AMD Polaris 10 Samples work at 1.27 GHz

[Det0x]

Netburst's problem was that it made this trade off and was a power hog/underperformer for it. NVIDIA seems to have driven frequencies up significantly at roughly similar perf/clock while improving power efficiency.

Netburst this is not.

power->watt->heat->milking edition cooler™->83°c throttling->underperforming

Nothing more need to be said on this matter.

 

[boozzer]

power->watt->heat->milking edition cooler™->83°c throttling->underperforming

Nothing more need to be said on this matter.

how loud is it? looks like hot and loud might get it's turn for this nv gen.

 

[96Firebird]

how loud is it? looks like hot and loud might get it's turn for this nv gen.

Not loud at all.

Not really hot either, just has a low target temp. at 82°C. Wonder if the new nodes are still sensitive to high temps?

 

[railven]

Not loud at all.

Not really hot either, just has a low target temp. at 82°C. Wonder if the new nodes are still sensitive to high temps?

Wonder if it's similar to the issue Intel saw with their early smaller node (and I think still experience to a degree.)

Also 37dba at load? What kind of fan profile is that? Give it a little more juice NV. Woof.

 

[Stuka87]

Not loud at all.

Not really hot either, just has a low target temp. at 82°C. Wonder if the new nodes are still sensitive to high temps?

The question is, how long was the card running under load before they measured it, as they do not say. Did they wait 20 mins for the temps to climb up to where it is throttling to measure?

 

[digitaldurandal]

power->watt->heat->milking edition cooler™->83°c throttling->underperforming

SNIP

Nothing more need to be said on this matter.

Wow, that is terrible.. looks kind of like the issue that original blower 290/290x had. I wonder if review sites will catch this, because with that cooler and at those settings your overclocks are going to be worthless. Who cares if it his 2500Mhz if it dips back to 1600Mhz after 20 minutes. Maybe Nvidia will learn the lesson that AMD had to (but probably not.)

 

[Mopetar]

Not loud at all.

Not really hot either, just has a low target temp. at 82°C. Wonder if the new nodes are still sensitive to high temps?

I'm assuming that's 37 dBa at stock settings. If you want to OC you need to crank the fan up or it's going to throttle. Tom's has a pretty good overview of the situation.

Basically if you want to sustain the OC, you need 100% fan otherwise it throttles. However he does say that even when it's loud (slightly less loud than a 980) it's not overly annoying:

The 1080's acoustic profile is really pleasant. It’s a nice broad whooshing/airflow noise, and the fan’s motor can never really be picked out. This is a lot better than the cheap axial fans that AMD used to put on its reference designs (and that some partners still use). In comparison, the Nvidia GeForce GTX 1080’s noise is tolerable.

 

[Arachnotronic]

power->watt->heat->milking edition cooler™->83°c throttling->underperforming

Nothing more need to be said on this matter.

Power consumption is roughly the same for ~70% more performance.

Again, this is not a Netburst, no matter how much you want to say it is.

 

[airfathaaaaa]

Power consumption is roughly the same for ~70% more performance.

Again, this is not a Netburst, no matter how much you want to say it is.

actually when you claim "overclockers dream" (but only for 10 mins) while the card goes back on the non boost clock well its even worse than roy telling that fury x was an overclockers dream...

 

[Slaughterem]

Dear AMD
I hate you because the way that you conduct business does not agree with what I as a millennial demand to be entitled to. If you are going to release a new gpu you better damn well make sure that it is the fastest card on the planet and you need to do it now. The cost of this card can not be more than $389. Don't you know that your competitor has paper launched an overclocked maxwell gpu and told us that the msrp is $100 less than their reference, err founders edition which all AIB will release in a few days. I don’t care if some say that NV marketing has created a Cluster F$$K with this brilliant strategy. It is your fault and how dare you not be competitive verse this offering which will cause me to spend a boat load of money for a high range card.

Personally I don't even think that you will release a new product, because you have failed to provide leaked information giving me and your competitor a clue about all of the said products. Their is no doubt that if you release any product it will be an unmitigated failure due to:
Slow clock speed, no overclocking potential, high power usage, the fan is loud - even if less than the NV 1080 OC at 48 db {which my dishwasher is currently rated at} Temperature of more than 83 for you is not acceptable even if your card doesn’t throttle at this point. http://www.computerbase.de/2016-05/geforce-gtx-1080-test/9/

Your marketing and business strategy absolutely sucks and everyone should be replaced by the experts found on these forums. I mean whoever heard of TAM - it is the most ridiculous acronym and should be replaced by LMFAO.

