News [WCCFTech] AMD to reveal Navi GPU at Computex (partial rumor)

[Jaskalas]

Yes, overall power consumption can matter. But Perf:Watt won't make a game play faster, its a talking point for all intents and purposes...

I'd say TDP does matter for those of us with, say, a PSU from Dell that comes with a regular XPS Desktop. I have no reason to believe I should try pushing it much past the power consumption of my 1060 6gb. It really was a factor in choosing between AMD and Nvidia. Within a certain price point, TDP is something I weigh with performance. With the hope that I never have to find out my system's limit the hard way.

 

[Yotsugi]

And you know that how ???

Specific kind of free form magic.

 

[AtenRa]

Specific kind of free form magic.

In other words, you simple dont know.

 

[Yotsugi]

In other words, you simple dont know.

Very bad argument.

 

[Stuka87]

Depends on many factors, really.
Either way Navi isn't focused on perf@W.

That's a no.
Also everything GFX10 comes with a new SMU and you'll see why.

If AMD was willing to toss out chips that require higher voltage, they could then lower the core voltage to a more sane level. Its not unheard of to drop 30-40W off a Polaris chip by undervolting it. My own RX480 shows this kind of drop. But, AMD isn't in a place to throw away chips that work, only they require more voltage to do so.

I'd say TDP does matter for those of us with, say, a PSU from Dell that comes with a regular XPS Desktop. I have no reason to believe I should try pushing it much past the power consumption of my 1060 6gb. It really was a factor in choosing between AMD and Nvidia. Within a certain price point, TDP is something I weigh with performance. With the hope that I never have to find out my system's limit the hard way.

I feel this is very much a niche case though. There certainly are people like you, but its a smaller minority.

 

[Yotsugi]

If AMD was willing to toss out chips that require higher voltage, they could then lower the core voltage to a more sane level

Its already on the sane level for either dGPU vendor.
AMD needs not lower voltage, but better uArch.
Which they'll get, sort of.

 

[crazzy.heartz]

This announcement couldn't have come at a better time; was just about to pull the trigger on a 1660Ti.

Would wait to see what new features AMD is offering with Navi, since this GPU will be powering both next gen XBox and PS5.

I am pretty sure AMD will offer one of it's variant with 8GB Memory (another reason i'm hesitant to purchase a 1660/2060 card).

 

[Yotsugi]

This announcement couldn't have come at a better time; was just about to pull the trigger on a 1660Ti.

Navi is August so you're in for a hefty wait.

 

[crazzy.heartz]

Navi is August so you're in for a hefty wait.

Been waiting for forever lol. I'll endure for few months more..

As I'll be upgrading both my GFX card and Monitor; this will a Major upgrade as I don't change components often.

Hope this will fulfill my criteria (8GB RAM, ~150W power consumption, ~ $300, 1440p High Settings/Ultra Textures)

 

[Yotsugi]

Hope this will fulfill my criteria (8GB RAM, ~150W power consumption, ~ $300, 1440p High Settings/Ultra Textures)

Shouldn't be that far off.

 

[Stuka87]

Its already on the sane level for either dGPU vendor.
AMD needs not lower voltage, but better uArch.
Which they'll get, sort of.

Well, the fact remains that undervolting Polaris and Vega chips by 100mV is common place. When you do this, not only does power consumption go way down, but performance goes up. This suggest the core voltage is being set to high, which we already know. It is being set high because some chips wont run with a lower voltage. And rather than toss out those chips, AMD sets the voltage high enough to where they will be stable. This specific case has nothing to do with uArch.

 

[Yotsugi]

undervolting Polaris and Vega chips by 100mV is common place

As if undervolting Maxwell/Pascal isn't a thing.

 

[tajoh111]

If AMD was willing to toss out chips that require higher voltage, they could then lower the core voltage to a more sane level. Its not unheard of to drop 30-40W off a Polaris chip by undervolting it. My own RX480 shows this kind of drop. But, AMD isn't in a place to throw away chips that work, only they require more voltage to do so.



I feel this is very much a niche case though. There certainly are people like you, but its a smaller minority.

This is not unique to AMD or Polaris.

GTX 1060 could do the same thing. 85 watt @1800mhz on the gtx 1060.

Nvidia Max q is basically based on this.

AMD needs a new architecture. Undervolting to barely enough volts will lead to RMA headaches(clock instability), lower yields and Nvidia doing the same thing making AMD go back to square 1, with both company having headaches for binning and lower yields for high end chips.

Radeon VII is just as overvolted as Turing at this point. Turing chips have more headroom under stock volts than a radeon vii.

https://docs.google.com/spreadsheets/d/1Iim9e_ejX3nkgxLIZ3vLu1seQ1m0lDTKUhClJpAO-Gk/edit#gid=0

While maintaining stock clocks, most chips undervolt less than 0.1 volts for Radeon VII. Nvidia chips can do this as well.

