AMD: Speculation, Rumors, and Discussion (Archive)

[CSI PC]

They don't list any sources nor data to back up their claim though

Painfully I think they are basing their own article on that 'article' from WCCFTech by Khalid.
The WCCFTech article was 2 days ago while the Tweaktown was 1day 23hrs ago.

And the WCCF article is based upon the embedded platform (mistake to do a direct comparison between the embedded and comparable discrete as can be seen when comparing the Tonga 380X to the Tonga E8950MXM embedded as the clocks would need to be 750MHz for the E8950MXM but are reported by AMD at 1000MHZ with a TDP <95W).

This can be seen that WCCF is extrapolating and assuming the spec is 100% correct as reported by AMD for the embedded platform (which in the past was not entirely correct for previous Tonga embedded part) and correlated to the comparable discrete GPU (again in the past even allowing for the clock rate anomaly downclocking to 750MHz still would not reach <95W in the same way we see with the discrete Polaris 480 downvolted to 0.8V, so there is something specific to the embedded platform solutions).
Following paragraph from WCCF is what makes me think they are jumping the gun like they have with the patent that recently came to light and they are saying is definitely part of Vega.

AMD has reportedly started rolling out new revisions of its Polaris 10 & Polaris 11 GPUs with a 50%+ improvement in performance per watt. Polaris 10 and Polaris 11 are AMD’s latest GPUs powering the company’s midrange RX 480 & RX 470 as well as mainstream RX 460 graphics cards. The new GPU revisions will reportedly debut first in the embedded market and will deliver the same and/or slightly better performance compared to the previous revisions but at significantly lower power consumption

http://wccftech.com/amd-polaris-revisions-performance-per-watt/

I think the AMD reported spec for the Polaris embedded platform is more skewed this time due to not only keeping clocks of the discrete but then wrongly using the TFLOPs from that as well and exacerbated by the more dynamic nature of Polaris and way reported by AMD (Tonga and Hawaii were more consistent with their official spec and measurements).
Time will tell, but it does seem plausible that like before AMD is just reporting the wrong clock and compounded it further.
Case in point the embedded mentions specifically 1120MHz clock, but full TFLOPs is only using the 1266MHz clock.
http://www.amd.com/Documents/ultra-high-performance-gpu-product-brief.pdf
Multiple factors that highlight why it is wrong to jump to the conclusions of a general improvement with Polaris and applicable to all platforms.
Personally I think it is a downclocked/volt '480' GPU with specific-unique design for embedded platform that also assists with hitting the <95W TDP spec, I would say similar as before with the Tonga variant that had the wrong marketed clocks but maybe the right TFLOPs; cutting a 380X from 3.97 TFLOPs and 190W TDP to 3TFLOPs and <95W TDP is a big ask (and why I do not think there is a direct correlation between embedded and full discrete GPUs).

Cheers

 

[sebbbi]

Painfully I think they are basing their own article on that 'article' from WCCFTech by Khalid.
The WCCFTech article was 2 days ago while the Tweaktown was 1day 23hrs ago.

And the WCCF article is based upon the embedded platform (mistake to do a direct comparison between the embedded and comparable discrete as can be seen when comparing the Tonga 380X to the Tonga E8950MXM embedded as the clocks would need to be 750MHz for the E8950MXM but are reported by AMD at 1000MHZ with a TDP <95W).

This can be seen that WCCF is extrapolating and assuming the spec is 100% correct as reported by AMD for the embedded platform (which in the past was not entirely correct for previous Tonga embedded part) and correlated to the comparable discrete GPU (again in the past even allowing for the clock rate anomaly downclocking to 750MHz still would not reach <95W in the same way we see with the discrete Polaris 480 downvolted to 0.8V, so there is something specific to the embedded platform solutions).
Following paragraph from WCCF is what makes me think they are jumping the gun like they have with the patent that recently came to light and they are saying is definitely part of Vega.

