News Halo Infinite Benchmarked: Master Chief Eats Tons of VRAM

[Alvar "Miles" Udell]

I think the real conclusion is that 313 has some work to do in optimizing the game for PC. Having to use a $600+ GPU to obtain 60fps minimums at 1920x1080 without sacrificing detail levels is insane.

 

[JarredWaltonGPU]

Yeah, ReBar is one difference, then the PC specs, especially the CPU and the most important one, the section of the game tested and how it was tested.

You say your test is valid, but for me is not. Why? Because I care about the worst case scenario, the most demanding part of the game as it is when I'm playing with the enemies and all that jazz, not a controlled sterile environment. And it shows the difference...

The truth is HUB made the best benchmark (without ReBar, Steve said a new test is coming with it ON and he did not use the MS store version, which is the worst - he specifically says that), because it's the closest representative of what we, all gamers, get on those demanding parts of the campaign. Look at his video, how he did the run, set on easy, running thru all the enemies on that section, without dying. There is something to learn from that.

I don't care if I have 100fps of 120 fps in low demanding zones, but I do care if I get 45fps or 60fps in the worst demanding parts of the game. It's the difference of having a smooth gameplay and not.

I'm sorry, but not all test are equal and not all testers are either, some are better and some need to get better. And you can do better, so that in the future your tests are better too and your work is not in vain. It's just the cold hard truth.

For the record, arguing from a point of ignorance and suggesting I "failed" as a hardware expert is an awesome way to get added to my ignore list. I just proved that the testing was valid to within 3% of the results with/without enemies, and then V doubled down and said it's still "not valid." Fine. You're wrong, but fine. I'm done trying to argue with someone that has their head in the sand. The only way to prove the Microsoft Store version is inferior would be to run all the tests on all versions of the game and show the results, carefully documenting how testing was conducted. That just doubled the amount of testing, and I strongly doubt it will matter.

 

[VforV]

For the record, arguing from a point of ignorance and suggesting I "failed" as a hardware expert is an awesome way to get added to my ignore list. I just proved that the testing was valid to within 3% of the results with/without enemies, and then V doubled down and said it's still "not valid." Fine. You're wrong, but fine. I'm done trying to argue with someone that has their head in the sand. The only way to prove the Microsoft Store version is inferior would be to run all the tests on all versions of the game and show the results, carefully documenting how testing was conducted. That just doubled the amount of testing, and I strongly doubt it will matter.

Please tell me how caring about the lowest FPS drops in the most demanding sections of the game makes me ignorant?

And how does showing a low demanding zone with no enemies is actually important or even matter? Does anyone play on an empty map without enemies and only in a section of the game where is low stress on the GPU/CPU? Is that how people normally play games?

HUB showed the former, you showed the latter. So what matters more, 100+ fps in low demanding zones or those drops under 60 fps in high demanding zones?

I watch these benchmarks and tests to see which GPUs have playable performance over 60 fps and which parts of the game are the more demanding ones, not to see test of looking at the sky getting 300 fps (exaggeration, but that's your point you're making trying to defend yourself).

How do you not understand basic logic here? That the most demanding zones and the lowest fps are the relevant ones, since those are the the parts with "issues", not the 100+ fps ones...

 

[Deleted member 14196]

Haha. I have it ignored too. Anyone with a best answers score of one who is throwing vitriol should really keep out of it.

 

[husker]

You say your test is valid, but for me is not. Why? Because I care about the worst case scenario, the most demanding part of the game as it is when I'm playing with the enemies and all that jazz, not a controlled sterile environment. And it shows the difference...

The single most important factor in determining if a test is valid is repeatability. This is why in any scientific test repeatability is the single most important factor. Also, validity is not a subjective property - it applies equally to all observers. Stating the test is not valid for you doesn't make sense. Are you saying you live in an alternate universe in which physical principles exbibit themselves in some kind of alternate way? It seems to me that this reviewer always stresses repeatability over everything else. Now, to your point, you may not like this particular test, maybe it didn't test the things you wanted, but that has nothing to do with the validity of the test. It is a perfectly valid test for the things that it claims to be testing for. Sometimes you have to do one or both of the following: 1) Take the information as given and extrapolate to other conditions. 2) Pick a different review with different test parameters, hopefully those that more closely match the conditions you are looking for.

 

[JarredWaltonGPU]

For the record, arguing from a point of ignorance and suggesting I "failed" as a hardware expert is an awesome way to get added to my ignore list. I just proved that the testing was valid to within 3% of the results with/without enemies, and then V doubled down and said it's still "not valid." Fine. You're wrong, but fine. I'm done trying to argue with someone that has their head in the sand. The only way to prove the Microsoft Store version is inferior would be to run all the tests on all versions of the game and show the results, carefully documenting how testing was conducted. That just doubled the amount of testing, and I strongly doubt it will matter.

