Intel 285K for VR usage - or stick to AMD X3D CPUs?

[lBlackMambal]

Hello all,

first of all, to avoid comparing my existing CPU to the 285K - where my CPU definitely will loose - the intention behind the topic title was to cover the X3Ds chips in general, which also includes the new 9800X3D but also the upcoming 9900X3D.

Ok, back to my setup, which lists as following

AMD Ryzen 7 7800X3D
64 GB RAM
RTX 4090

With this setup and running DCS (MT) with a Quest 3 via Virtual Desktop (God Mode) I reach solid 85 fps. When I activate the DCS service info, it says I'm CPU bound.

Looking at other postings here and on Reddit it reads that for VR the CPU has a not neglectable impact. Which brings me to sites like cpu-monkey and direct comparisons between different CPUs, where the 285K definitely outperforms all existing X3D variants (7800, 9800) for the single core performance. E.g.
285K vs. 7800X3D -> Cinebench R23 (2380 vs. 1811)
285K vs. 9800X3D -> Cinebench R23 (2380 vs. 2086)
285K vs. 9900X3D -> Cinebench R23 (2380 vs. 2231) Expected, not confirmed yet

What's your opinion on the 285K vs. the AMD lineup, specifically when it's about playing DCS in VR? I would highly appreciate your comments.

Thank you very much in advance.

 

 

Edited December 8, 2024 by lBlackMambal

[Dragon1-1]

The X3D series is unique. They do not have the raw single core clock speed of other CPUs, it's good, but not amazing on those chips. What they do have is copious amount of cache, which allows cutting down the time the CPU spends talking to the RAM. This has an outsized impact on VR because much of what CPU does in graphics heavy applications, such as rendering for a 4K headset, is slinging texture data back and forth. 

I don't know how those specific CPUs compare in actual in-sim performance, but I do know that tests focused on clock speeds only may give a skewed impression. Pushing for highest possible clock rates and core numbers makes sense if you do something that requires raw speed (such as heavy duty maths), but if the most demanding thing you do on that rig is DCS, then the X3D Ryzens might be the better choice.

[Aapje]

Why are you looking at Cinebench when you actually want game performance? The reviews are clear. The 285k is beaten by both the 7800X3D and 9800X3D in games.

If you are willing to spend a lot for relatively little gain, then I would wait for the stock situation of the 9800X3D to improve and to then buy that. Don't get the 9900X3D, because it's worse to have two 6 core chiplets, where one has the X3D-cache, rather than one single 8-core chiplet with X3D. You'll have more latency and there will be fewer cores on the X3D-chiplet.

There is a chance that the 9950X3D will be slightly better than the 9800X3D, but probably not.

[lBlackMambal]

vor 2 Stunden schrieb Aapje:

Why are you looking at Cinebench when you actually want game performance? The reviews are clear. The 285k is beaten by both the 7800X3D and 9800X3D in games.

If you are willing to spend a lot for relatively little gain, then I would wait for the stock situation of the 9800X3D to improve and to then buy that. Don't get the 9900X3D, because it's worse to have two 6 core chiplets, where one has the X3D-cache, rather than one single 8-core chiplet with X3D. You'll have more latency and there will be fewer cores on the X3D-chiplet.

There is a chance that the 9950X3D will be slightly better than the 9800X3D, but probably not.

Thanks for the reply to both of you. Dragon1-1 what makes me think that specifically for DCS it's more than shifting back and forth textures is the fact that - at least to my understanding - DCS is not the typical game but more the simulation resulting in processing even more "non-gaming" stuff. Taken the dynamics of the ego aircraft and how it behaves with the environment + position/action data of allied/enemy forces doesn't that consume a good amount of raw computational power - i.e. isn't this the heavy duty maths you were referring to?

Aapje, I took Cinebench assuming it probably reflects in a good way, how those kind of calculation demands can be handled best.
Yes, the 9950X3D is a hot topic for me too, hoping that it's slightly better than the 9800X3D and that we won't have the same situation as back in 2023 between 7800X3D/7950X3D - but I guess meanwhile selecting the correct cores isn't any longer a problem

[Aapje]

Cinebench is more relevant to what are commonly called 'productivity applications,' and not so much to gaming.

