AMD Vega 10, Vega 11, Vega 12 and Vega 20 Rumors and Discussion

Well, I can't argue with that.

 

I got myself a Vega56 at launch at MSRP because I already had a Freesync monitor and switching to an Nvidia card while keeping sync would have been too expensive. Fast forward a few months later, my monitor suddenly got a grey horizontal stripe across the middle of the screen, which is super annoying and It does not go away even If I connect my monitor to another device, using a defferent cable. Now I have two broken pieces of hardware next to me :/

 

Are we sure that those 'missing' features are really missing? What if they do benefit a few games, and they are enabled there in a way similar to shader replacement, and they are disabled by default when they don't.

Nobody ever said that an advertised feature needs to be used at all times. If AMD can point to one relatively popular game where, say, primitive shaders increase performance by 2%, then they've fulfilled their marketing promise in a, IMO, sufficient way to kill potential lawsuits.

The expectations of the upside of all those new features were probably way too high to begin with. A few percent improvement here and there is still valuable in this kind of market.

 

There are not going to be any lawsuits because there were no concrete performance or timeframe claims. How would you prove any sort of harm or breach, since no one knows how these features were supposed to manifest themselves anyway? For all we know, enabling some of them may actually decrease performance (ala async compute on Maxwell). There is no baseline as to what the marketing bullet points were supposed to actually mean, performance-wise.

Now, as soon as you make any sort of official statement as to anything being broken in hardware, you open the liability floodgates, having just admitted your product was defective. That’s why you will NEVER, EVER see any such statement.

Does anyone really think AMD didn’t know exactly what they had back in June? That the whole utterly cringe-inducing string of “blind tests” with disabled FPS counters was because they thought they had a real winner on their hands? That a ~500mm chip with expensive PCB and still-exotic memory was priced at $500 for the lolz, despite being a couple of driver updates away from beating 1080 TI by 20%, as some fellow contributors suggested? It was always in front of everyone from the beginning, and in AMDs defense, they really didn’t pretend otherwise.

 

The last official statement on the matter said they were. Whether or not that has changed since then is anyone's guess.

Maybe, but AMD provided some DSBR figures in the whitepaper. We've seen the difference culling makes from Fiji to Polaris and Vega should have improved that even further with primitive shaders and possibly packed math. We also have some idea of the gains Nvidia made with similar techniques.

In modern titles, we've already seen Vega beat a 1080ti seemingly without the features. Possibly even without the packed math. Approaching 20% from those two, possibly three, features seems almost trivial. Lack of packed math there would be huge considering all the async compute present in recent titles. The real concern with Vega seems to be the geometry bottleneck which both primitive shaders and DSBR should have addressed as Vegas new big features.

I'm still guessing compiler issues are breaking the features and it's not a hardware problem. Some unforeseen issue and most resources directed towards Apple or other contracts. MacPro is launching soon and Apple seems happy at least. Intel apparently showed an interest that they picked up AMD graphics. Also interesting that open Linux drivers we're up to 42% faster than AMDs internal, cross-platform version. It's unclear if Vega's features were even enabled there, just that the focus seems elsewhere. Regardless some communication on the matter would be helpful as even devs would surely like to know if they have a performance boost coming and certain techniques work better. If some techniques need to be implemented a certain way, advance notice would be welcome.

 

First time in history that a linus driver is faster than its equivalent windows version?

Nothing should surprise anyone about vega by now.

 

If the missing features in Windows were enabled too, Vega will catch up on this platform as well ?

 

Have we? The last time we went through this, you brought up Forza 7...

 

What is geometry boost at wx9100???

 

That was the most recent title to show Vega leading at the time and based on the newer APIs. Also have Wolfenstein to add to the list now and there are other titles coming. And yes without all the features apparently functioning. Not at all surprising with Vega having more texturing power, async support, and FP16 for added flops. Just consider that Vega isn't all that far from Titan V which just released. So yes we have seen it happen and it was predictable.

 

Uh No!



http://techreport.com/review/32766/nvidia-geforce-gtx-1070-ti-graphics-card-reviewed/3



https://www.hardocp.com/article/2017/10/16/forza_motorsport_7_video_card_performance_update/3

Vega exceeds 1080Ti at 1080p in that title, At 4K it's significantly behind.

 

Bullsh*t
It was with one single title, at one single resolution, by a single reviewer, and it was fixed by nvidia driver update.
Vega will never beat consistently 1080Ti. No now, not in 6 months, not in a year. Thinking otherwise is being disconnected from reality...

 

Wolfenstein also doesn't utilize the FP16 added flops you mentioned and touted by AMD and Bethesda on the Vega.

 

Just to refresh your memory, once Nvidia updated the drivers the so called-lead has evaporated

The grand total of zero titles list? 1080 TI actually outperforms Vega in Wolfenstein, which is a little sad seeing how the latter is a literal poster child for the former. And...what other titles coming?

We haven’t and it wasn’t. And, it won’t. I mean, that’s the literal reality. Such a strange hill to die on.

 

People put pressure on AMD, pleas read top quotes of this subreddit:


There are also 2 other subreddit blaming AMD because of primitive shaders. Each have nearby 400 upvotes

 

The concept of top quotes changes fairly frequently for an active reddit. If you want to refer to something in particular, you'll have to quote the specific text.

 

I think the primitive shaders are uncatchable on top

 

The top one for me is about Wattman fan curves. All I'm saying is quoting context within the forum is advisable when linking to something that is so transient.

 

you sortet best not top