Interesting, I'll update you once I get in touch with the tester.So, what Dx11 has to do with bad geometry throughput and general front-end performance, that as such limits access to vast overkill of ALU array GCN cards boast? I guess that's the time the old saying about bad dancer and his balls applies.
https://www.computerbase.de/2017-03.../4/#diagramm-rise-of-the-tomb-raider-dx12-fps
https://www.computerbase.de/2017-03...4/#diagramm-ashes-of-the-singularity-dx12-fps
http://www.guru3d.com/articles_pages/amd_ryzen_7_1800x_processor_review,16.html
http://www.guru3d.com/articles_pages/amd_ryzen_7_1800x_processor_review,17.html
http://www.guru3d.com/articles_pages/amd_ryzen_7_1800x_processor_review,19.html
Watch this and tell me why GCN is flawed.No, it is a logical consequence of GCN flaws, nothing else. But i digress, what business nV has hurting it's consumer base? It's not exactly in competition with Ryzen, y'know.
You do understand that within Dx11 this video agrees with me and brings even more evidence to it than i meant to.Watch this and tell me why GCN is flawed.
Uhm, isn't that obvious, considering we know that even mobo makers did not have too much time with Ryzen before release.
You do understand that within Dx11 this video agrees with me and brings even more evidence to it than i meant to.
Making draw calls multithreaded on DX11 in a way that is developer and game-engine agnostic was only possible due to the SW scheduler in the NVIDIA driver. With GCN, AMD hoped that developers would code their games such that the draw call thread is kept as free as possible so it can efficiently feed the ALUs via the command processor, ie. its hardware scheduler. Like Civilization V. But games continued to load the draw call thread with other tasks(this is the reason why Project Cars performed so poorly on GCN) and until recently, there haven't been many DX11 games that had the incentive to do it properly, because the NVIDIA driver took care of that.You do understand that within Dx11 this video agrees with me and brings even more evidence to it than i meant to.
Within Dx12 framework it gets more complicated but you understand as well as i do that making drawcall submission multithreaded is not exactly trivial even when it is possible on API level.
This exact reason, for the above, is why you NEED very fast CPU, rather than slightly slower, but with higher number of cores, in most game engines that are today. But brace yourselves. Shader Model 6.0 will change this.However, the SW scheduler of the NVIDIA driver has its own overhead, and this shows in comparatively recent DX11 games like COD:BLOPS 3 and Witcher 3 where it causes near-max CPU utilization across multiple cores in CPU-limited scenarios. This happens in Titanfall 2 as well, which the video doesn't show. Incidentally, the RX 480 is faster than the GTX 1060 in Titanfall 2. This is also carried over to DX12 where NVIDIA GPUs take a hit regardless of what CPUs are being used. The only way to overcome this, as the video points out, is when developers take their time to code in a way that prevents driver bottlenecks(which are due to features that were originally intended to help DX11 draw call multi-threading, but are redundant on DX12) on NVIDIA cards, examples of this being TW:W and Doom.
Yes, i have seen that.Making draw calls multithreaded on DX11 in a way that is developer and game-engine agnostic was only possible due to the SW scheduler in the NVIDIA driver.
Yes, i have seen that too.With GCN, AMD hoped that developers would code their games such that the draw call thread is kept as free as possible so it can efficiently feed the ALUs via the command processor, ie. its hardware scheduler.
Yes, i have seen even that.. Though it has to be noted that such near-max CPU utilization is already present there, nV driver just makes it worse in this case.However, the SW scheduler of the NVIDIA driver has its own overhead, and this shows in comparatively recent DX11 games like COD:BLOPS 3 and Witcher 3 where it causes near-max CPU utilization across multiple cores in CPU-limited scenarios.
*After a minute of checking* Technically correct, but i fail to see relevancy since it is hardly CPU heavy in the first place.Incidentally, the RX 480 is faster than the GTX 1060 in Titanfall 2.
Go on... I mean, AMD GPUs take a hit on regular basis too, but it is a fair statement that nV GPUs do so way more often.This is also carried over to DX12 where NVIDIA GPUs take a hit regardless of what CPUs are being used.
Yes, it is kind of twisted mirror of GCN Dx11 situation. Make no mistake, i have no issues admitting that nV's approach is flawed too, if that's your point.The only way to overcome this, as the video points out, is when developers take their time to code in a way that prevents driver bottlenecks
Wait, i listened to entire video, but that's one thing i did miss, did he mention it at some point or is that your addition?examples of this being TW:W and Doom.
It does because when i claimed GCN is flawed i meant the implementation not the very idea. A video highlighting the design flaw of GCN in certain situations is addition i totally did not have in mind, but granted it does more to explaining why badly made games run worse on AMD hardware than i ever hoped to, too bad i have seen it after writing my original post.This is basically the difference between AMD and NVIDIA software and hardware implementation that the video talks about, and if you claim that it only strengthens your claim that GCN is flawed, I suggest you rewatch it.
And here i am thinking it is because games are ultimately ALWAYS bound to a single threaded performance, or rather, performance of an event loop thread.This exact reason, for the above, is why you NEED very fast CPU, rather than slightly slower, but with higher number of cores, in most game engines that are today. But brace yourselves. Shader Model 6.0 will change this.
Heavily modified Source engine, CPU plays a part in it.*After a minute of checking* Technically correct, but i fail to see relevancy since it is hardly CPU heavy in the first place.
