TE
TechEcho
Home24h TopNewestBestAskShowJobs
GitHubTwitter
Home

TechEcho

A tech news platform built with Next.js, providing global tech news and discussions.

GitHubTwitter

Home

HomeNewestBestAskShowJobs

Resources

HackerNews APIOriginal HackerNewsNext.js

© 2025 TechEcho. All rights reserved.

Russian-Made Elbrus CPUs Fail Trials, 'A Completely Unacceptable Platform'

22 pointsby AuthorizedCustover 3 years ago

7 comments

codedokodeover 3 years ago
If you read the article, you&#x27;ll see that the difference in performance is not that huge:<p>&gt; SPEC CPU 2017 2.62 (base) ~ 3.15 (peak) times slower<p>&gt; PGbench&#x2F;PostreSQL 1.7 (read only profile) ~ 3.3 (read write profile) times lower<p>So, Elbrus is just 2-3 times slower in these tasks than powerful Intel&#x27;s CPUs that took millions of dollars to design. As Elbrus has lower clock frequency and larger process node, so this is expected. It is difficult to outperform a CPU that has smaller transistors and more of them; you need either to have more transistors or they have to be smaller; or both.<p>But Elbrus loses a lot in Java tests:<p>&gt; Java 23 ~ 26 times higher response time<p>I guess this could be because Java JIT doesn&#x27;t support Elbrus architecture.
codedokodeover 3 years ago
&gt; Therefore, by the time the ambitious 32-core Elbruses arrive, chips from AMD and Intel will be orders of magnitude faster and more efficient in terms of performance-per-watt than they are today.<p>I think that is wrong assumption because CPUs nowadays don&#x27;t become &quot;orders of magnitude faster&quot; in several years.
ab_testingover 3 years ago
I dont know how long they have been in development, but if it is their first attempt, then I would say that they are doing pretty good. The chips that these are being compared against chips that have 40 years of development dollars poured into them. May be if they keep at it, then in a xouple of years it will come out competitive.
johnklosover 3 years ago
I think this article slightly misses the point.<p>If the architecture was actually failing - that is, if it were dumping core when running regular workloads, or crashing, or corrupting data - that would be a real reason to not use it.<p>But it seems like it&#x27;s simply slower than they&#x27;d like, plus there are features they want (better casing, better LEDs, remote management). These don&#x27;t have much to do with actually running code.<p>Speed isn&#x27;t everything. We don&#x27;t put supercomputers in to space - we put slow but extremely reliable processors in to space. Slow and reliable is much, much more important to space and financial (non-trading) applications than fast and not-as-reliable.
评论 #29689664 未加载
d0mineover 3 years ago
It looks like kickbacks for buying Elbrus are not as great as for Xeons.
guenthertover 3 years ago
Uh, no mention of SPECTRE? I&#x27;d thought that would matter in a banking application; more than the difference in TPS even.
AuthorizedCustover 3 years ago
Sputnik V vaccine, Buran shuttle, USSR itself, I could go on. Will Russia ever shed its reputation of being a perpetual “me, too” with poor execution? Is the world better off without Russia overcoming it?
评论 #29692366 未加载