In games, the code that matters is nearly always the gameplay loop, which are the calculations that have to be done for every frame that is shown on the screen. This in large part uses the same data again and again, which is why the X3D-cache works so well for gaming, since the data that keeps getting reused is kept in this cache and so the CPU doesn't need to constantly retrieve it from RAM. The gameplay loop also pretty much never benefits from a huge amount of cores, since to get the calculation done quickly enough to get a high frame rate, you can't afford the latency costs of having a huge amount of cores work together.

In gaming, the CPU also can't afford to work ahead too much, since each new frame needs to adapt to the user input, the behavior of enemies, etc. This means that lots of cores can't be used effectively, to work ahead.

In 'productivity applications,' in many situations you have a lot of data that is not constantly reused for each calculation, so the X3D-cache doesn't help much. The data still needs to be retrieved from RAM, since mostly new data is needed for each calculation that is not present in the cache. Furthermore, these applications don't have to react to user input that much and can take a lot more time to do their work. For example, in Cinebench, making a single frame can take seconds. A game with a frame rate less than 1 FPS would be completely unplayable, so Cinebench is not representative of how games do things.

So for productivity applications, you tend to want: lots of cores and lots of bandwidth to the RAM. The speed of the cores, a very big cache and low latency to the RAM are less important.

For games, you tend to want: very fast cores, a big cache and low latency to RAM (although with a big enough cache this becomes much less important). The number of cores, and the bandwidth to the RAM are less important.

So essentially, the strengths of the 9800X3D match the demands of gaming very well, but not so much the demands for productivity applications. The 285k is more tailored for productivity applications. If you want to have strong performance for both kinds of applications, then the 9950X3D is almost certainly going to be the best, although that is achieved by more or less turning it into a 9800X3D for games (where the chiplet without X3D is not used for the game).

Edited December 9, 2024 by Aapje

[Hiob]

12 hours ago, lBlackMambal said:

Hello all,

first of all, to avoid comparing my existing CPU to the 285K - where my CPU definitely will loose - the intention behind the topic title was to cover the X3Ds chips in general, which also includes the new 9800X3D but also the upcoming 9900X3D.

Ok, back to my setup, which lists as following

AMD Ryzen 7 7800X3D
64 GB RAM
RTX 4090

With this setup and running DCS (MT) with a Quest 3 via Virtual Desktop (God Mode) I reach solid 85 fps. When I activate the DCS service info, it says I'm CPU bound.

Looking at other postings here and on Reddit it reads that for VR the CPU has a not neglectable impact. Which brings me to sites like cpu-monkey and direct comparisons between different CPUs, where the 285K definitely outperforms all existing X3D variants (7800, 9800) for the single core performance. E.g.
285K vs. 7800X3D -> Cinebench R23 (2380 vs. 1811)
285K vs. 9800X3D -> Cinebench R23 (2380 vs. 2086)
285K vs. 9900X3D -> Cinebench R23 (2380 vs. 2231) Expected, not confirmed yet

What's your opinion on the 285K vs. the AMD lineup, specifically when it's about playing DCS in VR? I would highly appreciate your comments.

Thank you very much in advance.

 

 

You’re looking at the wrong source in the first place. I recommend you to seek out the people that do actual proper gaming performance tests.

Gamers Nexus, Hardware unboxed, pc games hardware, Jayz2cents…..to name but a few. The result is clear as a sunny day. Any(!) X3D mops the floor with Intel (for gaming). Including the one you already own.

[lBlackMambal]

vor 2 Stunden schrieb Aapje:

Cinebench is more relevant to what are commonly called 'productivity applications,' and not so much to gaming.

In games, the code that matters is nearly always the gameplay loop, which are the calculations that have to be done for every frame that is shown on the screen. This in large part uses the same data again and again, which is why the X3D-cache works so well for gaming, since the data that keeps getting reused is kept in this cache and so the CPU doesn't need to constantly retrieve it from RAM. The gameplay loop also pretty much never benefits from a huge amount of cores, since to get the calculation done quickly enough to get a high frame rate, you can't afford the latency costs of having a huge amount of cores work together.

In gaming, the CPU also can't afford to work ahead too much, since each new frame needs to adapt to the user input, the behavior of enemies, etc. This means that lots of cores can't be used effectively, to work ahead.

In 'productivity applications,' in many situations you have a lot of data that is not constantly reused for each calculation, so the X3D-cache doesn't help much. The data still needs to be retrieved from RAM, since mostly new data is needed for each calculation that is not present in the cache. Furthermore, these applications don't have to react to user input that much and can take a lot more time to do their work. For example, in Cinebench, making a single frame can take seconds. A game with a frame rate less than 1 FPS would be completely unplayable, so Cinebench is not representative of how games do things.