The price is not what I am willing to pay for a mainstream product. You should have targeted the 5% of us with your products. The fact that you may have a product that is close to Fury performance at a much lower TDP means nothing because we already have products that cost more money providing this performance. And LMAO if you think that Apple and other OEM's will use these supposed products in laptops for back to school market.

Your drivers are never released on time, they cause my hardware to croak and for me to have hypertension even though I do not own your card.

DX12 is a sham and you want us to believe that your cards are better at DX12 games than Pascal? We all know that you are paying off the developers regardless of what the developers say. Async shaders is also a bunch of BS, even though you can do compute + graphics at the same time NV and tech review sites have decided that the architecture of Pascal is better, even though they do not have much of any performance gains from it. Lets not even believe for one minute that you played a slight role towards the release of DX12 and Vulcan, Nvidia was the company pushing these API’s so they could rightfully force customers who own Kepler and Maxwell to purchase new products.

A bit of advice for you when you have your supposed launch at computex, rehearse rehearse rehearse because if you have a slip up like NV did you will be crucified. And yes I believe it is a supposed launch because your announcement on your web site is not specific about what you are announcing and your private press meeting a few days before is a pipe dream.

Also remember if Roy Taylor makes any other comments about your supposed products this will lead to the bankruptcy of you dysfunctional company, so please have Roy on vacation during this supposed launch.

I hope that this advice helps you because I am an owner of AMD cards and will buy Nvidia this time. ( JK it just sounds better if I actually may have bought one of your products in the past)
Wishing you a quick bankruptcy
The paid Nvidia forum warrior.


Threadcrapping,flaming and trolling are not allowed
Markfw900

 

[Glo.]

Power consumption is roughly the same for ~70% more performance.

Again, this is not a Netburst, no matter how much you want to say it is.

No. It isn't. In that table you have one part with Power Target, and higher power target. The differences are meaningful. Normal Power target allows the GPU to be 180W GPU, but it throttles like a pig. Power target for that GPU in normal circumstances is 83 Degrees. Higher power target, the second part of the table is for 92 degrees(!). What happens then? The power consumption jumps by 39W up!. It is 4W lower than GTX 780 Ti, and that GPU according to TPU numbers uses 225W while gaming. So it is not anymore a 180W. Even then, because of higher power draw, and/or higher temperatures, the GPU is not able to sustain power target core clocks, and throttles like a pig. Everything is in the review of Computerese.de. Im staggered nobody have seen this before.

 

[Layer8]

Just one thought, we've seen 2 Polaris 10 benchmarks, 67DF:C7 and 67DF:C4. Where is 0x67C0 ?

From https://cgit.freedesktop.org/~agd5f...mdgpu/amdgpu_drv.c?h=drm-next-4.7-wip-polaris:

	/* Polaris11 */
	{0x1002, 0x67E0, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_POLARIS11},
	{0x1002, 0x67E1, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_POLARIS11},
	{0x1002, 0x67E8, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_POLARIS11},
	{0x1002, 0x67E9, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_POLARIS11},
	{0x1002, 0x67EB, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_POLARIS11},
	{0x1002, 0x67FF, PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_POLARIS11},
	/* Polaris10 */
	{0x1002, [B]0x67C0[/B], PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_POLARIS10},
	{0x1002, [B]0x67DF[/B], PCI_ANY_ID, PCI_ANY_ID, 0, 0, CHIP_POLARIS10},

 

[Face2Face]

Not loud at all.

Not really hot either, just has a low target temp. at 82°C. Wonder if the new nodes are still sensitive to high temps?

It's about as loud as the reference GTX 780, which I can say was very quiet up to 80% fan speed. By comparison my 7970 GE reference card is extremely loud at 80%, it's almost comical. I will say 82c isn't low, as Kepler and Maxwell throttle at or around that temp.

 

[96Firebird]

I will say 82c isn't low, as Kepler and Maxwell throttle at or around that temp.

You're right, 980 had a target temp. of 80°C, 84°C for the 980 Ti. I guess I never really paid attention to that, since I always look at aftermarket cards which keep the temps in check. Can't wait to see the performance of the aftermarket cards that don't throttle!

But this is all very OT...

 

[Concillian]

AMD has talked about a lot of hardware changes in GCN 4.

This is talking about power consumption... performance per watt. Of course a shrink better address this.

