what is so difficult to understand? Lack of competition => high prices, slow innovation.
Vega RX - too little, too late, too much power draw.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
The Vega RX Thread (Rumors and Info)
- Thread starter fendersrule
- Start date
- Sort by reaction score
what is so difficult to understand? Lack of competition => high prices, slow innovation.
Vega RX - too little, too late, too much power draw.
Me? ... For cMP I'm still happy with 2 x 7970
(The Prince134 "King of Mac" build)
The Vega is low power draw in comparison.
I may build a threadripper system with dual Vegas and try 'em in the Mac.
I just don't understand how the rest applies to this topic or to me.
Last edited:
Wolfenstein II uses FP16:
https://www.golem.de/news/id-software-wolfenstein-2-unterstuetzt-fp16-und-vulkan-1707-129201.html
https://www.golem.de/news/id-software-wolfenstein-2-unterstuetzt-fp16-und-vulkan-1707-129201.html
You have said, that their disappointment will lift in 6 months.
There is a certain very vocal group here that is NOT disappointed with performance of Vega, they are happy about it. You should've learned that by this moment if you read their posts about AMD over past years on this forum.
On the other hand. Im not sure it will take 6 months to improve performance of Vega.
Videocardz.com has a article where 'AMD Radeon Vega graphics cards explained', in it the AMD Radeon Pro 56 and AMD Radeon Pro 64 are listed. I'm assuming that these are the two that will be going in the iMac Pro. There is however no mention of their power usage at the moment.
According to Apple, that will be a 11 TFLOPS GPU inside the iMac Pro. And now, even the Vega 56 (10.5 TFLOPS) already cost 210W to achieve.
ATM, it looks like they will downvolt and downclock the Vega 64 to fit inside the iMac Pro. However, I don't think it's possible to cut down 50% of power (still ~150W), and only sacrifice 10% performance. If it's possible, then IMO, AMD is really stupid to marketing the Vega 64 as a 295W GPU.
ATM, it looks like they will downvolt and downclock the Vega 64 to fit inside the iMac Pro. However, I don't think it's possible to cut down 50% of power (still ~150W), and only sacrifice 10% performance. If it's possible, then IMO, AMD is really stupid to marketing the Vega 64 as a 295W GPU.
The GPU in iMac Pro has lower voltage and lower core clock @1.35 GHz, so the power draw will be lower.
Vega Nano has 1.2 GHz... At least Engineering Samples have had that clock...
If so then proof again AMD just wants to **** with consumers with their big power hungry component heavy models.
On the plus side for Vega, if High Sierra has plug and play support for all these cards you can use one now in your cMP and then install it in a future MP 7.1
Unless Apple keeps producing more crippled or proprietary bull crap.
Stacc, wrong link.
https://www.techpowerup.com/reviews/AMD/Vega_Microarchitecture_Technical_Overview/6.html
Whole presentation is giving a lot of information, on Vega architecture.
AMD blatantly is saying: we have released unfinished product, expect that Drivers might bring performance gains, but those are also relying on developers:
https://www.techpowerup.com/reviews/AMD/Vega_Microarchitecture_Technical_Overview/4.html
This pretty much explains why Vega is performing as it is in current state of software.
I think one of first games that will adopt Vega features will be Overwatch. AMD started very close collaboration lately with Blizzard, on their games, and improved performance quite a lot, lately.
Within 2 months Overwatch went on RX 460 from 70 FPS in 1080p Ultra preset to 84 FPS average, and RX 560, broke 106 FPS average framerate, and came close to GTX 1050 Ti.
"Use local video memory as a last level cache for system memory and storage".
That is essentially how HBCC works. Of course - you still need API, and the connection will go over PCIe to the GPU, instead - directly to the GPU.
https://www.techpowerup.com/reviews/AMD/Vega_Microarchitecture_Technical_Overview/6.html
Whole presentation is giving a lot of information, on Vega architecture.
AMD blatantly is saying: we have released unfinished product, expect that Drivers might bring performance gains, but those are also relying on developers:
https://www.techpowerup.com/reviews/AMD/Vega_Microarchitecture_Technical_Overview/4.html
This pretty much explains why Vega is performing as it is in current state of software.
