDs end due to amd needing to work on it i.e. it wasn't cdprs choice, I could be wrong on that though @Wrinkly
Not sure why you keep only referencing to control, yes control is a title with a fair bit of RT but there are several other better titles for RT.
If you want "fair" then metro ee is actually the best one here even if sponsored by nvidia given that 4a enhanced said themselves, a lot more can be done for amds RT perf. when it is implemented better hence why it RT runs pretty well on RDNA 2 (with furworks off), iirc, even on RDNA 2, it runs better than the rasterization/RT hybrid version, could be wrong though:
True, might be best to continue this in this thread instead:
forums.overclockers.co.uk
) summed it up quite well:
Raytracing-Showcase getestet: Ampere deklassiert Turing und RDNA 2 - ein Blick in die Zukunft?
Der Raytracing-Benchmark ART-Mark streckt selbst eine Geforce RTX 3090 und Radeon RX 6900 XT nieder - Zeit für einen genauen Blick und Benchmarks.
Although given that in games, turing seems to still match rdna 2, not sure how accurate that really is for "real world usage"
Was it confirmed that it was nvidias fault or/and the game developers fault or an issue on amds end?
I can't really recall of people mentioning crashing etc., I just recall of cp 2077 not having RT enabled for months on AMDs end due to amd needing to work on it i.e. it wasn't cdprs choice, I could be wrong on that though @Wrinkly
Not sure why you keep only referencing to control, yes control is a title with a fair bit of RT but there are several other better titles for RT.
If you want "fair" then metro ee is actually the best one here even if sponsored by nvidia given that 4a enhanced said themselves, a lot more can be done for amds RT perf. when it is implemented better hence why it RT runs pretty well on RDNA 2 (with furworks off), iirc, even on RDNA 2, it runs better than the rasterization/RT hybrid version, could be wrong though:
I think we're straying back into general RT vs non-RT discussion
True, might be best to continue this in this thread instead:
Do you care for Ray Tracing "now"?
Raytracing looks very nice but so does Ultra settings. For the last 20 years I've been using medium and high settings. I can see that Ultra looks very nice but I would rather buy a cheaper GPU. It's the same with Raytracing. Yes it's very nice but I'll go without and save the money. I think...
forums.overclockers.co.uk
Caporegime
Define cherry picked? Also, there are a good amount of comparisons on youtube showing RT performance comparisons, including from digital foundry when they do their RT in depth analysis videos. Essentially if it is just limited RT effects, amd holds up quite well but when you get various effects or/and higher resolution/more complex RT effects thrown in then amd buckles by quite a bit. Think this article from pcgamershardware (a fan favourite on this forum
Raytracing-Showcase getestet: Ampere deklassiert Turing und RDNA 2 - ein Blick in die Zukunft?
Der Raytracing-Benchmark ART-Mark streckt selbst eine Geforce RTX 3090 und Radeon RX 6900 XT nieder - Zeit für einen genauen Blick und Benchmarks.
Although given that in games, turing seems to still match rdna 2, not sure how accurate that really is for "real world usage"
Was it confirmed that it was nvidias fault or/and the game developers fault or an issue on amds end?
I can't really recall of people mentioning crashing etc., I just recall of cp 2077 not having RT enabled for months on AMDs end due to amd needing to work on it i.e. it wasn't cdprs choice, I could be wrong on that though @Wrinkly
Not sure why you keep only referencing to control, yes control is a title with a fair bit of RT but there are several other better titles for RT.
If you want "fair" then metro ee is actually the best one here even if sponsored by nvidia given that 4a enhanced said themselves, a lot more can be done for amds RT perf. when it is implemented better hence why it RT runs pretty well on RDNA 2 (with furworks off), iirc, even on RDNA 2, it runs better than the rasterization/RT hybrid version, could be wrong though:
True, might be best to continue this in this thread instead:
Do you care for Ray Tracing "now"?
Raytracing looks very nice but so does Ultra settings. For the last 20 years I've been using medium and high settings. I can see that Ultra looks very nice but I would rather buy a cheaper GPU. It's the same with Raytracing. Yes it's very nice but I'll go without and save the money. I think...
I have taken this to your suggested thread.