In some ways, 3DMark06 represents a significant upgrade over 3DMark05. Certainly, HDR rendering is gaining popularity and will be common in the future, so having a good graphics benchmark test for it is important. The shadow algorithm in 3DMark05 wasn't really used in any games we can think of, and cube shadow maps should be a much more common shadow solution. 3DMark06 also places the workload more on the pixel shader, where earlier 3DMark versions were largely vertex-shader bound. This puts it more in line with where the bottleneck is in real games.

3DMark06 的確比 05 有了大幅度的進步, 像是 HDR 的使用, 目前使用 HDR 的遊戲也越來越普及.

另外05裡面的陰影技術並幾乎沒有使用在真實遊戲上, 反而是 cube shadow map 比較常用 (06改回來了嗎?)

另外06也把重心逐漸的移到 Pixel Shader 上面, 跟之前版本以 Vertex Shader 為主不同. 更加的趨近最近的遊戲趨勢 (瓶頸多半在 Pixel Shader 上)



Still, we're left with quite a few worrisome questions. We can understand that Nvidia graphics cards won't run the SM3.0 tests with anti-aliasing—the hardware simply does not support floating-point blending together with AA (at least without using the pixel shaders for a very slow hack). We can't figure out why this should prevent those cards from producing a score, though. There's a formula for generating a 3DMark score using just the SM2.0 and CPU scores—why wouldn't a GeForce 7800 GTX card use that formula when AA is enabled? A GeForce 6200 supports SM3.0 but not floating-point blending. We still consider that a SM3.0 card, and yet it uses the formula for SM2.0 cards because it does not support the FP16 blending required for the SM3.0 + HDR tests. Why the discrepancy? Why doesn't the GeForce 6200 just produce no score? The rules are different when AA is enabled than they are when it isn't.

不過還是有些地方 Orz. 比如說 NV 的卡無法使用 SM3.0 的同時啟用 AA. (只能用 Pixel Sharder 硬幹, 會很慢)
不過詭異的是, 這時候計算成績就無法顯示了. 其實 GF6200 也有類似的困擾, 它的 SM3.0 無法正確的支援 HDR (Floating-point blending), 這時候計算成績時, 就使用 SM2.0 + CPU 加權來計算, 忽略掉 SM3.0 的成績.

所以6200 可以在無法使用 SM3.0 時計算出成績, 但是高階卡卻在開啟 AA 時無法使用 SM3.0 然後計算不出成績.

(注: 同樣的 6200 跟高階卡 開啟 AA 都無法執行 SM3.0, 一個有成績,另外一個沒成績)


Nvidia likely believes that having no score is better than having a low score. According to Tony Tamasi, Nvidia believes that an "NA" (not applicable) is the correct way to report the score.

Nvidia 傾向相信 "沒有成績會比低成績好", 回報 "NA" 是比較能夠反映正確成績的方法.
(ATI 抱怨中: 打不贏了就不要報成績, 這樣就沒人會比較, 也就不會說你比較爛, 因為無法比較)


Then there are the 24-bit depth stencil textures. This is supposedly an optional format in DX9, and yet it is required by Futuremark's new shadow algorithm. Only Nvidia supports this format, but rather than using a 16-bit depth stencil texture on ATI cards, they're forced into a costly workaround. If ATI is forced into a costly workaround for the depth stencil texture format, why isn't Nvidia forced into a costly workaround for HDR + AA? Here we have a major hardware manufacturer claiming that the 24-bit format for the DST is not required to preserve shadow quality, nor that it is commonly used by game developers. Futuremark claims that it is necessary. We'd like to see the benchmark using 16-bit DSTs just to compare quality and judge for ourselves.

另外有關於 24bit depth stencil textures. 只有 NV 支援, 而 ATI 無法支援時, 用了一個粉慢的方式去處理, 而不是使用 16 bit 的模式去執行.
(ATI抱怨中: 既然 NV 的 SM3.0 + AA 可以粉慢處理, 怎麼這個時候卻不叫它做呢 ?)

有硬體廠商抱怨, 這種 24 bit 模式很少遊戲會使用.(注: 除了 ATI 還有誰會抱怨這個啊 ) 但是FutureMark 卻強調說這是必要的. 其實蠻多人應該想知道 使用 16bit 模式的 DST 下 ATI 會不會比較快, 但是比較醜吧
(注: 我是有點想知道會不會比較快, 不過當年 nv 也是這樣吃鱉很久, 也是使用 FP16 之類的醜圖來撐, 結果網路上還是評價不高 )

Finally, where's the parallax mapping? Parallax mapping is a normal-mapping technique that takes the bump maps we're used to seeing to the next level, performing a series of computations based on the viewing angle to give textures an even greater sense of depth while reducing the distortion that earlier bump mapping methods can show at sharp viewing angles. We're already starting to see it in games like F.E.A.R., and it's the normal-mapping procedure of choice in future games. You're going to see it all over the place over the next year or so. If 3DMark06 is supposed to be a forward-looking benchmark, why isn't this pixel shader operation, essentially a "given" in upcoming games, all over the place?

最後, 那個很屌的 Parallax Mapping 跑到哪裡去了, 這是一種 Normal-Mapping 方式, 可以看到正很多的凹凸特效
目前 FEAR 已經用上了, 預計其他遊戲也會很快跟上, 不過 3DMark06 完全沒有用上這樣的技術.
(注:這點倒是蠻中肯的, 將來的遊戲應該都會卯起來用才是, 怎麼06 沒用上呢)