AMD "Vega 10" Bears Core-Config Similarities to "Fiji"

[TheoneandonlyMrK]

View attachment 87453

For once im fighting back English dictionary stlyey,

• the bare bones
  
  
  • 1The basic facts about something, without any detail.
    
    ‘the bare bones of the plot’
• • bare one's soul
    
    
    • Reveal one's innermost secrets and feelings to someone.
      
      ‘one feels vulnerable in baring one's soul to another’

 

[RejZoR]

Big Vega was said to have almost the same die size as Fiji, but on 14nm instead of 28nm. AND some are saying it will have the same shader count as Fiji???

Either this is bs or AMD actually comes out with a new pipeline micro architecture (which is very unlikely).

Sorry, but have you missed literally all the leaks and presentations so far? Vega is probably the biggest change to the rendering engine Radeons had in recent years. I personally believe GCN kinda lost it around R9 290X. That was GCN's last hurrah. It was just brute force, but today, GPU's need a bit more finesse, not just raw specs. Sure they help, but it has to be designed clever. And I think Vega is just that. And I also think evolution of Vega will actually bring full benefits of its features, especially since parts of it require specific coding which, apart from maybe games from Bethesda won't really be used in first generation (like Primitive Shaders). But may find a way into DirectX or Vulkan. Eventually. And it also depends on the trends, architecture may be a hit or miss. Both, AMD and NVIDIA had their hits and misses with architectures.

 

[Nokiron]

I set up darkcoin mining rigs, and all of mine did.

Frankly, I don't believe you did any experimental testing, and I doubt you even understand what you are talking about.

Your chart suggests memory made a major difference! If you can't see your own chart supports my argument, talking to you is a waste of time.

What? You are the one not understanding anything. I'm not saying overclocking the memory is a waste of time (I never did).
I'm saying that increasing core clock is without a doubt the highest priority, and especially for gaming (which my original comment was about as well). Of course there are gains (Again, already said this), but you are wasting potential core clock and boost clocks by doing so, parameters that WILL increase performance much more.

Would you take a maximum performance boost of 7% or 19%? Again, the 980 is the absolute best example of this since it already has a small bus-width to a powerful GPU.

Well, I mined Dogecoins and random alt-coins at the time. I had every incentive to test. And so far your argument is non-existent, atleast I have presented some actual data instead of pure speculation and guessworks

We already know what OCing core on 1080 does...and it's not spectacular. Without mem OC you're missing a few percent (about 5 if you have good chips). The only thing wasting boost clocks is nvidia's artificial limitation. OCing mem brings the perf increase up to near parity with the percentage of core increase.

So...guess what? Faster memory is needed or it won't scale.

Yes, thanks for that proof. I can totally see it now.

 

[Caring1]

For once im fighting back English dictionary stlyey,

• the bare bones
  
  • 1The basic facts about something, without any detail.
    
    ‘the bare bones of the plot’
• • bare one's soul
    
    • Reveal one's innermost secrets and feelings to someone.
      
      ‘one feels vulnerable in baring one's soul to another’

One means to expose, the other means carries, or presents.
Beware of Greeks bearing gifts, is not the same as, Beware of Greeks baring gifts.

 

[owen10578]

Mostly what I see across the web is nVidia fantrolls flaming every. Single. AMD/Vega thread. Doesn't matter what forum I visit. It's hilarious how dumb they look. Obnoxious, HaHaHaHaHaHa Right.

PS: Out of curiosity, would you consider me an AMD 'fanboy'?

You do sound like one. All I have to say.

 

[Caring1]

Obnoxious, HaHaHaHaHaHa Right.

Nothing more needs to be said really.

 

[Gasaraki]

Nvidia fanboys are still in denial about the supremacy of HBM (due to latency).

Don't worry, they'll praise nvidia for using it and say how it's so good when nvidia finally brings it to the consumer level.

Who the hell cares about latency of the graphics memory? It has never been an issue before until some fanboys thinks it's suddenly an issue. Graphics core and memory bandwidth is where it's at.

 

[Gasaraki]

I was waiting for someone to show up with an idea that a perfectly competing product should cost half the price for no logical reason. Dude, this isn't communism where free shit just happens until it runs out of free shit. Business needs to make profits and by making shit free for no reason doesn't make profits. When will you people get it in your heads? They can make product tiny bit cheaper to make it more attractive, but expecting same performance and half the price just because it's an AMD, sorry, that's just an insult to intelligence.

It's called increasing market share when your market share is in the toilet. You increase market share at the expense of short term profits to increase long term profits. Chinese companies do it all the time. They release phones that are almost as good as a Samsung at half the cost. Market share increases dramatically. Release something better at a higher price. Increased profits because now your name is popular for value/performance. Currently people run to Intel and nVidia because they are more popular even though AMD offers something that's the same price and performance. People still buy 7700Ks and 1060s when they could be buying Ryzen 1700s and RX580s.

You know what would make me buy AMD? Better performance at a lower price. Cause it sure isn't going to be stability and efficiency.

 

[RejZoR]

Yeah, well, wake up from dreams and come down to reality. Cheaper means 650 vs 699, not 350 vs 699...

 

[thesmokingman]

It's called increasing market share when your market share is in the toilet. You increase market share at the expense of short term profits to increase long term profits. Chinese companies do it all the time. They release phones that are almost as good as a Samsung at half the cost. Market share increases dramatically. Release something better at a higher price. Increased profits because now your name is popular for value/performance. Currently people run to Intel and nVidia because they are more popular even though AMD offers something that's the same price and performance. People still buy 7700Ks and 1060s when they could be buying Ryzen 1700s and RX580s.