Undervolting in it's current is mostly apologism, for AMD chips for it's performance per watt deficit. While it was valid for the 7970 and lesser extent 290x because both chip had significant overclocking headroom under stock volts(25% and about 15% respectively). This isn't the case today because at stock voltage, there simply isn't much headroom and Nvidia and AMD overclock to about 10% higher or less for the most part under stock voltage. This means for long term stability, these cards have the proper voltage settings.

Undervolting to the edge of stability is consumer resource highjacking/apologism. That is putting undervolting into the hand of the consumers so they do all the binning work/ responsible for any chip instability. It's the most affordable option for AMD which is why AMD is doing it and encouraging it in its viral marketing channels. Getting consumer to undervolting to barely enough voltage to run a chip is going to save power on any chip whether it is Nvidia or AMD.

But AMD fans propping it up as some AMD exclusive feature is getting tiresome. AMD undervolting every AMD chip to the exact voltage it needs to run stock is not a practical solution for AMD and is a complete overinflation of general performance/watt and is literally a best case scenario which is impractical. If AMD could undervolt chip to a specific voltage, they would.

Long terms solutions for AMD performance per watt woes are new architecture/software scheduler (big driver team) or reduce chip to chip variation via error analysis/correction at silicon level like Nvidia does currently(amd does it to but to lesser extent) which decrease chip to chip variance meaning chip to chip behavior is the same mostly. This undervolting every single card to an exact voltage with millions of videocards is impractical.

 

[Hitman928]

As if undervolting Maxwell/Pascal isn't a thing.

Nvidia cards can be undervolted as well to improve perf/w but I haven't seen (outside of maybe an outlier here or there) what is typical for Polaris/Vega where stock clocks are maintained if not increased while significantly undervolting the card.

Best cases I've seen AMD cards are able to match Nvidia in perf/w when Nvidia is at stock, but obviously some tweaking can bring Nvidia back into the lead though not as large as stock vs stock.

 

[maddie]

Nvidia cards can be undervolted as well to improve perf/w but I haven't seen (outside of maybe an outlier here or there) what is typical for Polaris/Vega where stock clocks are maintained if not increased while significantly undervolting the card.

Best cases I've seen AMD cards are able to match Nvidia in perf/w when Nvidia is at stock, but obviously some tweaking can bring Nvidia back into the lead though not as large as stock vs stock.

This is true. I know of no case where undervolting a Nvidia GPU leads to increased performance, but am open to being corrected. This is a critical clue to AMD running at higher than optimum voltages. Let us see if the CPU team with their advanced V control had any input into Navi.

 

[Hitman928]

This is not unique to AMD or Polaris.

Is there a link to this article to see if/how they are undervolting? Back when I had my 480 I could undervolt at stock clocks and get between 100-115 W during serious gaming.

Radeon VII is just as overvolted as Turing at this point. Turing chips have more headroom under stock volts than a radeon vii.

https://docs.google.com/spreadsheets/d/1Iim9e_ejX3nkgxLIZ3vLu1seQ1m0lDTKUhClJpAO-Gk/edit#gid=0

While maintaining stock clocks, most chips undervolt less than 0.1 volts for Radeon VII. Nvidia chips can do this as well.

If you look at only samples where overclocking and underclocking are not in play, the average undervolt is 103 mV. Are there any data sets showing Turing undervolt results?

I agree that AMD has been in long need of a new architecture to catch up to Nvidia in gaming perf and perf/w, but the undervolt results I've seen from AMD owners and from my own experience tend to support the narrative of AMD cards being significantly overvolted from the factory which has lead to their significantly worse perf/w stock results.

 

[Yotsugi]

software scheduler

Scheduling barely affects power.

 

[tajoh111]

Is there a link to this article to see if/how they are undervolting? Back when I had my 480 I could undervolt at stock clocks and get between 100-115 W during serious gaming.



If you look at only samples where overclocking and underclocking are not in play, the average undervolt is 1030 mV. Are there any data sets showing Turing undervolt results?

I agree that AMD has been in long need of a new architecture to catch up to Nvidia in gaming perf and perf/w, but the undervolt results I've seen from AMD owners and from my own experience tend to support the narrative of AMD cards being significantly overvolted from the factory which has lead to their significantly worse perf/w stock results.

All you have is reddit threads.

But the results are impressive.

https://www.reddit.com/r/nvidia/comments/9wpulp

https://www.reddit.com/r/nvidia/comments/ajwtlw

https://www.reddit.com/r/nvidia/com...nvolt_the_easy_way/?ref=share&ref_source=link

There isn't much undervolting talk of turing or pascal because it is not used in apologism as much to prop up the performance per watt. Turing and pascal cards already have good performance per watt so there isn't much effort into say the performance per watt is good.

However if we look at mobile chips, we see that the GTX 1060 and gtx 1070 are 95% percent the performance of their desktop counterparts but with a TDP of 66.66% of the desktop version leading to TDP's of 100 and 80 watts respectively or a 50 and 40 watt power reduction. This is real world application of undervolting that is commercially applicable and a better demonstration of real world undervolting and underclocking since these chips are binned for low voltage and underclocked for maximum performance per watt on a practical level at mass production level.