Actually, I have an update on this. I got the Steam version of the game. I reran the same three tests as before: "Orig" is my original shorter benchmark with an "empty" level but running through the most demanding area in that level. "Clean" is the same thing but a longer run where I traverse 90 seconds of the level. "Battle" is the same level but populated with Banished enemies with me fighting my way through them — which very much is not nearly as consistent (and I deleted the run where I died on the MS Store version, leaving just two runs). I've updated the above table as well to include the "Orig" results. Here you go, Steam version vs. Microsoft Store version.



So, there you have it. It's only on one card, but the "Battle" testing is again far more variable, particularly on the 99th and 99.9th percentile stuff. As far as the claim that the Steam version of the game performs better? It's complete garbage. On the "Clear" tests, average fps is within 0.5%, with MS Store version leading slightly. The lower percentile numbers do favor the Steam version, but those are always more variable — just look at the two runs on each test and you'll see anywhere from a 2% to a 15% gap on 99th percentile fps! There's even one case (my "Orig" test sequence, which was more repeatable) where the 99.9th percentile strongly favored the MS Store. But that's precisely the problem with looking at anything beyond 99th percentile: a single frame or two that takes longer to render can radically skew the results. The "97p Avg Min" incidentally is the average fps of the bottom 3% of all frametimes, which tends to vary a bit less than a strict 99th percentile.

I rest my case on how the testing was done. For the current release, there's every indication the Steam and MS Store versions perform the same, with the usual 1~2% margin of error you'll see in live gameplay testing.

 

[hotaru.hino]

Please tell me how caring about the lowest FPS drops in the most demanding sections of the game makes me ignorant?

And how does showing a low demanding zone with no enemies is actually important or even matter? Does anyone play on an empty map without enemies and only in a section of the game where is low stress on the GPU/CPU? Is that how people normally play games?

HUB showed the former, you showed the latter. So what matters more, 100+ fps in low demanding zones or those drops under 60 fps in high demanding zones?

I watch these benchmarks and tests to see which GPUs have playable performance over 60 fps and which parts of the game are the more demanding ones, not to see test of looking at the sky getting 300 fps (exaggeration, but that's your point you're making trying to defend yourself).

How do you not understand basic logic here? That the most demanding zones and the lowest fps are the relevant ones, since those are the the parts with "issues", not the 100+ fps ones...

Then we should discard all tests that don't represent any actual game play that someone would actually experience.

3DMark? Get rid of it.
Unigine demos? Get rid of those too
In-game benchmarks? Most of those aren't running a typical in-game scenario, so get rid of them.

Oh and we should make sure to have the OS be well used as well. Because nobody* runs their games on a band new, fresh install of their OS with no other apps other than what's strictly necessary for the game to run.

 

[drtweak]

I got a 960 4Gb (Glad i got the 4GB model 5 years ago XD) but yea it runs on low (Auto detected) but you know the graphics still look great. Even on low @1200P

 

[VforV]

The single most important factor in determining if a test is valid is repeatability. This is why in any scientific test repeatability is the single most important factor. Also, validity is not a subjective property - it applies equally to all observers. Stating the test is not valid for you doesn't make sense. Are you saying you live in an alternate universe in which physical principles exbibit themselves in some kind of alternate way? It seems to me that this reviewer always stresses repeatability over everything else. Now, to your point, you may not like this particular test, maybe it didn't test the things you wanted, but that has nothing to do with the validity of the test. It is a perfectly valid test for the things that it claims to be testing for. Sometimes you have to do one or both of the following: 1) Take the information as given and extrapolate to other conditions. 2) Pick a different review with different test parameters, hopefully those that more closely match the conditions you are looking for.

Alright, I admit using validity maybe was a poor choice of word(s), but then again English is not my native language and I used it as "pertinent", "relevant" (so maybe there's the issue). In which case my bad.

So in that regard, no, this test STILL is not relevant or important or matters, because when I play the game I will get worse FPS in those demanding zones and if this were the only test or all tests were done exactly the same with the same results and we would not have the HUB test where it shows a worse case scenario, well if we did not have that and only this one I would have been perplexed and **** off when playing the game and not getting the high FPS this test shows.

That's why this is not valid for me, but if the wording is wrong, then I'll change valid for relevant or important. I still stand by everything that I said, except this word which apparently is the issue...

I like it how my wording is the issue now, instead of how really not very useful this test is... the upside-down world.

