AMD: Navi Speculation, Rumours and Discussion [2017-2018]

Status
Not open for further replies.
Yes keep changing the goalposts, that's a good trolling tactic.

1 - Vega was such a failure HBM is so bad
2 - Vega sold well b-but it was only because of mining, not HBM
3 - Hashing was good because of HBM b-but it's still bad because it's not gaming
4 - ?
 
Yes keep changing the goalposts, that's a good trolling tactic.

1 - Vega was such a failure HBM is so bad
2 - Vega sold well b-but it was only because of mining, not HBM
3 - Hashing was good because of HBM b-but it's still bad because it's not gaming
4 - ?

I am not going to continue with this off-topic discussion, but to refresh your memory Legitreview come to conclusion that gtx1070 with ordinary gddr5 memory has best hashrate per Watt :) if its not thanks to HBM, I wonder what that may be ....

https://www.legitreviews.com/amd-radeon-rx-vega-64-vega-56-ethereum-mining-performance_197049

 
4 - B-but mining wasn't so great either because look at this single article I found about ethereum..

Yeah if you don't know about Vega having to be tweaked for lower core / higher mem clocks for mining, nor that Vega cards are much more profitable for XMR than Pascal is for Ethereum.
 
Yes Vega was sold out. But not because it was most successfull GPU in AMD history nor because it uses HBM. Actually it was mining craze who save Vegas and RTG ass...

GTX1070 was failure ? Well then I agree you trully must live in an alternative universe :)


You are talking in the past.

And are upset that AMD offered HBM memory to end-users for $499~$699. And while AMD's 7nm Pro cards, along with Nvidia's both use HBM. The fact that is cost more doesn't matter, it performs better too. So what is your point..?

HBM is coming down in price and getting faster.
 
You are talking in the past.

And are upset that AMD offered HBM memory to end-users for $499~$699. And while AMD's 7nm Pro cards, along with Nvidia's both use HBM. The fact that is cost more doesn't matter, it performs better too. So what is your point..?

HBM is coming down in price and getting faster.

Of course I am talking in the past, I am not prophet, I can't talk about future ;-) do you ?

And I will repeat myself again, I have nothing against HBM technology, it's actually good technogy. But looking at the technology from wider point of view.

Nvidia is much better financial wise than AMD and yet they used HBM only on high marging products, where price cover all expenses. On customer and gaming products they used more common memory and it turns out to be a better strategy than AMD did with Vega10. Nvida launched new GPUs on market sooner than AMD and with better margins. AMD choice of HBM caused them only troubles, delays, higher manufacturing cost due to TSV and HBM packaging and yet they were slower than Nvidias counterparts.The higher cost of HBM also prohibited AMD from releasing RX Vega 11 as Polaris replacement.... Vega Mobile got on laptops very late... etc.

Perhaps in future HBM will be more common on graphic and it will be good, but for now it didn't any good to AMD/RTG especially from gaming perspective and gaming GPUs. If Navi uses GDDR6 it will have a better chances to succeed than Vega 10 with HBM.
 
Do we know if GDDR6 need less power than gddr5 ? Because the main argument for HMB, imo, is power draw. Navi with gddr6 means power draw from the gpu must be under control, if gddr6 is like gddr5. They can't do a FuryX / Vega 10 again...
 
Like Vega and Polaris, Navi is a family of GPUs, not a single chip.
Higher-end Navi may use HBM. Or not.
There could also be a "premium mid-range" that uses HBM, like the Vega 12 in the MacBook Pro.

Claiming that the use of HBM is inherently "bad" regardless of target performance/efficiency is irrational, to say the least.
 
it's just a wishful thinking....

To save energy ? look at the power consumption of Nvidia Pascal/Turing and compare them with Vega. Nvidia doesn't need to use HBM to keep power consumption of their GPUs down and they don't need them to maintain high bandwidth either.
Wishful? It's middle school physics... HBM simply will use less power for the IO with the shorter distances to drive a signal. There are very good reasons why all the high bandwidth devices are using it. AMD, Nvidia, and others.

This whole line of reasoning is just plain stupid.

Do we know if GDDR6 need less power than gddr5 ?
Yes, only because it uses a more advanced process. Process aside, it's GDDR5 with some controller tweaks for narrower channels.
 
Yes, only because it uses a more advanced process. Process aside, it's GDDR5 with some controller tweaks for narrower channels.

IIRC, GDDR6 consumes more power "per width" than GDDR5, but with 75% higher throughput per width the cards can use 25-33% less memory channels and still gain a sizeable boost in bandwidth, while saving power.
I. e. a GTX1060 class could now use a 128bit bus and still gain overall bandwidth while saving power in the process.
 