Look at the mobile frequencies for RX 480 and 1060, you see why the rx480 is not a popular chip in laptops. 1077boost vs the 1680 which is a frequency degradation of 15%(normal boost of 1266) vs the 2.3% reduction(1709mhz) in favor of the gtx1060.

This allows the GTX 1060 to fit into much thinner laptops including some thing ones. The RX 480 mobile fit into the same profile of laptops as the GTX 1070 because of their TDP(100watts). The difference is the gtx 1070 mobile is 60-70% faster because the clocks are 3-4% lower vs the desktop variant compared to bigger downgrade in frequency for the rx480. Which again makes the rx480 as a mobile part undesirable.

Laptops are an aggregate purchase of a whole system which make such a difference in performance undesirable even with the cost difference. Since gaming laptops have a much more specialized purpose, the gaming performance matters more since it is largely not upgradable and is a specialized purpose where the rest of the usability characteristics(weight, battery life, looks) are compromised.

The 200 or 300 dollar difference in price(is much less when in an aggregate system) leads to the GTX 1070 having the better performance per dollar because the purpose of the device is gaming and the system is purchased as a whole. $1200 rx 480 laptop vs $1500 gtx 1070 laptop with the latter having 65% better performance leads to the latter having better performance per dollar even with the 25% price difference.

If the performance per watt could be narrowed with similar effort by both Nvidia and AMD, with undervolting/underclocking, we would see the performance per watt shrink in laptops since both companies are underclocking/undervolting in both those cases. However since the difference grows, the performance per watt of pascal vs polaris is not any better/perhaps even worse when both chips are undervolted/underclocked.

People undervolting a specific chip to a specific minimum undervolt with a limited stability testing when a chip is at it's best(newest) is not a realistic portrait of performance per watt.

 

[Leadbox]

Scheduling barely affects power.

How do you figure that? I saw a test, 7970 vs the 680. When measured at the card, the 7970 drew~45W more and when measured at socket, system power, it was pulling only ~10W more. Higher CPU usage on nvidia because of the software scheduler reduced the deficit.

 

[Yotsugi]

How do you figure that?

By actually having a basic understanding of how GPUs work.

Higher CPU usage on nvidia because of the software scheduler reduced the deficit

Has buggerall to do with compile time scheduling.

 

[Leadbox]

By actually having a basic understanding of how GPUs work.

Has buggerall to do with compile time scheduling.

You must be a hoot at parties with those one liners of yours. Keep em' guessing yeah?

 

[Yotsugi]

You must be a hoot at parties with those one liners of yours. Keep em' guessing yeah?

?

 

[maddie]

You must be a hoot at parties with those one liners of yours. Keep em' guessing yeah?

?

Guess it works both ways.

 

[Hitman928]

All you have is reddit threads.

But the results are impressive.

https://www.reddit.com/r/nvidia/comments/9wpulp

https://www.reddit.com/r/nvidia/comments/ajwtlw

https://www.reddit.com/r/nvidia/com...nvolt_the_easy_way/?ref=share&ref_source=link

There isn't much undervolting talk of turing or pascal because it is not used in apologism as much to prop up the performance per watt. Turing and pascal cards already have good performance per watt so there isn't much effort into say the performance per watt is good.

Looking through those thread, pretty much all are undervolting and underclocking. Again, not the same situation. Even those that say they're at stock it's difficult to say with the way Nvidia boost works. They'd have to run actual benchmarks to know for sure.

I owned both an Rx480 and a 1060 (still have this one actually) and the Rx480 I was able to drop power use by a good 30% or more while maintaining the same performance. With the GTX 1060 I was able to cut out probably 15% power for only about 5% drop in performance. With the RX480 undervolted the GTX1060 at stock was still slightly better in perf/w but it was pretty close at that point. Ended up keeping the 1060 though as it was the better card at the time.

Again, no one is arguing that Nvidia isn't winning in both perf and perf/w. People are just saying that AMD is trying to salvage every possible chip so they're using voltages beyond what they need to on most chips which has led to making their lagging perf/w seem worse than it could be. We'll see what Navi brings at 7nm. For AMD's sake, hopefully they get a lot better efficiency out of the chips because their last few attempts have not been great in relation to their competition.

 

[Thala]

I owned both an Rx480 and a 1060 (still have this one actually) and the Rx480 I was able to drop power use by a good 30% or more while maintaining the same performance. With the GTX 1060 I was able to cut out probably 15% power for only about 5% drop in performance. With the RX480 undervolted the GTX1060 at stock was still slightly better in perf/w but it was pretty close at that point. Ended up keeping the 1060 though as it was the better card at the time.

Nice story about your 2 samples...look at the curves from Toms Hardware, the GTX1060 even increases its perf/power lead when undervolting compared tp Rx480. GTX1060 is taking 50% power at 70% performance while RX480 still uses 65% power at 70% performance. This makes the GTX1060 a much better under-clocker than the RX480 - and is the main reason you find the GTX160 at a reasonable TDP in laptops without much performance regression.

It is pretty much given that both NVidia and AMD are working with the same voltage margins for shipping products and both are "undervolting" for their laptop products.