5 ms·
Weird how they include an HD 630 on the die. This eats up a fair chunk of die area, and eats into the thermal budget. It also means that the displays will proba
by sspiff 9y ago
Weird how they include an HD 630 on the die. This eats up a fair chunk of die area, and eats into the thermal budget. It also means that the displays will probably be wired to the HD 630, so the Radeon can power down to save power. Taking out the HD 630 could have lowered the power usage to a level more palatable to a laptop computer at full blast.
This means that you have to pay a penalty to use the Radeon to render, but send its framebuffer result back to the HD630 for displaying. This is also the case current Nvidia Optimus or AMD Enduro systems. Draw backs from this approach are frame tear / lack of proper V-Sync and added latency.
If this ends up being used by Apple however, it is likely that a hardware demux will be used to connect the display to both graphics chips, and switch it back and forth on demand. Other vendors don't do this, because it is more expensive, and passing the display from one GPU to another requires the pixel clock rate (among other details) to be shared between the different GPUs, which is not something that the vendor driver (Intel/Nvidia/AMD) stacks support.
Note that such a hardware demux is used in previous generations of MacBook Pros, and it is not publicly documented, which means very poor and/or very late support in Open Source operating systems such as Linux.
Some system builders (Lenovo and Asus have done this, probably others) wire the internal display of laptops to the Intel GPU, and the external HDMI/DP to the discrete graphics. In the past, this allowed you to power more displays (when Intel GPUs were restricted to 2 displays max), and it still allows you to avoid the Enduro/Optimus hassle for external displays, when you are likely to be powering the laptop with the A/C adapter anyway. It also meant that vendor/OEM specific drivers needed to be used in some cases, though.
- DiabloD3 9y agoIn Windows 10 and Linux, this can be nearly for free without any additional hardware. Linux calls this dma-buf, and Windows 10 basically requires it for any WDDM 2.x compliant drivers: GPU to GPU direct DMA mapping without CPU involvement (directly routed over PCI-E and not through the CPU memory controller when possible). Basically, it means Optimus style and hardware-muxer style setups are obsolete, and display framebuffer and render framebuffer are entirely disconnected naturally; and also does not require any overt cross-architecture support (ex: Nvidia Optimus support in Nvidia's driver requires explicit Intel support; dma-buf solutions don't). Presumably, OSX also can leverage this.
- sspiff 9y agoSure, but then you still don't have V-Sync, which causes terrible and easily noticeable tearing on almost anything (at least in the case of Nvidia, it seems to be much better with AMD from my annecdotal testing). I doubt Apple would find tearing acceptable on their products, honestly.
- DiabloD3 9y agoThat isn't related to dma-buf-type usage, though. That would be purely one of the drivers not signalling the end of the frame to the other (since all graphical APIs have suitable methods for this, and WDDM itself natively understands the concept). Also, in Windows 10, the window manager's compositor (equivalent thereof) does all the event signaling fine; ie, you could easily have a situation where a game is rendered on one GPU, displayed on the other, and the display GPU + monitor has VESA Adaptive-sync enabled, and the game correctly triggers new sync windows as if it had been running on the same GPU. Does this work in practice? Who knows, but its supposed to, especially with fixes that 1703 and 1709 added.
- mrec 9y agoIs this intended/expected to work on desktop setups with discrete GPUs too? I believe Optimus et al were always laptop-only, but some desktop users might care about heat/power/fan noise.
- DiabloD3 9y agoIt's expected to work. I've tested it on mine, Haswell i7's own iGPU, and a full sized GCN Radeon, with a game rendering on the Radeon but displaying on the iGPU. Multi-monitor setup with 2 on the Radeon (normally all 3) and 1 on the iGPU, seemed to work fine. No real visible lag either, from what I could tell with limited testing.
- rasz 9y agoWorks on desktop just fine. For example Pentium G3258 integrated GPU forwarding fullhd over VGA rendered on ancient Nvidia 7900GS (old, but still 2-4x faster than integrated intel). Overhead was ~10-30% compared to direct monitor connection to nvidia card.
- husam212 9y agoRIP Linux support :(