What I was talking about was performance per shader. I don't think anyone is expecting much change in "IPC" or performance per shader per clock speed with this gen. At the very least we can set a performance floor, as we know it's not going to be worse 'per core per clock' than existing cards.

Sure, there's possible upside, but I don't think anyone reasonably expects anything like 20+% improvement in performance per shader per clock.

 

[Bacon1]

This is talking about power consumption... performance per watt. Of course a shrink better address this.

What I was talking about was performance per shader. I don't think anyone is expecting much change in "IPC" or performance per shader per clock speed with this gen. At the very least we can set a performance floor, as we know it's not going to be worse 'per core per clock' than existing cards.

Sure, there's possible upside, but I don't think anyone reasonably expects anything like 20+% improvement in performance per shader per clock.

Look at all the "new" parts, its not just a die shrink but architecture changes too.

 

[Stuka87]

This is talking about power consumption... performance per watt. Of course a shrink better address this.

What I was talking about was performance per shader. I don't think anyone is expecting much change in "IPC" or performance per shader per clock speed with this gen. At the very least we can set a performance floor, as we know it's not going to be worse 'per core per clock' than existing cards.

Sure, there's possible upside, but I don't think anyone reasonably expects anything like 20+% improvement in performance per shader per clock.

AMD has already stated that their 2.5x perf:watt increase is due partly to power savings, and partly to IPC increases.

 

[atakall]

What I was talking about was performance per shader. I don't think anyone is expecting much change in "IPC" or performance per shader per clock speed with this gen. At the very least we can set a performance floor, as we know it's not going to be worse 'per core per clock' than existing cards.

Sure, there's possible upside, but I don't think anyone reasonably expects anything like 20+% improvement in performance per shader per clock.

In addition to the above-comments noting that AMD has repeatedly represented architecture changes that will improve performance, per AMD at this morning's JPMorgan conference:

Mark Papermaster:
"So first and foremost, both companies [NVDA & AMD] are going to leverage FinFET. But when you look at Pascal, it's largely just that. From our assessment, it's a technology map. What you'll see AMD do is drive, as I said a moment ago, both design and process. So you're going to see a generational performance provide advantage from both aspects."

He was kind enough not to note the Pascal regression.

Source Transcript:
http://seekingalpha.com/article/397...chnology-media-telecom-conference?part=single

 

[atakall]

As there have been several posts of rumors regarding delays in Polaris as well as bringing forward Vega to Oct. 2016, some might find AMD's statements this morning at JPMorgan's conference of interest:


"We're on track to roll it out mid this year. So it's coming up eminently.
***
And so it's a perfect follow-on, Harlan, to what we did on very, very high performance last year with Fury and now come back with FinFET, right into the sweet spot of the market, right on track, right when we promised we would be there with that next-generation Graphics Core Next.
***
So it's strong second half is what we're teeing up for. . . . Seeing Polaris in volume in the third and fourth quarter as you drive towards that will be important.
***
So if you see our new GPU roadmap that we rolled out in March, we have this Polaris family. And then that is followed-on by another family in 2017, which is actually a very tight cadence in terms of how you would normally launch products into the market. There might be a longer timeframe associated between new architectures. So we're very excited about the follow-on family. So if you imagine, Polaris will ramp over a series of quarters, as we nail that performance and mainstream portion of the market, then you follow it on, and buy a new family in terms of the ultra-enthusiast portion of the market."


Source Transcript:
http://seekingalpha.com/article/3977...ce?part=single

 

[boozzer]

No. It isn't. In that table you have one part with Power Target, and higher power target. The differences are meaningful. Normal Power target allows the GPU to be 180W GPU, but it throttles like a pig. Power target for that GPU in normal circumstances is 83 Degrees. Higher power target, the second part of the table is for 92 degrees(!). What happens then? The power consumption jumps by 39W up!. It is 4W lower than GTX 780 Ti, and that GPU according to TPU numbers uses 225W while gaming. So it is not anymore a 180W. Even then, because of higher power draw, and/or higher temperatures, the GPU is not able to sustain power target core clocks, and throttles like a pig. Everything is in the review of Computerese.de. Im staggered nobody have seen this before.

woah, thanks for another informative post.

 

[sirmo]

To be fair, Maxwell was already pretty good in terms of efficiency. There is only so much you could get out of it.

However this shady business of the "premium" Founders Edition card price, especially being power and thermally limited is a pretty big f-up.

