Vega/Navi Rumors (Updated)

[ShintaiDK]

I was thinking 275 W.

So with 1500-1600Mhz you mean 1150-1250Mhz?

 

[Glo.]

So with 1500-1600Mhz you mean 1150-1250Mhz?

FUDZilla, VideoCardz and WCCFTech claim that 12 TFLOPs/1.5 GHz 4096 GCN core chip has 225W TBP.

 

[jpiniero]

So with 1500-1600Mhz you mean 1150-1250Mhz?

That's why it needs to be at TSMC.

 

[ShintaiDK]

FUDZilla, VideoCardz and WCCFTech claim that 12 TFLOPs/1.5 GHz 4096 GCN core chip has 225W TBP.

And they have never been wrong before. Oh, and Vega is launching in October too, right? And Vega 11 will have 6144SP?

 

[ShintaiDK]

That's why it needs to be at TSMC.

What if the issue isn't the node? Its the same excuses with Zen. Node, node, node, but its just as much if not more the uarch. Nvidia said something that Pascal wouldn't have clocked more than around 1300Mhz, if it wasn't for a lot of R&D going in to tweaking the design.

 

[kraatus77]

Well, hawaii and fiji both have same tdp, so it's possible that 64cu 14ff gpu can have 225w. specially with hbm2.

they have done it before so why can't again ? heck it will have more tdp than pol10.

performance on the other hand is completely different topic. but i believe it won't be more than 1080.

 

[Glo.]

And they have never been wrong before.

So in essence, you are claiming that they are wrong, because you believe that? Do you have better sources, information than they have?

What if the issue isn't the node? Its the same excuses with Zen. Node, node, node, but its just as much if not more the uarch.

What are you basing this on? If you would be paying attention to Zen thread really, and be interested in the topic, actually, rather than trying to prove your agenda that AMD sucks/rubbish, you would knew that BOTH: microarchitecture and process are in play when it goes to efficiency and performance, in equal way. Great microarchitecture can be let down by process, and its the opposite, as well.

 

[linkgoron]

So with 1500-1600Mhz you mean 1150-1250Mhz?

You've already been shown an RX 480 running @ 1475 mhz in this very thread. It's not impossible that AMD have improved their arch, and glofo have improved their process (/AMD are using Samsung) and they have Vega running at 1500mhz.

Is it 100% certain? No. Is it likely? I don't know. However, it is not impossible. Nobody is saying that it's a fact, but all you are doing is thread-crapping. You weren't so skeptical when people talked about 2.4 ghz 1080 factory OC (are there any factory 1080 that are even over 2ghz OOTB? I think I saw ~1990 mhz). I'm not saying that being skeptical is bad, but you are basically thread crapping and derailing the thread.

 

[kraatus77]

It's a situational thing, when talking about foundry glofo sucks according to him, but here in this case it's amd's architecture which is the culprit here. it's funny to see how everything changes according to trash something.

 

[gamervivek]

I don't think it was necessarily tweaking Pascal's design for higher clocks but something more akin to 4870->4890 clockspeed bump. AMD are long overdue that kind of clockspeed bump for a process.

 

[ShintaiDK]

You've already been shown an RX 480 running @ 1475 mhz in this very thread. It's not impossible that AMD have improved their arch, and glofo have improved their process (/AMD are using Samsung) and they have Vega running at 1500mhz.

Is it 100% certain? No. Is it likely? I don't know. However, it is not impossible. Nobody is saying that it's a fact, but all you are doing is thread-crapping. You weren't so skeptical when people talked about 2.4 ghz 1080 factory OC (are there any factory 1080 that are even over 2ghz OOTB? I think I saw ~1990 mhz). I'm not saying that being skeptical is bad, but you are basically thread crapping and derailing the thread.

Using overclocks as reference, also without power draw is simply silly.

 

[ShintaiDK]

I don't think it was necessarily tweaking Pascal's design for higher clocks but something more akin to 4870->4890 clockspeed bump. AMD are long overdue that kind of clockspeed bump for a process.

It was even explained on Anandtech.
http://www.anandtech.com/show/10325/the-nvidia-geforce-gtx-1080-and-1070-founders-edition-review/6

And that's why R&D funding is so critical. From 1325Mhz to 1733Mhz on the same node.

Given the already significant one-off benefits of such a large jump in the voltage/frequency curve, for Pascal NVIDIA has decided to fully embrace the idea and run up the clocks as much as is reasonably possible. At an architectural level this meant going through the design to identify bottlenecks in the critical paths – logic sections that couldn’t run at as high a frequency as NVIDIA would have liked – and reworking them to operate at higher frequencies. As GPUs typically (and still are) relatively low clocked, there’s not as much of a need to optimize critical paths in this matter, but with NVIDIA’s loftier clockspeed goals for Pascal, this changed things.

 

[gamervivek]

I'm not sure if they mean that Pascal's arrangement differing from Maxwell is the reason for its higher clockspeeds. So I don't think that GCN has to undergo a complete overhaul either.

 

[linkgoron]

Using overclocks as reference, also without power draw is simply silly.

You can look at the video at ~9:40. The video states 133W watts average and a peak of 149W when benchmarking.

EDIT:
This was already stated by maddie.

 

[Arachnotronic]

I'm not sure if they mean that Pascal's arrangement differing from Maxwell is the reason for its higher clockspeeds. So I don't think that GCN has to undergo a complete overhaul either.

It's at the actual circuit level that these changes were done. High level uArch of Pascal relative to Maxwell didn't change dramatically, but the circuit designs are probably all new.

 

[Glo.]