Wishful? It's middle school physics... HBM simply will use less power for the IO with the shorter distances to drive a signal. There are very good reasons why all the high bandwidth devices are using it. AMD, Nvidia, and others.

This whole line of reasoning is just plain stupid.


Yes, only because it uses a more advanced process. Process aside, it's GDDR5 with some controller tweaks for narrower channels.
HBM wins at "signal power" but it loses out at total power due to crap IO dies on the HBM itself. Imho AMD used an immature technology too early for the mass market.
 
Of course I am talking in the past, I am not prophet, I can't talk about future ;-) do you ?

And I will repeat myself again, I have nothing against HBM technology, it's actually good technogy. But looking at the technology from wider point of view.

Nvidia is much better financial wise than AMD and yet they used HBM only on high marging products, where price cover all expenses. On customer and gaming products they used more common memory and it turns out to be a better strategy than AMD did with Vega10. Nvida launched new GPUs on market sooner than AMD and with better margins. AMD choice of HBM caused them only troubles, delays, higher manufacturing cost due to TSV and HBM packaging and yet they were slower than Nvidias counterparts.The higher cost of HBM also prohibited AMD from releasing RX Vega 11 as Polaris replacement.... Vega Mobile got on laptops very late... etc.

Perhaps in future HBM will be more common on graphic and it will be good, but for now it didn't any good to AMD/RTG especially from gaming perspective and gaming GPUs. If Navi uses GDDR6 it will have a better chances to succeed than Vega 10 with HBM.


Again, you are in the past....


Also, no need to look into the future, what about what is already here currently..? Vega 64 selling @ $400 bucks, it has HBM memory. Why are you so angry at that fact and not able to weigh that against how big a chip the $400 V64 is, vs something 7nm and smaller & natively faster, (w/possible 2Ghz HBM)..?

You seem perplexed. Such a chip won't cost more... because of TODAY's HBM prices. It just won't cost as little as GDDR6. Again, You keep reflecting on last years prices & performance, without understanding today's and tomorrow's prices. GDDR6 will be great on a pure cheap gaming card, but a high-end GPU would best be served with HBM modules.

Again, you don't have to know the future, but you just can't stand there and be ignorant about it either. Memory is coming down in price & getting faster.
 
Do you have any data on that? Everything official has power reduction per gb of bandwidth vs gddr.
I believe AMD gave a favorable comparison of the absolute power consumption of the memory subsystem in Fury versus the 290's GDDR5 subsystem.

The revised HBM on Vega nearly doubled the bit rate and for undisclosed reasons ran the HBM above the specified voltages. The higher speed and higher voltages could have negated a significant fraction of the advantage, but I have not seen a full accounting of the memory controller+bus+DRAM stack consumption.
 
Again, you are in the past....


Also, no need to look into the future, what about what is already here currently..? Vega 64 selling @ $400 bucks, it has HBM memory. Why are you so angry at that fact and not able to weigh that against how big a chip the $400 V64 is, vs something 7nm and smaller & natively faster, (w/possible 2Ghz HBM)..?

You seem perplexed. Such a chip won't cost more... because of TODAY's HBM prices. It just won't cost as little as GDDR6. Again, You keep reflecting on last years prices & performance, without understanding today's and tomorrow's prices. GDDR6 will be great on a pure cheap gaming card, but a high-end GPU would best be served with HBM modules.

Again, you don't have to know the future, but you just can't stand there and be ignorant about it either. Memory is coming down in price & getting faster.

You seem to take it kind of personally. Angry ? C´mon ..... You´re trying to convince me that HBM with packaging and TSV won´t cost more and calling me perplexed ?
 
More Navi rumours

Navi 12

Segment
- Mid-range (performance like Radeon RX Vega)
Hardware - 40 shader clusters = 2560 shader units
Chip - Slightly smaller than Polaris 10/20 (= appr. 200-200mm²)
Release - End of Q2 2019

Navi 10
Segment - High-end (performance like GeForce RTX 2080)
Hardware - Unknown
Chip - Unknown
Release - Middle of Q3 2019
 
But will rtx 2080 performance level be high end for nvidia one year from now (yeah, with amd history, I believe q3 2019 will be (very) late 2019) ?
 
Perhaps not for the majority, im thinking that if amd could be more competitive, Nvidia might take down their prices. Amd today isnt bad in price per performance, but not enough competition yet.
2080 performance in 2020 isnt that stunning, we need more competition from amd.
2080Ti performance for 500, like it should be. Nvidia then has no choice.
 
Status
Not open for further replies.
Back
Top