Monday, August 4th 2025
Battlefield 6 Heavily Favors CPUs with 3D V-Cache, 9800X3D Smokes i9-14900K
With closed beta access to Battlefield 6, popular Twitch streamers such as Bruhskey are able to tell us some early performance numbers for the game, and while their NDAs prevent them for sharing screenshots just yet, these can be counted as reliable sources. Bruhskey says that his AMD Ryzen 7 9800X3D processor with 3D V-Cache technology is able to yield significantly higher frame-rates than the Intel Core i9-14900K powered build of his unnamed friend that's also on the early access. Apparently, the latest version of the Frostbite engine powering Battlefield 6 heavily favors processors with large last-level caches, and the 9800X3D has 96 MB of it.
Bruhskey's rig features a 9800X3D, a GeForce RTX 5080 GPU, 32 GB of DDR4-6400 memory in a 1:1 FClk:UClk ratio, and an unoptimized NVIDIA graphics driver. Frostbite engine puts out framerates for the CPU and GPU ends of the graphics pipeline separately, so you get to identify where your performance bottleneck is. Bruhskey's rig scored 330 to 370 FPS on the GPU end, and 310 to 330 FPS on the CPU end, with a total system latency of 6.7-8.5 ms. Here's where it gets interesting: Bruhskey claims that his friend's i9-14900K powered rig yields a CPU FPS of around 210 FPS, which puts the Intel processor at a roughly 30% performance disadvantage.
Intel, which has until now given its client processors heavy dedicated L2 caches, is pivoting toward larger shared L3 caches with its upcoming "Nova Lake-S" desktop processors. Currently, the fastest "Arrow Lake-S" processor comes with large 3 MB dedicated L2 caches per P-core, but with only 36 MB of L3 cache that's shared among eight P-cores and sixteen E-cores. "Nova Lake-S" sees Intel come up with a maximum core configuration of 16P+32E+4LPE—double the current core-count with a low power island cherry on top—but with a large L3 cache of 144 MB.
Sources:
Bruhskey (Twitter), Tom's Hardware
Bruhskey's rig features a 9800X3D, a GeForce RTX 5080 GPU, 32 GB of DDR4-6400 memory in a 1:1 FClk:UClk ratio, and an unoptimized NVIDIA graphics driver. Frostbite engine puts out framerates for the CPU and GPU ends of the graphics pipeline separately, so you get to identify where your performance bottleneck is. Bruhskey's rig scored 330 to 370 FPS on the GPU end, and 310 to 330 FPS on the CPU end, with a total system latency of 6.7-8.5 ms. Here's where it gets interesting: Bruhskey claims that his friend's i9-14900K powered rig yields a CPU FPS of around 210 FPS, which puts the Intel processor at a roughly 30% performance disadvantage.
73 Comments on Battlefield 6 Heavily Favors CPUs with 3D V-Cache, 9800X3D Smokes i9-14900K
Total water is wet scenario.
Unless you're saying that going from 6000 2000 1:1 to 6400 2133 1:1 adds 130-170fps?
What's even more hilarious is that it's an Intel sponsored game. That would assume specific optimizations. Maybe only for Arrow Lake/Battlemage then?I have yet to see game where even my 5800X3D compiles shaders for 10 minutes.
Besides which one is preferable?
Waiting 6 minutes for shadercomp with higher performance and lower power or waiting 3 minutes for shadercomp with lower performance and higher power?
The choice is pretty clear.
1. Sponsorships don't necessarily mean "optimize for our gear at the detriment of the competition", this is a folk tale that outstayed its welcome
2. There is absolutely nothing new in this trend, every other game exhibits the exact same behavior to varied degrees (and in some cases even more extreme, see: Factorio)
3. ARL doesn't introduce anything relevant regarding instruction sets vs. ADL/RPL other than the weak NPU (see image)
4. ADL/RPL architecture is literally 4.5 years old, that they're even spoken of in the same line is much more merit than it ever deserved, it's actually amazing how well it aged
I just :confused:It's a hypothetical scenario and I actually agree with you there, but just saying
i9 14900K on auto settings..(voltages and frequency, complains crash on shader compile when its his dumb@$$ fault)
memory on auto settings..(speed and timings)
GPU on auto overclock and probably doesn't know what Re-Bar is..