So for productivity applications, you tend to want: lots of cores and lots of bandwidth to the RAM. The speed of the cores, a very big cache and low latency to the RAM are less important.

For games, you tend to want: very fast cores, a big cache and low latency to RAM (although with a big enough cache this becomes much less important). The number of cores, and the bandwidth to the RAM are less important.

So essentially, the strengths of the 9800X3D match the demands of gaming very well, but not so much the demands for productivity applications. The 285k is more tailored for productivity applications. If you want to have strong performance for both kinds of applications, then the 9950X3D is almost certainly going to be the best, although that is achieved by more or less turning it into a 9800X3D for games (where the chiplet without X3D is not used for the game).

Thank you once more for your reply. Yes, I totally get the point that the X3Ds are playing in another league in terms of gaming performance. And 7800X3D, 7900X3D, 9800X3D and the upcoming 9900X3D aren't the power houses if it's about production software.

Thing is, and that's why I brought it up in the DCS forum, I'm not sure if DCS can be considered as this typical "gaming scenario" - where the X3Ds generally outperform like crazy.

If you look at this YT comparison between a 7800X3D and a much older 12900K in a DCS VR usage, it's showing interesting results:

 

For sure, the clip isn't the most recent one, i.e. using an older build of DCS, different GPU drivers, chipset drivers etc. and someone could say, it's because "back then the CPU was new and systems/apps haven't been perfectly adjusted to it yet". Though based on the theoretical facts (gaming benchmarks back then), the X3D should have already beaten the 12900K by far.

Thing is, exactly the non-equal distribution across the cores is what I notice on my current setup with the most recent DCS build as well - it mainly sets the load on 2 cores, one of them all the time close to 100%.
Having in addition the feedback of a friend of mine, who is easily getting 90fps without the need of adjusting e.g. fov, pixel density, by running an Intel 14900KF (which, according to gaming benchmarks should also be outperformed, even by the "old" 7800X3D) makes me think twice if X3Ds really are the CPUs of choice for DCS - specifically when playing in VR.

P.S. it not about having the bigger e-p... against my friend I just want to be sure I get the best bang for the buck.

[Hiob]

DCS is a prime gaming specimen (specifically DX11). Why do you think it isn’t? If anything it will profit even more than more modern games.

As an example on how effective the 3D cache is. A 5800X3D provides 20-25% better CPU frametimes than a (faster by clock speed) 5900X.

[MoleUK]

1 hour ago, lBlackMambal said:

Thank you once more for your reply. Yes, I totally get the point that the X3Ds are playing in another league in terms of gaming performance. And 7800X3D, 7900X3D, 9800X3D and the upcoming 9900X3D aren't the power houses if it's about production software.

Thing is, and that's why I brought it up in the DCS forum, I'm not sure if DCS can be considered as this typical "gaming scenario" - where the X3Ds generally outperform like crazy.

If you look at this YT comparison between a 7800X3D and a much older 12900K in a DCS VR usage, it's showing interesting results:

 

For sure, the clip isn't the most recent one, i.e. using an older build of DCS, different GPU drivers, chipset drivers etc. and someone could say, it's because "back then the CPU was new and systems/apps haven't been perfectly adjusted to it yet". Though based on the theoretical facts (gaming benchmarks back then), the X3D should have already beaten the 12900K by far.

Thing is, exactly the non-equal distribution across the cores is what I notice on my current setup with the most recent DCS build as well - it mainly sets the load on 2 cores, one of them all the time close to 100%.
Having in addition the feedback of a friend of mine, who is easily getting 90fps without the need of adjusting e.g. fov, pixel density, by running an Intel 14900KF (which, according to gaming benchmarks should also be outperformed, even by the "old" 7800X3D) makes me think twice if X3Ds really are the CPUs of choice for DCS - specifically when playing in VR.

P.S. it not about having the bigger e-p... against my friend I just want to be sure I get the best bang for the buck.

The X3D CPU's tend to 1: Perform well in games generally, 2: Perform well in VR, 3: Perform well in atypical simulation heavy titles with lots of data being slung about and 4: Perform well with janky/old game engines.

That pretty much makes them perfect for a game like DCS. And all other simulation games in particular.