I think one of first games that will adopt Vega features will be Overwatch. AMD started very close collaboration lately with Blizzard, on their games, and improved performance quite a lot, lately.
Within 2 months Overwatch went on RX 460 from 70 FPS in 1080p Ultra preset to 84 FPS average, and RX 560, broke 106 FPS average framerate, and came close to GTX 1050 Ti.
You do not have to rely on the SSG GPUs.So... Jarred Land (president of RED Digital Cinema) posted on his FB page that AMD gave him an alpha version of a Vega-based Pro 2TB SSG GPU to test out. He compares it with the TITAN Xp. It's an insane beast. I hope Apple uses this bad boy, if they stick with AMD! Check it out:
"Use local video memory as a last level cache for system memory and storage".
That is essentially how HBCC works. Of course - you still need API, and the connection will go over PCIe to the GPU, instead - directly to the GPU.
Last edited:
Yes, I'm interested in how spectacularly ATI managed to miss performance targets while consuming more power than most systems can provide.
And if they prematurely introduced something before the drivers and software were ready - I'm curious as to why.
Here are 1000 words to describe the Vega release:
If so, Im pretty sure this post will be interesting for you:
https://forums.macrumors.com/threads/the-vega-rx-thread-rumors-and-info.2056361/page-2#post-24846631
I have posted previously: in perfect world, with properly optimized software, 1.6 GHz, 512 GB/s Vega should be two times faster than Fiji.
So far, it appears that Vega with the same memory bandwidth, and 45% higher clock speed it is up to 25% faster than Fiji. So definitely something is holding the architecture up. But also definitely - its not the hardware. Im sure that after reading my post describing Vega architecture, and comparing it to the technical details AMD released and Techpowerup reposted you would understand - why I am saying this.
P.S. What performance targets you say that AMD missed...?
Driver support only may not help this architecture. The applications have to reworked to utilize some of Vega arch features(FP16 for example is a must, the same goes for Primitive Shaders, but this feature also relies on driver support, and its Vulkan and DX12 only).
So don't expect that Vega will magically happen to be great in DX11 games. Vulkan and DX12 - that is completely different story.
Im sure they were talking about performance per watt from Vega Nano.
P.S. Gaming is different story about efficiency, and professional workloads are different story. I thought this is professional forum, but once again, we are only talking about gaming performance.
What makes you think this is a "professional" forum? It's to discuss the Mac Pro, which is a range of computers sold by Apple (that used to have a PCIe slot and could use a wide variety of graphics cards, official or otherwise). People use Mac Pros for all kinds of things, including playing games.
Links?
I don't remember that, and I have dozens of them. Nvidia's been shipping drivers and CUDA ready for the latest GPUSs for quite some time.
We got pretty much immediate speedups from upgrading - just upgrade CUDA to the new version that supports Maxwell/Pascal and things are faster immediately. (Note that my focus is compute-only - no games, and no 3D graphics.)
If you have apps that don't use the new features - no speedups until the apps are updated. That can be fixed by a strong 3rd party early adopter program to help apps be ready on launch date. Nvidia mostly seems to have that - CUDA 8 was out when Pascal shipped for real. If your apps use the CUDA libraries - they're mostly ready on day 1. (Many GPU advances enhance existing APIs without code changes. Using new APIs do need some app code changes.)
CUDA 9 with Volta support is towards the end of Beta, and with GV100-based cards showing up should soon CUDA 9 be public.
ATI should be embarrassed for coming out with a 345 watt liquid-cooled card - and saying that "the software/drivers aren't ready".
[doublepost=1501533965][/doublepost]
You should apply to be the next Trump White House communications director. Your belief in alternative facts makes you a shoe-in.
You keep moving the goal posts here. First, graphics workloads are professional. Apple is advertising the iMac Pro for VR, featuring a Vega graphics chip. There are many other "professional" workloads that stress graphics performance.
Second, Vega FE/Vega RX 64 with a TDP of 295 W with GTX 1080 graphics performance is pathetic efficiency wise. Sure, a downclocked chip may slightly improve this. But even if there is zero performance loss between the 295 W RX 64 and the 210 W Vega Nano, it will still be less efficient than the 180 W GTX 1080.