SoylentNews is people
SoylentNews
Intel Rocket Lake Desktop CPUs Will Launch in March, Gigabyte Confirms - ExtremeTech:
Gigabyte has confirmed that Intel will launch its Rocket Lake CPU refresh in March, as part of an announcement touting its own PCIe 4.0 support. Gigabyte announced today that if you own a Z490 motherboard, you'll be getting a UEFI update to support Rocket Lake CPUs with full PCIe 4.0 support.
The rest of the PR goes into detail on how Gigabyte engineered their motherboards to handle the higher heat output of PCIe 4.0, and the fact that addressable BAR support is coming to the company's motherboards as well. Addressable BAR is the same feature AMD debuted as Smart Access Memory earlier this year.
The March 2021 date confirms what we've heard previously — late March is more likely than early March. It's going to be genuinely interesting to see how Cypress Cove performs against AMD's Zen 3. Generally speaking, based on leaked benchmarks and early data, we're looking at impressive gains for Intel in single-thread performance. Multi-thread performance estimates for the Core i9-11900K have varied. In some cases, the 11900K is almost a match for the 10-core Core i9-10900K. In a few leaked results, it's actually been faster on eight cores than Comet Lake was on 10.
Are any of my fellow Soylentils doing PC builds right now, and if so what are you building? Let us know in the comments!
takyon writes: Intel announced more details about Rocket Lake at CES 2021. While dropping the top core count from 10 to 8, Intel estimates a 19% IPC increase for Rocket Lake-S. It also adds AVX-512 and "Deep Learning Boost" support. The integrated graphics should be about 50% faster, and can be used for stream encoding while discrete graphics is being used for gaming. AV1 video decode is supported. New Z590, B560, and H510 motherboards will support both Rocket Lake and Comet Lake. Intel's comparison of the 8-core i9-11900K to AMD's 12-core Ryzen 9 5900X shows the former performing 2-8% faster at several games at 1080p.
Also at Tom's Hardware and Wccftech.
(Score: 5, Interesting) by fakefuck39 on Thursday January 14 2021, @05:38PM (1 child)
I wonder why more applications don't use both graphics cards - the discrete GPU for the actual graphics, and the integrated one for some kind of vector computation. Nvidia has that Physx thing, and dx12 supports using two different brand cards at the same time. But I don't see any actual applications using this.
Since I'm not a gamer, here's a good example: business meetings. I'm on a conference call with 20 people, seeing 20 streams, sending my own. There's compression, decompression, noise reduction for audio, etc. I'm also running a computationally intensive script in the background and drawing up a visio with high-res graphics files. It's all using the same GPU and CPU, the integrated graphics card is idle. Why isn't "the computer" offloading some of that work onto the crappy intel card in my laptop, resulting in faster less jittery experience for my visio usage? I'm sure that underpowered GPU can handle at least encoding and decoding the video in my webex meeting, and because this is not exactly a memory-bandwidth restricted process, there shouldn't be a bandwidth bottleneck between the two cards.
I think the reason this isn't done, is the way dx12 has it set up, it's still up to the application to handle this. What we need is kind of like the thread model in Solaris - you get a bunch of user threads, and they're mapped to real threads by the OS. We do it for CPU, I don't see why not for GPU. If it's left up to each application, it's probably not happening.
(Score: 2) by hendrikboom on Sunday January 17 2021, @04:37AM
I think the hardware initialisation is such that it configures itself to make only one video card available so that the poor OS that it's about to boot doesn't get confused.
-- hendrik