找回密码
 加入我们
搜索
      

Ultra200 官方PPT 技术介绍和前瞻

查看数: 4943 | 评论数: 40 | 收藏 0
关灯 | 提示:支持键盘翻页<-左 右->
    组图打开中,请稍候......
发布时间: 2024-10-16 15:36

正文摘要:

来自于:新潮电子杂志,微博的链接:https://weibo.com/ttarticle/x/m/ ... 80849?_wb_client_=1 主要基于官方PPT和一些线下会议沟通(文章到提到),之前已经泄露了一部分PPT就不重复介绍了。 我提炼下一些比较有 ...

回复

kinglfa 发表于 2024-10-17 10:14
分割为了更细的16.67 GHz.............
fromiss 发表于 2024-10-17 10:07
用ultra 200内置显卡解码,能通过nvidia显卡输出么?
格文多琳 发表于 2024-10-17 10:01
不习惯没超线程的英特尔
aibo 发表于 2024-10-17 10:00
pdvc 发表于 2024-10-16 23:44
编解码不支持H266/VVC么?

我之前发过一个帖子

https://www.chiphell.com/thread-2638715-1-1.html

有人说月亮湖可以
难道桌面这个还吸回去了
aibo 发表于 2024-10-17 09:59
gartour 发表于 2024-10-16 20:40
decode only 是什么意思?只能解码不能播放?

大概率是渲染带宽不够

从7代U开始就发现,单通道根本没法解4k60,双通道2400有时候还会缺帧,到2666才基本能稳定
h155的本子,8k60的av1也是缺帧很多,差不多就对上这个8k30了
panzerlied 发表于 2024-10-17 09:55
过时技术研究所 发表于 2024-10-17 09:43
战地是公认的线程优化垃圾,有啥看的意义

战地用的线程比较多。
赫敏 发表于 2024-10-17 09:51
dens 发表于 2024-10-16 02:54
小核IPC提升32%?那特么都不叫小核了吧。和12代大核CPU-Z跑分基本没差了

算力有了,IO还是残废。等把缓存加大IO加强面积又压不下去了
过时技术研究所 发表于 2024-10-17 09:43
panzerlied 发表于 2024-10-16 17:38
你可以试试战地。

战地是公认的线程优化垃圾,有啥看的意义
BFG9K 发表于 2024-10-17 07:49
panzerlied 发表于 2024-10-16 15:40
亮点还是有的。

有一说一确实辛苦特尔的实验室了,辛辛苦苦挑项目最后-5%。哪怕是能挑项目到打平也不会这 ...

等U9 295出来了,一切都会好起来的(
pdvc 发表于 2024-10-16 23:44
编解码不支持H266/VVC么?
gartour 发表于 2024-10-16 20:40
decode only 是什么意思?只能解码不能播放?
ioko 发表于 2024-10-16 19:23

应该是16.67hz,
theshy 发表于 2024-10-16 18:01
16.67 GHz?
天道太酬勤 发表于 2024-10-16 17:51
”系统功耗降了188W“,这个到底是怎么测出来的?同帧数下?
同显卡占用率CPU功耗能低这么多?这提升不是起飞了
panzerlied 发表于 2024-10-16 17:38
dens 发表于 2024-10-16 17:30
小核这B玩意以前我是很嫌弃的,WIN10到手就关了。
直到两个月前升24H2系统,我就改成开小核关超线程用了 ...

你可以试试战地。
gladiator 发表于 2024-10-16 17:34
IO和SOC Tile采用N6制程工艺
AMD:intel也干了
dens 发表于 2024-10-16 17:30
r147909 发表于 2024-10-16 16:35
只有IPC高没用啊 cache呢
AI 9的zen5C也是拖后腿的,只有生产力能用
什么有人不满意了? 说来比一比多核 ...

小核这B玩意以前我是很嫌弃的,WIN10到手就关了。
直到两个月前升24H2系统,我就改成开小核关超线程用了。
你别说,玩吃鸡的时候8个小核都均衡的吃上了,而且帧数还不低,搞不懂

捕1.JPG


小心夹脚 发表于 2024-10-16 16:56
提示: 作者被禁止或删除 内容自动屏蔽
毛茸茸 发表于 2024-10-16 16:55
KimmyGLM 发表于 2024-10-16 16:49
个人以为,分布式P核 E核布局,有效降低积热,也是不错的创新

就看实测效果如何了
CyanCloverFern 发表于 2024-10-16 16:51
本帖最后由 CyanCloverFern 于 2024-10-16 16:59 编辑
KimmyGLM 发表于 2024-10-16 16:31
Arrow Lake also features an updated Thread Director. For the P-Cores, the new telemetry structure  ...


