To run Time Spy and Night Raid benchmarks, you need Windows 10, a graphics card that supports DirectX 12, and a processor that supports SSSE3.To run Speed Way, you must have Windows 11 or the Windows 10 21H2 update, and a graphics card with at least 6GB VRAM and DirectX 12 Ultimate support.Contact us Get quoteĪll the currently supported benchmarks and feature tests in 3DMark are compatible with Windows 11. My bet is still on AMD edging out nvidia in this one specific scenario, though, but I also don't think it's a deciding feature when comparing the two camps.3DMark Professional Edition Get quote Press license From $1695 per yearĪsk us about Testdriver - our easy benchmark automation solution for enterprise IT, production line QA, and PC lifecycle management.īuy two or more benchmarks to qualify for a bundle discount. We don't have as detailed information on the differences between AMD's ACE scheduler and Nvidia's dynamic scheduler, so again, many of the predictions about future async compute on nvidia and AMD hardware will have a lot of speculation involved. ![]() Not only is it application-dependent, the amount of hardware chip designers throw at the problem (in the case of POWER8, larger fetch buffers and issue queue) also determine how much of a payoff you get from it. In general it's a good idea, but it's going to be heavily application dependent.Īnandtech had an interesting article recently on POWER8 where they look at SMT gains with 2- 4- and 8- threads, and compared it to the hyperthreading gains on Xeon. Yeah, I'm glad that not too many people have compared async computing with hyperthreading, because that's only going to lead to more screwball stories like this one, but it's actually not a bad comparison. Is Time Spy's 20% a lot? A little? Is it a useful tool to predict actual game code use moving forward? Futuremark seems to think they've settled on a good balanced workload, but I expect some developers will be further away on either side of the spectrum depending on their respective aims. If async compute (or more specifically, concurrent execution) becomes a bigger part of game engines moving forward, I think that AMD has a larger potential gain, but for most currrent titles, nvidia seems to be getting more performance out of fewer transistors and less power than AMD with their current architectures. It sounds like nvidia is doing a better job of keeping its execution units fed all the time, so has less to gain than AMD does when async compute is added to the mix. ![]() Throwing compute into the mix doesn’t accomplish anything if the graphics queue can sufficiently saturate the entire GPU. That 20% is still enough to show up in GPU results between NVIDIA and AMD cards.Īnd to quote Anandtech again (from their 1070/1080 review, addressing async compute in general)įor async’s concurrent execution abilities to be beneficial at all, there needs to be idle time bubbles to begin with. It does use DX12 async, just not that much of it. ![]() In the case of async compute, Futuremark is using it to overlap rendering passes, though they do note that "the asynchronous compute workload per frame varies between 10-20%." But in brief, everything from asynchronous compute to resource heaps get used. Futuremark has published an excellent technical guide for the benchmark, which should go live at the same time as this article, so I won't recap it verbatim. That said, Futuremark has definitely set out to make full use of FL 11_0. At the same time it doesn't use any of the features from the newer feature levels, so while it ensures a consistent test between all cards, it doesn't push the very newest graphics features such as conservative rasterization. Under the hood, the engine only makes use of FL 11_0 features, which means it can run on video cards as far back as GeForce GTX 680 and Radeon HD 7970.
0 Comments
Leave a Reply. |
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |