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

Vegas geometry performance is not week. You can see here: (must select cards and settings in the drop down menue)
http://www.pcgameshardware.de/Radeo...3/Tests/Benchmark-Preis-Release-1235445/3/#a2

You see on Polygoneoutput Vega beats up Titan X Pascal. And Titan X Pascal have 6 Rasterizer, AMD only 4.

Only when 100% culling is used, than Vega falls behind. That's the key where Primitive shader can help.
I've seen those benchmarks before... I had forgot about them. You're right in that primitive shader's will help with the 100% culling and it should help with 50% as well.
I am not particularly familiar with that benchmark suite although I have seen it before but I don't know enough details. But I question the validity of micro benchmarks. I'm wondering if in the polygon test the output of the vertex shader is thin and doesn't fill the parameter cache in the case of AMD. (I think nvidia purely uses the L1 and L2 for parameters) This would give AMD an advantage with regards to feeding the rasterizer. The interesting thing is that Vega 64 keeps up with the Pascal Titan X in some of the benchmarks but this doesn't translate to real world (game) performance.

But MDolenc is right, I don't know what I was expecting with regard to primitive shaders maybe its just hope for AMD being more competitive.
 
Lets see what future brings. I'm disapointed about the journalists. They can't get any Information about primitve shader from AMD
 
I wonder where did the Vega custom cards go?
Because early September was promised and yet its late October now and there's no custom cards in sight anywhere.
AMD?
 
My guess is, there is not a lot of money to be made of custom Vega (because it's already pretty costly, so they can't really chard more and sell them...), so, nobody push custom cards out of the doors, except Asus...
 
My guess is, there is not a lot of money to be made of custom Vega (because it's already pretty costly, so they can't really chard more and sell them...), so, nobody push custom cards out of the doors, except Asus...
Even XFX/Sapphire?
Let's not be silly for a minute.
They are already making absolutely overkill reference; nothing stops them from slapping a custom cooler on it.
 
I'm disapointed about the journalists. They can't get any Information about primitve shader from AMD
After AMD totally took the piss out of journalists with the Vega RX launch, I think we have to conclude AMD doesn't give a damn what the press writes or how well informed gamers are when they make a purchase.
 
After AMD totally took the piss out of journalists with the Vega RX launch, I think we have to conclude AMD doesn't give a damn what the press writes or how well informed gamers are when they make a purchase.
But than you have 2 opttertunitys:
1. Press don't inform customers very well about These Topics
2. Customers are to stupid to understeand whats going on.
For my self i didn't reed any article about the broken Features the last month. For my oppinion press informs me very bad.
 
Even though the meta discussion is off-topic, it's pretty laughable to condemn online-press when half the internetzes runs around with adblockers, basically denying online-press their wage. Who want's to work for nothing in return? Glory does not pay your rent.
 
Even though the meta discussion is off-topic, it's pretty laughable to condemn online-press when half the internetzes runs around with adblockers, basically denying online-press their wage. Who want's to work for nothing in return? Glory does not pay your rent.

The historic publication to explain these details was the "Fachzeitschrift", and they aimed at people which can understand what it is about by education and experience. I'm sure they still sell on paper (and internet) and take money for access and survive. The error to me seems that the general internet press tries to simplify in a rather brutal lossy way, either because they don't understand the issue themselfs, or because it's simply not simplifyable.
 
But than you have 2 opttertunitys:
1. Press don't inform customers very well about These Topics
2. Customers are to stupid to understeand whats going on.
For my self i didn't reed any article about the broken Features the last month. For my oppinion press informs me very bad.

I think we have a very classic example where the above argument breaks down.
Look at how the press followed up on the 970 4GB debacle, plenty of good articles that never let this lie and quite a lot of customers irritated enough that there was the lawsuit.
It did still sell well, but then at the time it did not unfortunately have much in way of comparable competition when taking into account various factors such as performance/efficiency/noise-cooling.
 
Last edited:
The historic publication to explain these details was the "Fachzeitschrift", and they aimed at people which can understand what it is about by education and experience. I'm sure they still sell on paper (and internet) and take money for access and survive. The error to me seems that the general internet press tries to simplify in a rather brutal lossy way, either because they don't understand the issue themselfs, or because it's simply not simplifyable.
There's more to it I guess. Even Fachzeitschriften nowadays are feeling the need to adress larger audiences. And the online-press is under heavy pressure time-wise. They feel they absolutely need to get the article out at launch and after that there's the next review waiting already. It's not helping that bascially everyone takes „free“ work for granted.
 
There's more to it I guess. Even Fachzeitschriften nowadays are feeling the need to adress larger audiences. And the online-press is under heavy pressure time-wise. They feel they absolutely need to get the article out at launch and after that there's the next review waiting already. It's not helping that bascially everyone takes „free“ work for granted.
But where is the Quality? The best mess you see at Wolfenstein. Only one site reportet that there is an issue with one Special scene. But no fouther Testing whats the Issue! Other sides didn't even stated that there is an issue. Is this Quality?

