AMD Polaris architecture – GCN 4.0

[Deleted User 234324]

Even though Polaris 10 is a smaller die size doesn't the 14nm size mean at least double the transistor density so it could easily be more powerful than the much larger tonga die.

 

[Seanspeed]

Even though Polaris 10 is a smaller die size doesn't the 14nm size mean at least double the transistor density so it could easily be more powerful than the much larger tonga die.

Well yea, the architectural improvements and process shrink should mean a far more efficient GPU, but if you generally think about double the efficiency, then I'd say this is likely to put it in the 'maybe a bit faster than a 390X' category. Which is fine if it can release at a great price, but is there any reason to be excited about it other than that? People right now are drooling for more performance, not necessarily just better prices. There's just something very unexciting about a new lineup of cards on a long-time-coming new process and not actually have them push things on any.

Maybe they'll surprise, but it looks like Nvidia with their 300mm GP104 cards will be the ones that people flock to for more exciting performance results.

 

[bru]

Even though Polaris 10 is a smaller die size doesn't the 14nm size mean at least double the transistor density so it could easily be more powerful than the much larger tonga die.

Yup, this is where the 2*perf/watt come into play, on the last conference call AMD were talking about 2*per/watt, so look at Tonga even though it is a considerably bigger chip (size wise we would need to be looking at something in between Pitcarn and Tonga, but I will use Tonga to allow for the GCN improvements) you either get twice the performance for the same power or the same performance at half the power. Now there have been rumours of 100W+ for Polaris 10 which would make it about the same performance as Tonga being a 190W chip and seeing as AMD themselves have said this is their most efficient chip yet, I cannot imagine it gulping down wattage but merely sipping it.

So to sum up, I reckon something between Pitcarn and Tonga in size at half the power, with GCN improvements, gives us something about Tonga performance, a little over half the power usage (100W+) it could be pushing close to 390 performance or more depending on how much power it actually uses.

 

[CAT-THE-FIFTH]

Maybe they'll surprise, but it looks like Nvidia with their 300mm GP104 cards will be the ones that people flock to for more exciting performance results.

Maybe on tech forums,but unless Nvidia have something truely better under £250,I am not sure.

I have one of the most popular cards out there,ie, a GTX960 and something like R9 390 level performance at £175 to £200 will be a 50% improvement over a GTX960. I think plenty of people like me would get a card with that level of performance.

The same goes with the GTX750TI,a significantly better bus powered card woul probably be a decent seller - if Polaris 11 is 150MM2,that is far bigger than previous estimates and would place it at closer to R9 380/R9 380X level performance. That would be nearly 40% faster than a GTX950.

 

[Seanspeed]

Maybe on tech forums,but unless Nvidia have something truely better under £250,I am not sure.

I have one of the most popular cards out there,ie, a GTX960 and something like R9 390 level performance at £175 to £200 will be a 50% improvement over a GTX960. I think plenty of people like me would get a card with that level of performance.

The same goes with the GTX750TI,a significantly better bus powered card woul probably be a decent seller - if Polaris 11 is 150MM2,that is far bigger than previous estimates and would place it at closer to R9 380/R9 380X level performance. That would be nearly 40% faster than a GTX950.

Yea of course it's still a big market for these cards, but just lowering prices on existing cards would mostly do the trick for that market, too.

I'm not saying it's a bad thing at all. It's just not 'sexy', ya know? £250-320 is still a big market and it'll be quite exciting to see what kind of performance you can get in that range, but it seems like AMD will be sitting that out unless they just knock the Fiji prices down a good chunk.

 

[bru]

if Polaris 11 is 150MM2,that is far bigger than previous estimates and would place it at closer to R9 380/R9 380X level performance. That would be nearly 40% faster than a GTX950.

Only if it used the same power as a currant 150mm^ card (something like the 360 100W) twice the performance would bring it up to the 380 level, which isn't going to happen on Polaris 11. Polaris 11 is the super small super low power chip from what has been rumoured so far, Polaris 10 is the supposed 100W+ chip certainly not Polaris 11.