这段话出处是wccfetch不是牙膏ppt。

PPT(牙膏最终发布的ppt里没有这图)里的图片看着挺像,但Pcore1更密集,无法确认为具体是不是任务启动首先运行到Ecore。
图二里P core工作更密集,而E core就不干活了。



牙膏官网的回复更是白开水:
Yeah it's still the same scheduling behavior but intense workloads communicates hint to the CPU to shift the workload over to the P core. We fully expect that the vast majority of hours that the person hours, the man hours, of all the collective  CPUs that will be sold, most of those would probably be in a desktop case. The gamers game a few hours a day so we have to think about the total power consumption footprint of the product over its life cycle.

Starting with E-cores first is the way to go for power because people aren't playing game all the time. When they do, the game will immediately snap to P cores like they do on Lunar because that's what the games tell the Thread Director to do.

Robert Hallock (Intel VP & General Manager Client AI and Technical Marketing)


个人PC一般而言响应型请求 >> 吞吐型请求,降优先级的调度好过升优先级的调度,如果是升优先级我不看好牙膏的改良调度器。


KimmyGLM 发表于 2024-10-16 16:49
毛茸茸 发表于 2024-10-16 16:48
反正性能没提升,最大创新不是去掉超线程?

个人以为,分布式P核 E核布局,有效降低积热,也是不错的创新
毛茸茸 发表于 2024-10-16 16:48
反正性能没提升,最大创新不是去掉超线程?
mhsk 发表于 2024-10-16 16:38
步子迈太大,缓缓。
r147909 发表于 2024-10-16 16:35
本帖最后由 r147909 于 2024-10-16 16:36 编辑
dens 发表于 2024-10-16 15:54
小核IPC提升32%?那特么都不叫小核了吧。和12代大核CPU-Z跑分基本没差了


只有IPC高没用啊 cache呢
AI 9的zen5C也是拖后腿的,只有生产力能用
什么有人不满意了? 说来比一比多核
kaiwenwu 发表于 2024-10-16 16:34
喝茶 发表于 2024-10-16 15:47
如果从用户场景来看,英特尔也公布了与AMD最强游戏处理器7950X3D的比较结果,5个主流游戏的平均成绩来看, ...

总不可能挑微软飞行模拟吧直接被吊锤,再说内存用的还是6400也没啥说法起码安装新平台的特性来个9200 9600吧
KimmyGLM 发表于 2024-10-16 16:31
CyanCloverFern 发表于 2024-10-16 16:18
翻了一遍PPT,没找到E核优先于P核的信息。(月亮湖的LPE大于P这个确定,来源于实测,但低功耗岛一开始就是 ...

Arrow Lake also features an updated Thread Director. For the P-Cores, the new telemetry structure has been designed so that threads can be directed accurately and according to the workload, and for the E-Cores, the new HW-based prediction model can be leveraged for better classification to see if the workload can be accommodated on just the E-Cores or needs to be moved to the P-cores for higher performance. Lastly, the prediction model has been improved to offer more informed & controlled scheduling across the cores.

结合这段,基本可以确认是E核优先
KimmyGLM 发表于 2024-10-16 16:29
CyanCloverFern 发表于 2024-10-16 16:18
翻了一遍PPT,没找到E核优先于P核的信息。(月亮湖的LPE大于P这个确定,来源于实测,但低功耗岛一开始就是 ...

看采访:

Yeah it's still the same scheduling behavior but intense workloads communicates hint to the CPU to shift the workload over to the P core. We fully expect that the vast majority of hours that the person hours, the man hours, of all the collective  CPUs that will be sold, most of those would probably be in a desktop case. The gamers game a few hours a day so we have to think about the total power consumption footprint of the product over its life cycle.

Starting with E-cores first is the way to go for power because people aren't playing game all the time. When they do, the game will immediately snap to P cores like they do on Lunar because that's what the games tell the Thread Director to do.

Robert Hallock (Intel VP & General Manager Client AI and Technical Marketing)
CyanCloverFern 发表于 2024-10-16 16:18
翻了一遍PPT,没找到E核优先于P核的信息。(月亮湖的LPE大于P这个确定,来源于实测,但低功耗岛一开始就是待机目的)
ITD只有一句,vary by os。
Built into the hardware, Intel® Thread Director is provided only in performance hybrid architecture configurations of 12th Gen or newer Intel® Core processors; OS enablement is required. Available features and functionality vary by OS.
dens 发表于 2024-10-16 15:54
小核IPC提升32%?那特么都不叫小核了吧。和12代大核CPU-Z跑分基本没差了

Archiver|手机版|小黑屋|Chiphell ( 沪ICP备12027953号-5 )沪公网备310112100042806 上海市互联网违法与不良信息举报中心

GMT+8, 2025-4-28 13:55 , Processed in 0.014114 second(s), 10 queries , Gzip On, Redis On.

Powered by Discuz! X3.5 Licensed

© 2007-2024 Chiphell.com All rights reserved.

快速回复 返回顶部 返回列表