It was even explained on Anandtech.
http://www.anandtech.com/show/10325/the-nvidia-geforce-gtx-1080-and-1070-founders-edition-review/6

And that's why R&D funding is so critical. From 1325Mhz to 1733Mhz on the same node.

That is Nvidia's marketing in its best, in which idiots believe.

The same thing can be applied to AMD. Without tweaking the architecture for particular process, Polaris would not exceed 900 MHz:

Again: You always have to optimize.

 

[Arachnotronic]

That is Nvidia's marketing in its best, in which idiots believe.

Come on. This kind of statement is beneath you.

 

[Glo.]

You can look at the video at ~9:40. The video states 133W watts average and a peak of 149W when benchmarking.

Somehow I missed this. If the water cooled GPU has lower clocks, and higher power draw than the one in XFX GPU - then we are looking at best binned chip, or new revision of them.

Interesting.

 

[Glo.]

Come on. This kind of statement is beneath you.

I do apologize. But pay attention to how JHH can spin out most obvious things as achievements, and people are able to believe him!

I was watching the Pascal GTC presentation, and was amazed by how high clocks Pascal is able to achieve. Later, in Zen thread on this very forum I have learned its always matter of optimization of microarchitecture for the process. And that bad process can let down even the best microarchitecture on this world. Actually, most of people do not understand this, Pascal, consumer GPUs are just Maxwell GPUs with 2 tweaks: Dynamic Scheduling and next generation Memory Compression on 16 nm FF+ process. The only true new, next generation chip is GP100, with different architecture layout.
And Maxwell was able to achieve 1.5 GHz clocks, already on 28 nm process on Air.

All of this is off-topic, so lets move on.

 

[gamervivek]

It's at the actual circuit level that these changes were done. High level uArch of Pascal relative to Maxwell didn't change dramatically, but the circuit designs are probably all new.

Yeah, 4890 was almost complete unchanged from 4870. AMD didn't do any such improvement with 28nm, even regressing in some ways, so it's long overdue for them.

 

[rainy]

I'm not saying that being skeptical is bad, but you are basically thread crapping and derailing the thread.

We have on AT forums very strict rules - he's doing constantly thread crapping/trolling on AMD related threads and still doesn't get infractions/warnings for that.
From my perspective, it's a best proof that some members here have more rights than others.

 

[krumme]

We have on AT forums very strict rules - he's doing constantly thread crapping/trolling on AMD related threads and still doesn't get infractions/warnings for that.
From my perspective, it's a best proof that some members here have more rights than others.

Who is the mod? And how is it adressed?

 

[tajoh111]

No, it was running at roughly 1.05V stock and 1.18 overclocked, thus at lower than release voltage.

Power was ~ 100W gaming, ~ 133W [1475 Mhz] overclocked and ~90W in a heaven loop.

Design revision, an improved process, or both, which if true, bodes very well for Vega to hit high performance at competitive wattage.

If this is truly indicative of Polaris perf/watt, then all of the previous arguments as to Vega's inability to compete have been rendered worthless.

This is likely more the exception than the rule. Come on, one piece of evidence is enough to invalidate a wall of evidence and highly unscientific.

The only thing this shows evidence for is bias on the person doing the research.

When there is a extreme data point outside the norm, then there is likely an anomaly. It is not the norm and it's why two extremes on a set of data are often thrown out.

We don't keep the single extreme and throw out the hundred other data points. Extremes can be explained by more plausible explanations than he ones you gave.

There was likely something wrong with the test bench or a cherry picked sample or something is up with jays2cents.

The thing is this is huge publicity for XFX and will help them sell thousands more videocards compared to their competitors. No other card has overclocked this well for polaris while using this little power.

A power circuitry redesign won't save 120 watts of power. That's just impossible when this represents a 50% savings in power.

 

[Glo.]

This is likely more the exception than the rule. Come on, one piece of evidence is enough to invalidate a wall of evidence and highly unscientific.

The only thing this shows evidence for is bias on the person doing the research.

When there is a extreme data point outside the norm, then there is likely an anomaly. It is not the norm and it's why two extremes on a set of data are often thrown out.

We don't keep the single extreme and throw out the hundred other data points. Extremes can be explained by more plausible explanations than he ones you gave.

There was likely something wrong with the test bench or a cherry picked sample or something is up with jays2cents.

The thing is this is huge publicity for XFX and will help them sell thousands more videocards compared to their competitors. No other card has overclocked this well for polaris while using this little power.

A power circuitry redesign won't save 120 watts of power. That's just impossible when this represents a 50% savings in power.

When there is exception from the rule, then there is no rule.

I suggest watching the linked video in the first place.

 

[zinfamous]

When there is exception from the rule, then there is no rule.

I suggest watching the linked video in the first place.

The one thing that it shows (if his numbers are true), is that there certainly isn't an issue with Polaris uarch. It really only takes one sample to show what the design can actually do. The problem, then, lies in process/manufacture/vendor settings. Whether it's properly setting voltage, cooling, or simply the issues with manufacturing from glofo, it does only only take one outlier from this case with the XFX card to show that the design for Polaris is closer to what AMD advertised than what their detractors want it to be--and from what the ave release seems to be showing.

This isn't saying that the power issue with actual Polaris chips in actual cards isn't a problem. Stock performance on that card (mild OC) with great temps and great power draw is really quite impressive. Remains to be seen if that will eventually become the norm was Polaris matures...and whether or not review sites actually update those numbers if it does become the norm.

The only thing that I am certain of is that AMD detractors will never revise their opinions of the situation to reflect the current status of the chips if it is warranted.