 

[GFX-Kid]

I am hoping this can achieve the same performance of a Fury X or be within 10% of it. Really need it to get to 5000 or thereabouts on 3dmark firestrike ultra when overclocked for me to be interested.

 

[nashathedog]

The launch was rubbish just like with Fury. If AMD had not skimped on the reference cooler and had decent third party cards out,and not allowed Nvidia to troll them,then things would be different.

There as bad as each other and that occasion was probably payback for the 4gb means 4gb jibes that AMD were using as free promotional material with the Hawaii chips after the 970's ram fiasco.

 

[flopper]

I am hoping this can achieve the same performance of a Fury X or be within 10% of it. Really need it to get to 5000 or thereabouts on 3dmark firestrike ultra when overclocked for me to be interested.

I suspect some games will run faster and some slower than a 980ti depending on what design they choose. dx12 and dx11 add some variables also

 

[Freddie1980]

Even though Polaris 10 is a smaller die size doesn't the 14nm size mean at least double the transistor density so it could easily be more powerful than the much larger tonga die.

The 14 and 16 NM process GF and TSCM are using isn't as advanced as Intel's. It more of 20nm hybrid process but the finfets allow chips to have power draws as if it were a genuine 14/16nm shrink but I wouldn't expect to see a doubling of the transistors

 

[humbug]

The Substrate is 20nm, the transistors are 3D 14nm, or 16nm for TSMC, exactly there same as Intel. ^^^^^^

Samsung’s 14nm FinFET Process Offering ⁃ 14LPE and 14LPP.

14LPE (Early edition) targets the early technology leaders and time-to-market customers such as mobile application SoCs to meet the latest mobile gadgets’ aggressive schedule and improved performance/power requirements. 14LPE is the first foundry process technology manufactured in the foundry industry with the successful volume ramp-up. 14LPE offers 40% faster performance; 60% less power consumption; and, 50% smaller chip area scaling as compared to its 28LPP process.

14nm LPE vs 28nm LPP

# 50% smaller (Half the size so double the transistor density)
# 40% faster
# 60% reduction in power consumption.

Or if using 14nm LPP add another 10% performance on top of that.

14LPP (Performance boosted edition) is the 2nd FinFET generation which the performance is enhanced up to 10%. 14LPP is the single platform for every application designs with the improved performance for computing/Network designs and the lowered power consumption for Mobile/Consumer designs. 14LPP will be the main process technology offering in 2016 and after.

http://www.samsung.com/semiconductor/foundry/process-technology/14nm/

 

[GFX-Kid]

I suspect some games will run faster and some slower than a 980ti depending on what design they choose. dx12 and dx11 add some variables also

That would be fine by me. Just hoping they are genuinely overclockers dream

If they overclock like the 980Ti or better, then I will grab one and watercool it. They should offer watercooled edition for £50 extra, I would pay that

 

[CAT-THE-FIFTH]

Only if it used the same power as a currant 150mm^ card (something like the 360 100W) twice the performance would bring it up to the 380 level, which isn't going to happen on Polaris 11. Polaris 11 is the super small super low power chip from what has been rumoured so far, Polaris 10 is the supposed 100W+ chip certainly not Polaris 11.

Emm?? You do realise the 212mm2 Pitcairn chip in the HD7870 was 10% faster than an HD6970 and consumed 80% less power on average?? Pitcairn was nearly half the size of the chip in the HD6970 and that was going from 40NM to 28NM AND IT HAD MORE TRANSISTORS TOO. This is going from 28NM to second generation 14NM wih Finfets.

The 14NM process has double the transistor density with a 60% to 70% reduction in power.

So,that means if you shrunk Tonga directly down with no uarch changes,you could have an R9 380X probably consuming no more than 100W. That would put it close to 180MM2.

Tonga is a 360MM2 chip,but even the R9 380X does not have the 384 bit memory controller enabled.The R9 380 has 2/3 of the memory controller enabled and 87.5% of the shaders enabled.

Polaris 11 is also using a 128 bit memory controller too,so less RAM chips and less power consumption over a partially disabled 384 bit one.

It also lots of other changes which have been detailed - one of them appears to be power gated CPUs,etc.