I can personally attest to this as I went from a 5800X to a 5800X3D while playing in VR on the single-threaded build. The effect was immediate and enormous, despite the loss in clockspeed my performance went WAY up.

You can also look at the F4E radar performance issues that many had on release where the X3D's still massively overperformed due to brute forcing through it via cache.

Long story short, if a big gaming focus of yours is DCS you want an X3D. But it's still best to avoid the two CCD X3D's due do windows scheduling problems, so the 7800/9800X3D's remain the best options.

Edited December 9, 2024 by MoleUK

[Raven (Elysian Angel)]

Do we know already what the architecture of the 9900X3D will be like? Will it again be 1 CCD with V-cache and the other without, or will it be a different architecture?
DCS MT scales with more cores, so I'm keeping my eye out on the 9900X3D when it releases. But I haven't made up my mind yet. I am currently on a 5900X but will very likely replace that sometime next year by either 9800X3D or 9900X3D depending on how the latter performs.

[LucShep]

22 hours ago, lBlackMambal said:

Thank you once more for your reply. Yes, I totally get the point that the X3Ds are playing in another league in terms of gaming performance. And 7800X3D, 7900X3D, 9800X3D and the upcoming 9900X3D aren't the power houses if it's about production software.

Thing is, and that's why I brought it up in the DCS forum, I'm not sure if DCS can be considered as this typical "gaming scenario" - where the X3Ds generally outperform like crazy.

If you look at this YT comparison between a 7800X3D and a much older 12900K in a DCS VR usage, it's showing interesting results:

 

For sure, the clip isn't the most recent one, i.e. using an older build of DCS, different GPU drivers, chipset drivers etc. and someone could say, it's because "back then the CPU was new and systems/apps haven't been perfectly adjusted to it yet". Though based on the theoretical facts (gaming benchmarks back then), the X3D should have already beaten the 12900K by far.

Thing is, exactly the non-equal distribution across the cores is what I notice on my current setup with the most recent DCS build as well - it mainly sets the load on 2 cores, one of them all the time close to 100%.
Having in addition the feedback of a friend of mine, who is easily getting 90fps without the need of adjusting e.g. fov, pixel density, by running an Intel 14900KF (which, according to gaming benchmarks should also be outperformed, even by the "old" 7800X3D) makes me think twice if X3Ds really are the CPUs of choice for DCS - specifically when playing in VR.

P.S. it not about having the bigger e-p... against my friend I just want to be sure I get the best bang for the buck.

You're taking a comparison with an older version of DCS, and multiple Agesa updates to AMD motherboards since then that have allowed better/stable performance.

The thing is, as others said here, the X3D chips from AMD have a very clever "cheat", which is the 3D V-Cache. And that on itself, for now, is an unbeatable crutch over any downsides that the AMD Chiplet design does have. A particular feature so good with CPU limited games including simulation genre (DCS, Assetto Corsa Competizione, etc) - especially with the single CCD ones like the 7800X3D and 9800X3D - that makes them the best gaming chips. Though definitely not as good for production/work tasks, are low in stock, and also (IMO) became too expensive.

That said, it makes sense why the i9 12900K holds so well, even today - it's a good Monolithic CPU design, which has significant advantages for latency (noticeably lower) over Chiplet designs, such as those from AMD and newest Intel (Arrow Lake) as well. And another upside, it's now much cheaper (half the price of the 9800X3D) and it doesn't suffer from any degradation whatsoever (as seen in the later Raptor Lake).
If Intel 12th gen Alder Lake and 13th/14th Raptor Lake would have had 3D V-Cache, they'd certainly be unbeatable CPUs for gaming but, alas.... it's an AMD feature (so far).

Although it is now a dead platform, the i9 12900K is still a valid solution and a great price/performance choice if somewhat on a budget.
In such situation, I'd still recommend it with a mid-range Z690 or Z790 motherboard (Asus TUF Gaming Plus, MSI Tomahawk), new or used. These also also exhist in DDR4 versions (not just DDR5 versions), which means it may allow for re-usage of DDR4 memory (in case that you already have 64GB of DDR4 RAM) with negligible difference to gaming performance.

One thing is certain... avoid Intel 15th gen Arrow Lake (or better, "Error Lake"), because of its inconsistency and huge latency penalty for gaming.
It teased with great expectations but turned out to be the biggest flop I've seen since the AMD Bulldozer days. 

Edited December 10, 2024 by LucShep

[hannibal]

7800X3D and 4090. u have a good setup!