TE
科技回声
首页24小时热榜最新最佳问答展示工作
GitHubTwitter
首页

科技回声

基于 Next.js 构建的科技新闻平台,提供全球科技新闻和讨论内容。

GitHubTwitter

首页

首页最新最佳问答展示工作

资源链接

HackerNews API原版 HackerNewsNext.js

© 2025 科技回声. 版权所有。

GpuScan and SSD-To-GPU Direct DMA

197 点作者 matsuu超过 8 年前

11 条评论

exDM69超过 8 年前
There is no explanation how it works. Does it work on top of existing APIs in user space? Or is there a custom kernel driver bypassing user space?<p>I&#x27;ve done some high throughput streaming from HD&#x2F;SSD to GPU before, and it&#x27;s pretty easy to beat the naive solution but getting the most out of it would require kernel space code.<p>I was doing random access streaming of textures using memory mapped files for input and copying to persistent&#x2F;coherent mapped pixel buffers on the CPU with memcpy with background threads. This was intended to take advantage of the buffer caches (works great when a page is reused) and intended for random access. If I would have been working on a sequential&#x2F;full file upload, my solution would be entirely different.<p>Edit: here&#x27;s the source: <a href="https:&#x2F;&#x2F;github.com&#x2F;kaigai&#x2F;ssd2gpu" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;kaigai&#x2F;ssd2gpu</a><p>It has a custom kernel module.
评论 #12525029 未加载
zokier超过 8 年前
This is very interesting in the light of recent AMD announcement of their &quot;Solid State Graphics&quot;, ie GPU with SSD ducktaped on: <a href="http:&#x2F;&#x2F;www.anandtech.com&#x2F;show&#x2F;10518&#x2F;amd-announces-radeon-pro-ssg-fiji-with-m2-ssds-onboard" rel="nofollow">http:&#x2F;&#x2F;www.anandtech.com&#x2F;show&#x2F;10518&#x2F;amd-announces-radeon-pro...</a>
foobar2020超过 8 年前
This would be incredibly useful for distributed machine learning - imagine a Tensorflow implementation that almost entirely bypasses CPU.
评论 #12526294 未加载
评论 #12525297 未加载
witty_username超过 8 年前
So, if I understand correctly, data is being loaded directly from the SSD to the GPU and then filtered by the GPU before the CPU handles the more difficult queries.<p>Neat.
justinclift超过 8 年前
This is very awesome. If further developed + made into a feasible option for PostgreSQL, this has potential to do interesting things to TPC benchmarks. :)
nl超过 8 年前
See also <a href="https:&#x2F;&#x2F;developer.nvidia.com&#x2F;gpudirect" rel="nofollow">https:&#x2F;&#x2F;developer.nvidia.com&#x2F;gpudirect</a> and to some extent <a href="https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;NVLink" rel="nofollow">https:&#x2F;&#x2F;en.wikipedia.org&#x2F;wiki&#x2F;NVLink</a>.<p>NVLink is in the Power9 servers Google is using.
评论 #12524992 未加载
评论 #12526657 未加载
carbocation超过 8 年前
I&#x27;m really hoping that Optane delivers on the hype, in which case our durable storage could be just 10x slower than RAM. At least, I imagine that it would be really helpful for speeding up even this approach.
Razengan超过 8 年前
I hope this brings us closer to widespread external GPUs, where you could use a slower-than-PCIe bus like Thunderbolt 3 or USB 3.1 to upload all assets to the EGPU&#x27;s SSD during a one-time loading screen.
foobarbecue超过 8 年前
Direct Direct Memory Access? That&#x27;s pretty direct.
评论 #12526254 未加载
musha68k超过 8 年前
Amazing results! We need more of that kind of thinking - GPU&#x2F;SSD accelerate all the things!
MrBuddyCasino超过 8 年前
Who is providing the DMA engine in this case? Has the GPU access to PCIe device memory?
评论 #12525015 未加载