- 精华
- 0
- 帖子
- 219
- 威望
- 0 点
- 积分
- 228 点
- 种子
- 10 点
- 注册时间
- 2013-12-22
- 最后登录
- 2024-10-28
|
发表于 2020-4-22 15:03 · 广东
|
显示全部楼层
本帖最后由 浅唱离歌笑天涯 于 2020-4-22 15:19 编辑
https://www.eurogamer.net/articl ... series-x-full-specs
反复对比了DF的原文稿确定我当初没有理解错,DF这B绝对收了微软的钱,明明就是个缺点却当优点语气来说,N卡的机械学习可以直接从tensor里面调用资源运算,而XSX的机械学习(光追降噪等)用到的性能要从12T FP32里面拆,这个49TOPS INT8 和 97 TOPS int4 是把12T FP32完全换算过来的,然而可能吗?游戏不可能调用所有FP32用来机械学习,只能划分出一小部分,说白了就是架构对比图灵拉胯,这玩意光追效率肯定不行,竟然好意思说是2080ti光追性能的3.8倍,且不说那3800billion 和 10Gigaray 单位都不同,连算数都算错了,他出来算来是3800W:1000W ,然而10Giga 明明是100W ,按他这个说法xsx是2080ti的38倍,可笑不可笑?那么这些闹得满天飞的谣言是哪来的?除了地摊英语水平的营销号水军和 听风是雨到处找显卡碰瓷带节奏的软软们还有谁?
不过话又说回来,NV和DF也都说过,游戏在光追方面的机械学习部分其实占用的性能并不多,我个人保守估计12T里面划分2个T出来应该是够用的
It was an impressive showing for a game that hasn't even begun to access the next generation features of the new GPU. Right now, it's diffi** to accurately quantify the kind of improvement to visual quality and performance we'll see over time, because while there are obvious parallels to current-gen machines, the mixture of new hardware and new APIs allows for very different workloads to run on the GPU. Machine learning is a feature we've **ed in the past, most notably with Nvidia's Turing architecture and the firm's DLSS AI upscaling. The RDNA 2 architecture used in Series X does not have tensor core equivalents, but Microsoft and AMD have come up with a novel, efficient solution based on ** shader **. With over 12 teraflops of FP32 compute, RDNA 2 also allows for double that with FP16 (yes, rapid-packed math is back). However, machine learning workloads often use much lower precision than that, so the RDNA 2 shaders were adapted still further.
"We knew that many inference algorithms need only 8-bit and 4-bit integer positions for weights and the math operations involving those weights comprise the bulk of the performance overhead for those algorithms," says Andrew Goossen. "So we added special hardware support for this specific scenario. The result is that Series X offers 49 TOPS for 8-bit integer operations and 97 TOPS for 4-bit integer operations. Note that the weights are integers, so those are TOPS and not TFLOPs. The net result is that Series X offers unparalleled intelligence for machine learning."
|
|