Can FSR 4 be enabled on a GTX 1080? Has anyone succeeded?

[qwp12311]

Can FSR 4 be enabled on a GTX 1080? Has anyone succeeded?

[MLuc24]

No. FSR4 is AMD-only hardware, and a 1080 isn't in the ballpark — the README lists it as "officially, RDNA4 and RDNA3 dGPUs only" every time it comes up. Pascal isn't RDNA at all, so there's no path to it, working or otherwise.

What OptiScaler can do for you on that card is swap a game's existing upscaler for FSR 3.1 or XeSS. Both run on Nvidia cards. Worth setting expectations though: XeSS on Pascal falls back to a slower code path since the card has no DP4a, so FSR 3.1 is usually the better pick there.

[MLuc24]

You're right, and I got that wrong — sorry. The 1080 is GP104, compute capability 6.1, which is exactly the generation where Nvidia added the INT8 dot-product instruction. The Pascal part that lacks DP4a is GP100 (Tesla P100, CC 6.0), not the consumer 10-series.

DP4a availability is really a Shader Model 6.4 question, and Pascal meets it. OptiScaler treats it that way too — the status flag it exposes is gated on the shader model, not on a hardware generation list:

{ "DP4A_ACTIVE", 0x20000000, "OF pipeline using dp4a-accelerated SSD (SM 6.4+)" },

(that one's the OptiFG optical flow path rather than XeSS, but the capability question is the same.)

So XeSS on your 1080 runs the DP4a path, not some slower generic fallback. What's still true is that Pascal executes those INT8 ops on its regular integer ALUs — there's no XMX or tensor hardware behind them — so the path works but costs a lot of frame time on a card of that age. That's a performance trade-off, not a missing feature, which is the part I described incorrectly.

FSR 3.1 still tends to be the better pick on a 1080 for that reason, but it's worth just trying XeSS and comparing — on your card the gap is a matter of measurement rather than something ruled out in advance.

The FSR4 part of my earlier answer stands: RDNA4/RDNA3 only, nothing to do with DP4a.

[qwp12311]

The GTX 1080 should theoretically be able to run the INT8 version of FSR 4. However, when I tried forcing it via OptiScaler in Zenless Zone Zero and Atomic Heart, the GPU doesn't seem to work properly.

[MLuc24]

That tracks, and it's a different wall from the DP4a one — FSR4 isn't a shader OptiScaler compiles for whatever GPU you have, it loads AMD's own amdxcffx64.dll and that library in turn talks to amdxc64 on the driver side, so what gates it isn't whether your card can do INT8 math but whether AMD's driver stack is there to answer at all. The INT8 model is a fallback path inside that library for RDNA3, not a generic route any INT8-capable GPU can take, which is why the support flags in OptiScaler are things like amdxcffx64SupportsInt8 rather than a hardware capability query. So the 1080 having DP4a is real and it genuinely matters for XeSS, it just buys you nothing for FSR4 — DP4a is dot4add_i8packed in an ordinary shader, while FSR4 goes through AMD's cooperative-vector path in their driver, and those aren't interchangeable. Forcing it will fail the same way in every game, so ZZZ and Atomic Heart aren't telling you anything about those two titles specifically, and I'd put the time into FSR 3.1 versus XeSS instead since both actually run on your card.

[qwp12311]

FSR 3.1 and XeSS both don't perform very well — the anti-aliasing quality is not great on either of them

[qwp12311]

这说得通，而且它与 DP4a 的瓶颈不同——FSR4 不是 OptiScaler 为你的 GPU 编译的着色器，它加载的是 AMD 自家的着色器库amdxcffx64.dll，而驱动程序端又会amdxc64与该库进行交互。因此，关键不在于你的显卡是否支持 INT8 运算，而在于 AMD 的驱动程序栈是否能够响应。INT8 模型是 RDNA3 架构在该库中的一个回退路径，而不是任何支持 INT8 的 GPU 都可以采用的通用路径。这就是为什么 OptiScaler 中的支持标志是类似这样的，amdxcffx64SupportsInt8而不是硬件功能查询。所以，1080 支持 DP4a 是真实存在的，而且对 XeSS 来说确实很重要，但它对 FSR4 没有任何帮助——DP4a 位于dot4add_i8packed一个普通的着色器中，而 FSR4 则通过 AMD 驱动程序中的协同向量路径来实现，两者无法互换。强行测试在每款游戏中都会以同样的方式失败，所以 ZZZ 和 Atomic Heart 并不能告诉你关于这两款游戏的具体信息，我建议你把时间花在 FSR 3.1 与 XeSS 的对比上，因为这两款游戏实际上都可以在你的显卡上运行。

那20系为什么可以运行fsr4 int8