You know what would make me buy AMD? Better performance at a lower price. Cause it sure isn't going to be stability and efficiency.

Yeah, well, wake up from dreams and come down to reality. Cheaper means 650 vs 699, not 350 vs 699...

This. Cheaper sure, not giving it away committing seppuku in the process. smh

 

[Captain_Tom]

Yes, thanks for that proof. I can totally see it now.

Thank you for admitting you had no idea what you were talking about.


Most cards are built with a decent balance of bandwidth and compute. If you increase the core clock, don't expect much more performance unless you can increase the bandwidth by the same amount.


As for "The 980 being a good example", no it isn't. The 980 has a paltry 5.0 TFlops of compute, and the aggressive memory compression introduced with Maxwell gave it FAR more effective bandwidth than previous 256-bit cards. If you want a good example of the bandwidth issue just look at the 1070/1080 and Polaris cards: They get tiny performance increases from MASSIVE clockspeed increases. In fact a 1070 at 2.1 GHz rarely gets more than a 5-10% performance increase unless you can overclock the memory a ton.

 

[Captain_Tom]

@RejZoR is right. People shouldn't expect AMD to give the Vegas away at cost. AMD has been trying the approach of selling their chips for cheap in order to gain market share for a long time now and it's not working. It does absolutely no good to gain market share selling too cheap because you don't make any more profit that way. If a Vega performs at around a 1080 level then people should expect to pay close to what a 1080 would cost and if a Vega performs at around a 1080 Ti then people should expect to pay close to what a 1080 Ti would cost.

Somehow people, even AMD fans, have got it in their heads that AMD should behave like a non-profit charity. They are a business just like Nvidia and Intel and they should charge as much as the market will bear just like Nvidia and Intel and that's not greed, that's business.

Polaris is AMD's bargain-bin line-up, and it will remain that way. There's a reason AMD is making both Polaris and Vega at the same time: Polaris is a crazy simple design that is dirt-cheap to produce.


Vega on the other hand will be priced higher like you are saying, but don't expect them to sell it for the same price as Nvidia. The 1080 Ti came out first, and thus Vega will need to be cheaper if it has comparable performance. You can't just sell the same product later and expect success.


My guess is there are really 3 directions for how Vega will pan out:

1. AMD manages to get 1550 - 1600 MHz clocks, and Vega turns out to be as big as a leap as Maxwell was for Nvidia. Expect a $600 - $700 card that matches or beats the Titan XP.
2. AMD reaches decent 1400 - 1500 MHz clocks, and Vega is a nice new arch (but still not perfect). Expect a $500 - $600 card that trades blows with the 1080 Ti.
3. Something goes horribly wrong and AMD can't get above 1200-1300 MHz, and Vega brings a modest 10% IPC increase. The card will cost $400 - $500 and land 15% stronger than the 1080, but 10% weaker than the 1080 Ti.

Still not sure what I would bet on yet, but rumors will probably give us a hint within a couple weeks...

 

[Nokiron]

Thank you for admitting you had no idea what you were talking about.


Most cards are built with a decent balance of bandwidth and compute. If you increase the core clock, don't expect much more performance unless you can increase the bandwidth by the same amount.


As for "The 980 being a good example", no it isn't. The 980 has a paltry 5.0 TFlops of compute, and the aggressive memory compression introduced with Maxwell gave it FAR more effective bandwidth than previous 256-bit cards. If you want a good example of the bandwidth issue just look at the 1070/1080 and Polaris cards: They get tiny performance increases from MASSIVE clockspeed increases. In fact a 1070 at 2.1 GHz rarely gets more than a 5-10% performance increase unless you can overclock the memory a ton.

That was not a reply for you, but sure. Again, you are not providing proof.

That 3% and 4% average increase sure sounds like a lot!

http://www.sweclockers.com/test/237...orce-gtx-1080-och-gtx-1060-med-snabbare-minne

https://www.techpowerup.com/reviews/MSI/GTX_1080_Gaming_X/24.html
https://www.techpowerup.com/reviews/MSI/GTX_1080_Gaming_X_Plus_11_Gbps/30.html
(The Gaming X 11 Gpbs also maintains a higher core clock. surely that does not matter...)

 

[efikkan]

My guess is there are really 3 directions for how Vega will pan out:

1. AMD manages to get 1550 - 1600 MHz clocks, and Vega turns out to be as big as a leap as Maxwell was for Nvidia. Expect a $600 - $700 card that matches or beats the Titan XP.
2. AMD reaches decent 1400 - 1500 MHz clocks, and Vega is a nice new arch (but still not perfect). Expect a $500 - $600 card that trades blows with the 1080 Ti.
3. Something goes horribly wrong and AMD can't get above 1200-1300 MHz, and Vega brings a modest 10% IPC increase. The card will cost $400 - $500 and land 15% stronger than the 1080, but 10% weaker than the 1080 Ti.

1550-1600 MHz would yield ~13 TFlop/s (boost), and based on the performance per TFlop gap of Pascal vs. Polaris, it would still need a 10% improvement over Polaris.

If Vega has roughly the same performance per TFlop as Polaris, it would require over 14.1 TFlop/s to match GTX 1080 Ti, or ~1730 MHz average boost.

Additionally, Pascal is up to ~80% more energy efficient vs. Polaris. If Vega10 are to match GTX 1080 Ti, it needs to close most of this efficiency gap as well.