字号·· | 护眼
thenextweb

OpenAI称其AI在AMD旗舰GPU设计中实现14%提升OpenAI says its AI found 14% gains in AMD’s flagship GPU design

点「原文对照」整页切到原文,或双击某段只看那段的原文。

OpenAI表示,其模型在AMD旗舰GPU设计上将功耗、性能和面积三项指标提升了高达14%,部分变更已应用于一款已送制造的芯片,且在AMD下一代旗舰CPU的关键部分实现了个位数提升。OpenAI还详细披露了其自研Jalapeño芯片的AI优化成果,并正与Synopsys合作训练GPT-Synopsys模型。

OpenAI says its models improved power, performance and area by up to 14% on AMD’s flagship GPU designs, with some changes in a chip sent for manufacturing and single-digit gains on AMD’s next flagship CPU. It also detailed AI-made gains on its own Jalapeño chip and is training a GPT-Synopsys model with Synopsys.

OpenAI周三在一篇博客文章中称,其AI模型已改进AMD旗舰显卡芯片的设计,部分变更已进入一款送制造的芯片。文章报告称,在功耗、性能和面积——芯片设计师需要权衡的三项指标——上,相较于AMD高度优化的设计,最高实现了14%的提升。

OpenAI says its AI models have improved the design of AMD’s flagship graphics chips, with some changes already in a chip sent for manufacturing. In a blog post on Wednesday, it reported gains of up to 14% in power, performance and area, the three measures chip designers trade off, against AMD’s heavily optimised designs.

该模型重写了描述芯片电路的代码,并使用Synopsys软件验证每项变更仍能正常工作。OpenAI表示,这相当于在不到五个月内完成了约一年的进展。根据OpenAI披露,AMD也确认其下一代旗舰CPU的关键部分实现了个位数提升。

The model rewrote the code that describes a chip’s circuits, and used Synopsys software to check each change still worked. OpenAI says this was about a year’s worth of progress in under five months. AMD also confirmed single-digit gains in key parts of its next flagship CPU, according to OpenAI.

OpenAI在其首款自研芯片Jalapeño上采用了相同方法,该芯片从设计到制造仅用了九个月。模型建议的变更将芯片两个关键部分的面积分别缩减了约13%和10%,且在成品硅片上验证成功。初创公司也在追逐这一思路:Architect Labs于6月融资2400万美元,帮助企业利用AI设计定制芯片。

OpenAI used the same approach on Jalapeño, its first in-house chip, which went from design to manufacturing in nine months. Changes suggested by its models cut the size of two key parts of the chip by about 13% and 10%, and worked in the finished silicon. Startups are chasing the same idea: Architect Labs raised $24M in June to help companies design custom chips with AI.

OpenAI还用其五代模型测试了两款小型开源处理器的设计。最新模型GPT-6.2 Astra在改进现有设计和根据文字需求从头构建处理器两项测试中均得分最高。OpenAI指出,第二项测试较易,因为模型可省略原设计所具备的功能。

OpenAI also tested five generations of its models on designing two small open-source processors. Its newest, GPT-6.2 Astra, scored highest both when improving existing designs and when building a processor from written requirements. OpenAI noted the second test was easier, because models could leave out features the originals had.

接下来,OpenAI正与Synopsys合作训练GPT-Synopsys,后者正将重心转向AI辅助设计,旨在运行工程师已在使用的设计工具。并非所有人完全信服。OpenAI芯片负责人Richard Ho上月表示,他仍不完全信任AI设计变更,因此每一项变更都会经过检查。

Next, OpenAI is training GPT-Synopsys with Synopsys, which is refocusing on AI design, to run the design tools engineers already use. Not everyone is fully convinced yet. OpenAI’s own chip chief, Richard Ho, said last month he still does not fully trust AI design changes, so every one is checked.

阅读媒体原文 ↗