It is there in the 14:12 mark. Creative Assembly and the TW series is mentioned before that.Wait, i listened to entire video, but that's one thing i did miss, did he mention it at some point or is that your addition?
Well dang. That pretty much explains everything doesn't it. Well except why Nvidia stops scaling on DX12 and why it might be only recently. But that would explain just about everything. Why Radeon's are so consistently better at DX12. Why Nvidia sees near consistent performance hit on DX12. So than the real question again is why does it seem like the Nvidia driver caps the MT scheduling at 4c. Probably goes back to what I was thinking. They thought they could eek out better performance on the their cards by doing that considering the performance of the 7700k.Watch this and tell me why GCN is flawed.
Fair, i guess, though it is still hardly CPU heavy from what i see in benchmarks and stuff.Heavily modified Source engine, CPU plays a part in it.
Oh yeah, i noticed it now, but i believe TW is mentioned in different context still (TW is mentioned in context of single threaded games i believe).It is there in the 14:12 mark. Creative Assembly and the TW series is mentioned before that.
Very enlightening video. Explains lots of things and lolfail9001 got more gas in his fire to continue his crusade.Watch this and tell me why GCN is flawed.
Set P0 to what you desire, keep the rest unchanged. Use High Performance power profile and set the Minimum Processor State to a low value and report back.That's what I followed. It will not let P0 activate. P1 is as high as it goes.
Set P0 to what you desire, keep the rest unchanged. Use High Performance power profile and set the Minimum Processor State to a low value and report back.
Watch this and tell me why GCN is flawed.
https://msdn.microsoft.com/en-us/library/windows/desktop/dn899217(v=vs.85).aspxIn D3D12 the concept of a command queue is the API representation of a roughly serial sequence of work submitted by the application. Barriers and other techniques allow this work to be executed in a pipeline or out of order, but the application only sees a single completion timeline. This corresponds to the immediate context in D3D11.
And here we are, talking about DX12 in a CPU related thread.We went over this nonsense last year in the GPU section. This whole video is just a bunch of nonsense which doesnt make any sense and doesnt reflect the reality.
Straight from Microsoft:
https://msdn.microsoft.com/en-us/library/windows/desktop/dn899217(v=vs.85).aspx
Multi-Threaded Rendering is an API concept. Work gets only submitted through one command queue. You can fill the command queue from multiple threads. After the GPU got the queue the hardware is scheduling the workload on the GPU. The application can create multiple command queues. The number depends on the hardware.
particular, the following scenarios can be addressed with D3D12:
- Asynchronous and low priority GPU work. This enables concurrent execution of low priority GPU work and atomic operations that enable one GPU thread to consume the results of another unsynchronized thread without blocking.
- High priority compute work. With background compute it is possible to interrupt 3D rendering to do a small amount of high priority compute work. The results of this work can be obtained early for additional processing on the CPU.
- Background compute work. A separate low priority queue for compute workloads allows an application to utilize spare GPU cycles to perform background computation without negative impact on the primary rendering (or other) tasks. Background tasks may include decompression of resources or updating simulations or acceleration structures. Background tasks should be synchronized on the CPU infrequently (approximately once per frame) to avoid stalling or slowing foreground work.
- Streaming and uploading data. A separate copy queue replaces the D3D11 concepts of initial data and updating resources. Although the application is responsible for more details in the D3D12 model, this responsibility comes with power. The application can control how much system memory is devoted to buffering upload data. The app can choose when and how (CPU vs GPU, blocking vs non-blocking) to synchronize, and can track progress and control the amount of queued work.
- Increased parallelism. Applications can use deeper queues for background workloads (e.g. video decode) when they have separate queues for foreground work
No, it is a logical consequence of GCN flaws, nothing else. But i digress, what business nV has hurting it's consumer base? It's not exactly in competition with Ryzen, y'know.
The video states that command can be issued from multiple workers. With nvidia the driver handles the scheduling while in AMD the hardware does it. Hence amd has lower overhead.
The video then goes on to say due to this, NV will load up the main worker with scheduling task hence leaving less room for other things like player management, UDP calls, etc. Perfectly logical in reasoning assuming the developers are not interest in accounting lower IPC ST cpu due to NGW or otherwise, I.e. amd's platform.
The nVidia driver doesnt handle the scheduling. There is no difference to AMD. "inter warp scheduling" has nothing to do with it.
And this wrong, too. nVidia creates helper threads under DX11/OpenGL to support the main render thread - aka offloading workload to other threads.
Instead of listening to people on youtube you could watch GDC videos from nVidia about DX11 and DX12: http://www.gdcvault.com/play/1023517/Advanced-Rendering-with-DirectX-11
And here is one for OpenGL and Vulkan: http://www.gdcvault.com/play/1023516/High-performance-Low-Overhead-Rendering
Maybe you guys can now go back to Ryzen. This is so more interesting than this nonsense about nVidia, AMD, DX11 and DX12. The last two pages are really offtopic.
eurocom said:Eurocom is launching the most powerful laptop ever created. The 19.1” EUROCOM Tornado F9 is fully configurable and can support dual AMD Ryzen 7 series processors, dual SLI NVIDIA GeForce GTX 1080Ti with 16GB GDDR5X VRAM or AMD Radeon RX580 dual CrossFireX desktop graphics setup, eight memory slots for up to 128 GB DDR4 memory and six M.2 NVMe solid-state drives.