Then we should discard all tests that don't represent any actual game play that someone would actually experience.

3DMark? Get rid of it.
Unigine demos? Get rid of those too
In-game benchmarks? Most of those aren't running a typical in-game scenario, so get rid of them.

Oh and we should make sure to have the OS be well used as well. Because nobody* runs their games on a band new, fresh install of their OS with no other apps other than what's strictly necessary for the game to run.

You're exaggerating and talking nonsense too.

1. Unigine demos and In-game benchmarks have fixed settings, so everyone using that preset gets the same test out of that software, only their hardware differs.

2. Those are tests only, this is real gamplay, or it should be as close as possible to real gameplay (which is exactly the issue, that it's not in this case). Two different things. I want to see how it is when I play (before I play/buy) not just a test done in a void, having very little in common with actual gameplay.

3. You don play Unigine, you do play this game (and other games). Playing is different than running a fixed settings test.

Benchmarks of games that do not have a built-in benchmark is harder yes, but that does not mean the one who does them should not strive for as close to real gameplay as possible... As hard as it is, HUB proved it's doable, so it's not like no one did it...

Like I said, I still stand by all that I said.

 

[hotaru.hino]

2. Those are tests only, this is real gamplay, or it should be as close as possible to real gameplay (which is exactly the issue, that it's not in this case). Two different things. I want to see how it is when I play (before I play/buy) not just a test done in a void, having very little in common with actual gameplay.

Having a test with a clear map provides a baseline of which to compare the so-called "actual gameplay" performance you want. Being able to compare the two numbers may be able to tell you something about how the game is implemented. If the game gets 300 FPS in a clear map but immediately tanks to 60 FPS with a handful of stuff going on, then that's an immediate red flag the game isn't well implemented. If all you knew it was 60 FPS, then that doesn't really provide much information about how the game is implemented.

I would like to think articles like this aren't just seeing how hardware performs on the game, but a critique on how the game was implemented.

But I mean, if you don't like the content presented here, you're free to just not visit the site anymore.

 

[VforV]

Having a test with a clear map provides a baseline of which to compare the so-called "actual gameplay" performance you want. Being able to compare the two numbers may be able to tell you something about how the game is implemented. If the game gets 300 FPS in a clear map but immediately tanks to 60 FPS with a handful of stuff going on, then that's an immediate red flag the game isn't well implemented. If all you knew it was 60 FPS, then that doesn't really provide much information about how the game is implemented.

I would like to think articles like this aren't just seeing how hardware performs on the game, but a critique on how the game was implemented.

But I mean, if you don't like the content presented here, you're free to just not visit the site anymore.

Here is the gist of it:

1. I mainly trust HUB and GN and very very sporadically, some other sites and other YT channels make a good piece of content from time to time, the rest of the time is mostly entertainment for me, from all the other places on the net regarding tech. I don't have something against this site in particular... It's about that none compare to HUB and GN. And to be even clearer even thou Steve and Tim (fom HUB) both do written posts also on TechSpot (they have a collaboration), no I don't care about TS either, they are in the same boat and category as this site all the other sites, as in just entertainment for me.

2. The way I am wired is whenever and wherever I see an issue of any kind (in the fields that I'm interested and watch, of course, in this case IT and tech), I intervene/reply/post and make my stance known about that issue.

So that's what I did here and I do everywhere, except on some sites that there is no more reason to try do do so, like Videocardz and WCCFtech, because those places are lunacy asylums and reason and reasoning has long left there...

Like I did here, I may been harsh, but all I did was point an issue (wording it wrong admittedly), that this test was less relevant that what HUB (and another channel, which tested the same section as HUB) did and I suggested in the future these tests should be done better, because it's possible to do better as proven by HUB and the other channel, for testing games without built-in benchmarks, even if it's harder to do.
That was my only intention and why I did it. I consider this constructive criticism, but some got offended so... that is that.

What I don't understand is why the results are inversed? nvidia is stomping AMD in HUB's test while in your case AMD is above nvidia...

I got my answer for my own question.

The reason Radeon GPUs do better in low demanding scenes vs nvidia, but get beaten by nvidia in higher demanding scenes it's because of (at least) these two reasons:

1. Infinity Cache on RDNA2 works better with lower demanding scene because it gets more cahce hits with fewer variables, while in the more demanding scenes the Infinity cache is not so potent, but the grunt power of Ampere is better... something like that.
2. The game still needs more patches and optimizations.

This game dev and engineering is explaining it better here @1:16:14 minute mark:

View: https://youtu.be/DKBHsD4UUOo?t=4573