So taking that into consideration Polaris 11,can probably easily match a GTX960 or R9 380 and have massively lower power draw.

If you even look at the average power figures for the R9 380X for example:

https://tpucdn.com/reviews/ASUS/R9_380X_Strix/images/power_average.png

A 60% reduction in that would put the power consumed at under 100W.

The R9 380 is consumes less power than that.

FFS,AMD had a bus powered HD7850 years ago - do you really believe Polaris 11 is going to be slower than an HD7850??

There is a bus powered GTX950 made on 28nm too.

 

[nox_uk]

Have any of the Polaris cards been physically pictured yet? Wondering about physical size for my mini itx...

 

[CAT-THE-FIFTH]

Let's look it from another perspective:

http://tpucdn.com/reviews/ASUS/GTX_950/images/perfrel_1920_1080.png

That is review of the bus power GTX950.

The 232MM2 Polaris 10 with a 256 bit memory controller is rumoured to be R9 390X level performance.

Even if AMD went with Polaris 11,being half a Polaris 10,ie a 116MM2 die it would at least GTX950 level performance.

The latest rumour is placing it at 150MM2,which is 30% bigger.

The different between a GTX960 and R9 380 is around 20% to 25% in performance.

 

[Mauller]

Let's look it from another perspective:

http://tpucdn.com/reviews/ASUS/GTX_950/images/perfrel_1920_1080.png

That is review of the bus power GTX950.

The 232MM2 Polaris 10 with a 256 bit memory controller is rumoured to be R9 390X level performance.

Even if AMD went with Polaris 11,being half a Polaris 10,ie a 116MM2 die it would at least GTX950 level performance.

The latest rumour is placing it at 150MM2,which is 30% bigger.

The different between a GTX960 and R9 380 is around 20% to 25% in performance.

Considering the 270X is Pitcairn XT, as well as Polaris 11/Baffin XT having the same number of CU's as Pitcairn XT, It will have greater performance than the 950 since Pitcairn XT already beats the 950 in the majority of cases. And by a fair amount in good a few.

 

[bru]

If you even look at the average power figures for the R9 380X for example:

https://tpucdn.com/reviews/ASUS/R9_380X_Strix/images/power_average.png

A 60% reduction in that would put the power consumed at under 100W.

The R9 380 is consumes less power than that.

What....the 380 is nowhere near under 100W. it is a 190w TDP card.

Don't forget 2*perf/watt is either twice the performance at the same power or the same performance at half the power.(or of course anything in between)

So 150mm^ is about Bonaire in size, so that is twice a 360 in performance or half Bonaires 100wTDP

 

[bfar]

I suspect some games will run faster and some slower than a 980ti depending on what design they choose. dx12 and dx11 add some variables also

If Polaris 10 performs that well we could be looking at a very competitive year. RTG's relative silence and Nvidia's rush to be first to market is a very good sign. I doubt RTG are gonna win the overall performance crown, and yet I have a suspicion Polaris will hit all the right notes when it finally arrives. Low power, low price, great performance.

 

[flopper]

If Polaris 10 performs that well we could be looking at a very competitive year. RTG's relative silence and Nvidia's rush to be first to market is a very good sign. I doubt RTG are gonna win the overall performance crown, and yet I have a suspicion Polaris will hit all the right notes when it finally arrives. Low power, low price, great performance.

232mm2 core.
depending on how one calculate the number of transistors, performance watt etc...
One can say at least this, a 390 beats the 980ti today with Dx12.
so Polaris will continue that trend and be great in dx12.
How well it will far in older engines as dx9, dx11 as those will still be here for some time is more unknown.

I foresee a 9700 type of card one we will recall fondly when the next generation card hit us and who dosnt want to own such one?

 

[AlamoX]

GTX 1080 about 20% over 980Ti, no competition
polaris 10XT 480X with perf equivalent of Fury Pro/Nano competing with GTX1070
polaris 10Pro 480 best ratio perf/dollar will be around 290 perf
polaris 11XT 470X around 380 perf competing with 1060Ti
polaris 11Pro 470 around 270X perf