Runs on a potato AIO..
his 9800X3D, oh well, nothing is new..
here we go again with the AMD architecture vs Intel stuff..blah blah blah.. (comedy and Irony)
These kinds of articles are more interesting than about 90% of the other articles on this site (not a snide to TPU).
Intel being Intel I would only expect any optimizations for the latest hardware, and Intel optimizing their CPU in order to make the competition look worse is what Intel has done going back decades.Which is how most people will run their gaming system with auto CPU settings and if they're lucky, turn on XMP. Expecting everyone to tweak and adjust things is something only enthusiasts do and why 13th and 14th gen chips frying themselves was such a major issue.
So yeah, I'd argue this should be here considering other rumors we see as well.
to each of our own, I'm just here to jest in a healthy way I can, til I see the AMD boi's comin..
The extra four LPE cores are physically located in a separate tile, which should allow for large power savings as the entire P+E-core complex tile can then be entirely turned off at idle. This is of more interest to laptops than desktops, but as long as thread coordination is effective (and Intel does have a hardware feedback function for this), it should not be detrimental to performance overall.The point is that it's speculative and based on hearsay, even if it sounds plausible (and is not unheard of before)
And as execution ports, buffers and queues changes with every generation, so even if one generation shows large advantages from a certain combination of instructions, it's very likely that this balance will change with the next generation. When dealing with tiny functions compiled down to just a few lines of assembly, you can certainly see some which heavily favors this architecture or another. But as mentioned, most of these differences aren't universally for one party or the other, and it's not like you can choose which instructions can do your task. The only attribute that I've seen remain consistently different between AMD and Intel is that AMD has an advantage with dense math vs. Intel performs better at complex logic. (But even this is subject to change when APX arrives.) This means inside a function scope, the more you remove e.g. branching and other unnecessary operations (even register shuffling), the more it generally tend to favor AMD. (But I don't think "any" games are low-level optimized today.) But going the other way is hard, that is making it favor Intel, practically impossible unless you want to sabotage the code with bloat or are just incompetent. Large L3 caches adds another attribute into the mix, but to intentionally make your code L3 cache sensitive would be beyond stupid, and would entail adding extra layers of abstraction on top of everything to make the active code too big for L2. It is very possible though to unintentionally achieve this by overengineering and bad coding practices.The fact that shader compilation is an issue at all is a testament to poor development practices. There is no good reason why you should keep recompiling them.
"auto settings is normal and fair"
- fully tuned balls to the wall 9800X3D with memory speeds most won't be able to get, no mention or acknowledgment.
Fascinating. Expected attitude though. You're perfectly happy to be duplicitous if it favours your favourite manufacturer. The "hilarious" thing is, as others have mentioned, Raptor Lake is an old architecture, vs the 2025 9800X3D it's expected which one will generally perform better in games, so there's no need to test a heavily OC system vs assumingly a stock one.
Or to ignore that's what's happening here. But you do anyway, with glee.
Now test a max tune 14900K.
I also cant agree on the notion that because of hardware variety precached files could not be included with games due to size.
I've looked at various compiled shader caches on my PC and these usually are below 400MB per game.
Including 10 most common hardware configuration precaches with a game would only cost 4GB extra space. In the age of 150GB games this is small.
Companies could even offer precompiled packages for a a variety of hardware as optional downloads with the game itself. Some sort of automated system would detect the hardware and suggest the best package to download based on the hardware. Reducing the need to include any in the installer. Like some already do for languages, SP/MP etc. This can significantly reduce installer size.