I did some testing about my self with this issue, thats what i get:


Realy strange. If you look more in the sky the Frame get worst by 10%?
And whats also strange, thise Scene is cpu bounded. I started a recording tool i get 70fps without the tool i had 90-100 fps here, but the same behavior of the 10% . Anyone can say where the Problem is?
 
Last edited:
There's more to it I guess. Even Fachzeitschriften nowadays are feeling the need to adress larger audiences. And the online-press is under heavy pressure time-wise. They feel they absolutely need to get the article out at launch and after that there's the next review waiting already. It's not helping that bascially everyone takes „free“ work for granted.

You are generalizing. In the engineering space I have not observed any out-of-its-own-space fishing. We have a few of them coming in every month or quarter or bianually and circulating the office. That doesn't mean there are none doing, just that there are some not doing it. There are more engineers than ever, no need to diversify, would be my guess. I also didn't observe rushing with stories, they are published when the author feels he covered all (or the relevant for the article) bits, which means sometimes half a year of latency (not because of publishing frequency).

I know there is the more interesting question lingering below the meta-discussion: Are number of CUs a good markting instrument for consumers? If yes, what do you need to do in your articles to make everyone understand what exactly it is and what it means? If not, what do you have to do to break the vicious circle of marketing departments obsession over "number races" and "checkbox accumulations"?
I'd prefer to stay away from any marketing effords of the current IHVs (as a writer, hypothetically), there is not much innovation going on, there is no justification to write an in-depth piece for every release, there is not enough information to make a piece correct.

Then there is the question of using games for benchmarking, which I think is very questionable. Because you have no chance to decompose the games software effect from the hardware effect, and tend to attribute everything to hardware, until contradictions appear, and then the attribution is simply like sampling an RNG.
For as long as games are not looked at with RGA or Nsight or RenderDoc I don't see conformation for an attempt to even be serious about analyzing the situation and trying to find answers. Or for as long as a writer doesn't blame the engine first and foremost for observed performance (which requires skill and knowledge or good contacts, or guts).

I don't believe in big picture statistics (smash all games benchmarks in a statistic and pretend that it's a good or valid thing to do). Not about age, or gender, or populations or graphics cards. No thing is the best for all, put the pretention exists, and companies are living off it.

Lastly, every publication chooses it's audience, if the audience is known to be volatile, you can't blame the audiance for it's attitude. Choose your audience more wisely. You can always change the publication you work for (in principle), just pick one with an audience you'd prefer to work for. :)
 
it's pretty laughable to condemn online-press when half the internetzes runs around with adblockers, basically denying online-press their wage.
I can't testify as to what other people do and think, but I for one like having a free and open internet, and thus I don't run adblockers.

Then again, I don't have a capped internet connection, so it doesn't bother me that some websites (undoubtedly desperate for more ad revenue and clixxxzzz) run absolutely crazy insane javascript bullshit that reloads like six or eight or maybe even ten adbanners every few minutes and can snarfle hundreds of megabytes of data per hour per page.

These javascript monstrosities also tend to be unstable and extremely resource intensive from a browser point of view; keep a page open for an extended time period and the browser might eat vast quantities of RAM, and/or stop responding to almost any input on that page, including scrolling and F5/reload.

So I don't blame people who adblock TOO much either. For example. it's super hard and annoying to post on a forum where it takes several seconds for any text to show up because of a damn ad-flogging javascript going berzerk as you're typing.
 
Some adds plainly insult our inteligence, see Anandtech .
Absolutely. Online ads are a double-edged sword. (Still better than broadcast ads I'd say, mainly because 99.99% of the time they STFU rather than play an idiotic song to me about how awesome some bullshit I'm not the least interested in is...)
 
You are generalizing. In the engineering space I have not observed any out-of-its-own-space fishing. We have a few of them coming in every month or quarter or bianually and circulating the office. That doesn't mean there are none doing, just that there are some not doing it. There are more engineers than ever, no need to diversify, would be my guess. I also didn't observe rushing with stories......

The numbers game has always been for chumps as far as marketing goes. Apple realized it quickly enough, their SOCs have often been much faster than the competition, but things like number of cores would have suggested otherwise. So they just ditched the whole thing. As did video games. I still remember a decade ago when video game marketing liked "numbers" a lot, and Peter Molyneux was mocking the entire idea by stating Fable 2 had 20k bluebell flowers or some such.

The average consumer just cares what the price is versus how fast it makes their game go. Even the average coin miner only cares about the hash per second rate versus power draw. That's still nothing necessarily to do with CU number. I'm slightly impressed Nvidia got enough people onto stupid hill by making performance/tdp a "thing" for even a handful of gamers. Of course in the end it doesn't really matter other than what your power supply is rated for. The cost of even doubling the watts you draw for a gamer is negligible. But if it works, even a little and temporarily, it works. Still, AMD could counter it by getting out of the "numbers!" game all together.
 
Back
Top