http://wccftech.com/amd-polaris-revisions-performance-per-watt/

I think the AMD reported spec for the Polaris embedded platform is more skewed this time due to not only keeping clocks of the discrete but then wrongly using the TFLOPs from that as well and exacerbated by the more dynamic nature of Polaris and way reported by AMD (Tonga and Hawaii were more consistent with their official spec and measurements).
Time will tell, but it does seem plausible that like before AMD is just reporting the wrong clock and compounded it further.
Case in point the embedded mentions specifically 1120MHz clock, but full TFLOPs is only using the 1266MHz clock.
http://www.amd.com/Documents/ultra-high-performance-gpu-product-brief.pdf
Multiple factors that highlight why it is wrong to jump to the conclusions of a general improvement with Polaris and applicable to all platforms.
Personally I think it is a downclocked/volt '480' GPU with specific-unique design for embedded platform that also assists with hitting the <95W TDP spec, I would say similar as before with the Tonga variant that had the wrong marketed clocks but maybe the right TFLOPs; cutting a 380X from 3.97 TFLOPs and 190W TDP to 3TFLOPs and <95W TDP is a big ask (and why I do not think there is a direct correlation between embedded and full discrete GPUs).

Cheers

Anandtech lists 1.26 GHz as the turbo clock of the 95W embedded Radeon (same as RX 480). Don't know whether they just guessed it, or if they got the info from AMD.

http://www.anandtech.com/show/10710/amd-announces-embedded-radeon-e9260-e9550

 

[CSI PC]

Anandtech lists 1.26 GHz as the turbo clock of the 95W embedded Radeon (same as RX 480). Don't know whether they just guessed it, or if they got the info from AMD.

http://www.anandtech.com/show/10710/amd-announces-embedded-radeon-e9260-e9550

I linked the brochure from AMD in my post, which is the official guideline IMO.
Anandtech also adjusted the clock spec for the Tonga embedded as well, which they stated doing.
So the official AMD brochure is showing too high clocks for the Tonga version and if going with the 5.8 TFLOPs figure (personally I do not) they then use the wrong low clock for that Polaris version.
So unfortunately you either go by the AMD brochure or make adjustments and assumptions, point is the information seems to be even more skewed and wrong than even before and this is from AMD, therefore it is really wrong to conclude like WCCFT did that embedded platform has a massive performance improvement that correlates to the discrete GPU.
Cheers

Edit:
Yep here it is for Tonga:
http://www.anandtech.com/show/9682/...on-gpu-line-releases-radeon-e8950-e8870-e6465

Unfortunately AMD’s specs are somewhat at odds with each other, so it’s not clear what this part is clocked at. AMD cites a 1000MHz GPU clockspeed, however the 3 TFLOPS peak number implies 750MHz. Given the 95W TDP, we’re going to assume for the moment that it’s the latter.

If it was clocked at 1000MHz then that would mean it was operating just above the 380X discrete GPU with 190W TDP and would have the same TFLOPs, giving 50% improvement just like WCCFT is reporting for the Polaris embedded.

And AMD has stayed with the 1000MHz even now with the latest brochure late 2016 that shows spec of of E9550, I did the same as Anandtech and calculated the Tonga TFLOPs by downclocking to reach same figure (albeit it would still be above <95W just like Polaris 480 that the discrete GPU cannot hit at its lowest stable voltage envelope).
Here is the link: http://www.amd.com/Documents/ultra-high-performance-gpu-product-brief.pdf

edit 2:
I should had also pointed out that AMD actually shows both clocks for the discrete 480, unlike embedded:
LOCK SPEEDS (BOOST / BASE) 1266 MHz / 1120 MHz
http://www.amd.com/en-gb/products/graphics/radeon-rx-series/radeon-rx-480

 

[Alexko]

So GlobalFoundries would have us believe that while they needed to license Samsung's 14nm process to get to that node (possibly with lower performance?) they're going to get to 7nm ahead of Samsung?

Am I the only one who's skeptical, here?

 

[Razor1]

They have done the same things in the past to no avail (more than once too). So most likely the same thing will happen again. Its just an announcement lol that's the way I take it. IF it really happens great for them but I wouldn't put any money on their horses.

 

[Ryan Smith]

So GlobalFoundries would have us believe that while they needed to license Samsung's 14nm process to get to that node (possibly with lower performance?) they're going to get to 7nm ahead of Samsung?

Am I the only one who's skeptical, here?

Keep in mind that since they originally tried to develop 14XM, GloFo picked up IBM's foundry arm, and this seems to be where all new development is coming out of.