Also Nvidia hasn't done anything to improve their architecture in terms of supporting new APIs. Preemption is a poor excuse for hardware asynchronous compute. And it's basically a bandaid.

It's particularly egregious when you consider this card is billed at being designed for VR.

Have a listen at what Oculus Rift folks think of Nvidia's VR capability: https://www.youtube.com/watch?v=v_eRwxqhAGo&feature=youtu.be&t=538

 

[C@mM!]

At the very least we can set a performance floor, as we know it's not going to be worse 'per core per clock' than existing cards.

Sure, there's possible upside, but I don't think anyone reasonably expects anything like 20+% improvement in performance per shader per clock.

I wouldn't quite be so so sure.

Pascal took a hit to IPC compared to maxwell. Nothing saying that it won't happen to AMD as well. That being said, I do feel like AMD has been preparing their cards for this moment for quite a while, whilst as Maxwell was a reactionary architecture.

 

[railven]

As there have been several posts of rumors regarding delays in Polaris as well as bringing forward Vega to Oct. 2016, some might find AMD's statements this morning at JPMorgan's conference of interest:


"We're on track to roll it out mid this year. So it's coming up eminently.
***
And so it's a perfect follow-on, Harlan, to what we did on very, very high performance last year with Fury and now come back with FinFET, right into the sweet spot of the market, right on track, right when we promised we would be there with that next-generation Graphics Core Next.
***
So it's strong second half is what we're teeing up for. . . . Seeing Polaris in volume in the third and fourth quarter as you drive towards that will be important.
***
So if you see our new GPU roadmap that we rolled out in March, we have this Polaris family. And then that is followed-on by another family in 2017, which is actually a very tight cadence in terms of how you would normally launch products into the market. There might be a longer timeframe associated between new architectures. So we're very excited about the follow-on family. So if you imagine, Polaris will ramp over a series of quarters, as we nail that performance and mainstream portion of the market, then you follow it on, and buy a new family in terms of the ultra-enthusiast portion of the market."


Source Transcript:
http://seekingalpha.com/article/3977...ce?part=single

Interesting that he says another family in 2017.

Does that mean a new generation 500-series, or the Vega "family" of GPUs?

Does that mean whatever in 2017 will be treated as a refresh or extension of 400-series?

Unless they got something else planned for 2017 and Vega does see release in 2016 to fill up the top.

EDIT:

Actually this peaks my interest:

So if you imagine, Polaris will ramp over a series of quarters, as we nail that performance and mainstream portion of the market, then you follow it on, and buy a new family in terms of the ultra-enthusiast portion of the market.

Does this imply buy the mainstream card and then buy the enthusiast card? A trade up option? Que?

 

[SPBHM]

I wouldn't quite be so so sure.

Pascal took a hit to IPC compared to maxwell. Nothing saying that it won't happen to AMD as well. That being said, I do feel like AMD has been preparing their cards for this moment for quite a while, whilst as Maxwell was a reactionary architecture.

it took a hit? if you are comparing 1080 to 980 ti you have to consider things like 256vs384bit, 2vs3MB l2, 64vs96 ROPs and so on, not just the amount of CUDA cores...

the super high clocks from Pascal are looking like big trouble for AMD if they are stuck at 1.3GHz,

also Maxwell was already a significant upgrade from Kepler, while GCN 1.0 to 1.2 less so,

 

[atakall]

Interesting that he says another family in 2017.

Does that mean a new generation 500-series, or the Vega "family" of GPUs?

Does that mean whatever in 2017 will be treated as a refresh or extension of 400-series?

Unless they got something else planned for 2017 and Vega does see release in 2016 to fill up the top.
Actually this peaks my interest:

Quote:
So if you imagine, Polaris will ramp over a series of quarters, as we nail that performance and mainstream portion of the market, then you follow it on, and buy a new family in terms of the ultra-enthusiast portion of the market.
Does this imply buy the mainstream card and then buy the enthusiast card? A trade up option? Que?

(1) My interpretation is that he is repeating that Vega will be the new family after Polaris and will be 2017. Of course, if AMD's is now planning to pull Vega forward to 2016 (eg., the October 2016 rumor), he wouldn't disclose such and would stick with current "2017."

(2) It's an incorrect transcription. "Buy" should be "by" such that the proper quote is "then you follow it on by a new family in terms of the ultra...." She (Ruth Cotter of AMD) should have used "with" instead of "by." The statement is really just a repeat of what he said a few times about the cadence of Polaris to Vega (see earlier quotes in the post).