 

[Deleted member 2197]

AMD says: Radeon RX 470 is better choice than GeForce GTX 1050 Ti

Now, what makes this presentation so interesting. As far as I remember, it has not happened before for the competitor to release more performance figures than manufacturer themselves before reviews are even out.

Officially all we have from NVIDIA is one graph at GTX 1050 product page comparing GTX 1050 to 1050 Ti and GTX 750 Ti in Overwatch, Gears of War 4 and GTA5.

http://videocardz.com/63971/amd-rad...choice-than-geforce-gtx-1050-ti#disqus_thread

 

[DavidGraham]

Why would AMD admit RX 480 has lower performance than 1060?

 

[Alexko]

Well, it would appear to be true, albeit not by very much.

 

[ImSpartacus]

Why would AMD admit RX 480 has lower performance than 1060?

I can admire their honesty.

And I also get a kick out of the 750 ti being included. I don't think they wanted the 460 at the very bottom of the pack.

 

[tunafish]

IMHO inclusion of 750 Ti makes sense. It was basically the low-end GPU to beat for a long time, and while it's old now, there are a lot of people who have it and who would compare against it. I wonder if the 1050 Ti will take it's place now -- it does have 15W more TDP...

 

[lanek]

Asus Strik RX470 here had a discount for a week, just taken one for my brother ( he was running one of my old 5870 since some times as his 460 have die ). Nice little card, quiet, taking everything ( at least on the game he play ). he play at 1080p, so not a problem. Not have much time for bench it, as it is not for my system and i dont want to go to his home for install benchmarks and everything. But so far, seems good.for the price.

 

[Deleted member 13524]

AMD says: Radeon RX 470 is better choice than GeForce GTX 1050 Ti

Well it's probably worse for AMD to be selling the RX470 at $170, but IMO it's much better for a consumer getting a 4GB RX470 for $170 (I've seen it at $160 actually) than a 1050 Ti at $140.
Unless it's someone using a severely constrained PSU (under 300W?) then the 1050 Ti gets the best performance at <75W, I guess.

 

[Razor1]

Well whats interesting is Apple got fully functional P11 chips, this is the same thing we saw with Tonga, Apple got the fully functional chips first. Now a fully functional P11 at the rx460 mhz although higher in power consumption than its competition, should be able to compete well against a 1050ti. And it should be much easier to price that lower comparing to a cut down P10.

The way I'm looking at it, is they aren't getting enough fully functional P11's (apple allocation is low volume so that should not be affecting them) to put the squeeze on the 1050ti. By dropping the rx 470 that tells us, two things, AMD doesn't want the 1050ti variants, overclocked to gain traction and the supply issues of the P10 cards, wasn't just for them, its also for the P11. What ever that reason is.....

 

[mczak]

Well whats interesting is Apple got fully functional P11 chips, this is the same thing we saw with Tonga, Apple got the fully functional chips first.

Strictly speaking this is not entirely correct. Noone ever got fully functional Tonga chips, not even apple .

Now a fully functional P11 at the rx460 mhz although higher in power consumption than its competition, should be able to compete well against a 1050ti. And it should be much easier to price that lower comparing to a cut down P10.

That 8% or so more performance (just going with the sort of established tradition that a X% increase in CUs will net about X% / 2 in real world performance) isn't going to help much against a 1050Ti. Might be enough to help against the 1050, though.

 

[RedVi]

Will we see fully enabled Polaris 11 and a respun Polaris 10 return in the 5xx series? They'd have to move down a number but at least that way they would be closer to matching NV model number to model number. Unfortunately I don't think either are really compelling enough to be the next bonaire or pitcairn, but knowing AMD they will definitely be re-used for at least one more 'generation'.

 

[Deleted member 2197]

Chinese site publishes review of the AMD Radeon RX 470D
http://hexus.net/tech/news/graphics/98878-chinese-site-publishes-review-amd-radeon-rx-470d/

 

[Malo]

What is a 470D?

 

[Anarchist4000]

What is a 470D?

It's a Polaris 10 cut down more than a 470.

 

[CSI PC]

I wonder if the 470D will cannibalise sales of the 470 in China, depends upon games one looks at as they can be either very close or around 9% apart.
Would be interesting to know if there is a price point that is a threshold for majority of discrete GPU owners in the region and whether the 470D fits the highest accomodating price while the 470 is just beyond.
Cheers