[Benchlife] R9 480 (Polaris 10 >100w), R9

[Silverforce11]

Expensive node, crap yields (being at GloFo), nVidia's pricing.

AMD will change their tune when they see what nVidia is charging. If the 1070 is slightly slower than 980 Ti performance for $499, there's not much reason to be much cheaper than that if it's close in performance.

We don't know the yields. A small chip should not cause problems for a node that's already mass production for over a year. A big 600mm2 chip will certainly have issues though.

The question here is what is AMD trying to achieve with Polaris?

Maximum revenue/profit will mean they have to provide an incentive for users to upgrade, for a LOT NVIDIA users since they are the majority of the PC gaming market. You cannot entice them to switch with poor perf/$. Offering 980Ti performance for $299 is a sure way to get many to switch back to AMD.

GP104 is also very likely to be 20-30% faster than 980Ti, so it's premium and "NV tax" will mean it will sell due to the halo effect. If Polaris is priced at $399 like you say, many NV gamers will just buy the 1070 instead.

 

[Silverforce11]

Depends on the game. Especially with AMD's overhead problems with directx 11. But when it comes to power draw, the 970 is far superior.

Since you have a great history of accurate maths and prediction, such as AMD's Polaris being 6 months behind Pascal...

Hmm.

 

[3DVagabond]

Depends on the game. Especially with AMD's overhead problems with directx 11. But when it comes to power draw, the 970 is far superior.

Overall in DX11 the 390 is faster. And TPU's bench suite is not AMD-centric.

 

[jpiniero]

The question here is what is AMD trying to achieve with Polaris?

Get competitive on perf/w to regain mobile market share.

GP104 is also very likely to be 20-30% faster than 980Ti, so it's premium and "NV tax" will mean it will sell due to the halo effect. If Polaris is priced at $399 like you say, many NV gamers will just buy the 1070 instead.

Many of them will buy the 1070 anyway. But a $100 difference would be pretty good if the perf and perf/w is similar.

 

[Silverforce11]

Get competitive on perf/w to regain mobile market share.

Many of them will buy the 1070 anyway. But a $100 difference would be pretty good if the perf and perf/w is similar.

The best ways to regain notebook share is to give excellent perf/w and perf/$. These two go hand in hand.

You nailed it on the 1070, if AMD wants current 950, 960 and 970 owners to switch to Polaris 10, they need to offer beyond great perf/$ to attract them. Otherwise they will go with the 1070.

When you're the underdog and you want to gain mindshare and marketshare, you have to go all out and give a great deal. No wishy washy in-between.

 

[parkerface]

Or you could have gotten a 970 and not had to worry about power draw.

A 970 didn't offer me anything more in performance than a 390 for the games I play, cost the same at the time of my purchase, and can't make me money like a 390 has.

I lost interest in NVidia products years ago.

 

[maddie]

Expensive node, crap yields (being at GloFo), nVidia's pricing.



AMD will change their tune when they see what nVidia is charging. If the 1070 is slightly slower than 980 Ti performance for $499, there's not much reason to be much cheaper than that if it's close in performance.

Get competitive on perf/w to regain mobile market share.



Many of them will buy the 1070 anyway. But a $100 difference would be pretty good if the perf and perf/w is similar.

One thing for sure, you are very consistent about ignoring the division head and projecting your own beliefs.

I guess the statement about making the entry requirements for VR more affordable is a trivial one. Explain how it applies to mobile in a very significant way.

Few months to go.

 

[Adored]

It's pretty bizarre just how badly the 970 is performing now tbh. I think it is capitulating worse than even the OG Titan did.

Maybe that's the wrong way to look at it though, it was probably very much overperforming due to AMD underperforming for a long time.

 

[KaRLiToS]

It's pretty bizarre just how badly the 970 is performing now tbh. I think it is capitulating worse than even the OG Titan did.

Maybe that's the wrong way to look at it though, it was probably very much overperforming due to AMD underperforming for a long time.

It is starting to make the GTX 1070 very attracting don't you think.

 

[Game_dev]

Since you have a great history of accurate maths and prediction, such as AMD's Polaris being 6 months behind Pascal...

Hmm.

I did say "up to". So far I'm not wrong. I certainly won't count a paper launch.

I wasn't expecting AMD to abandon the high end all together. I guess they need to focus on consoles.

 

[Excessi0n]

For some reason I was under the impression that the Tesla wasn't GP100, just because it's solely aimed at HPC. The block diagram that AT and other sites showed for it has all the raster engines (ROPs) removed and if that's the case it's worthless as a gaming card. It's easily possible it was accidentally omitted, but it makes sense to ditch them if they're trying to make the beefiest HPC card imaginable.

The ROPs being absent on the diagram doesn't mean much. I'm pretty sure the GK110 diagrams from the old Tesla presentations omitted ROPs as well and that chip obviously made it into graphics cards.

GP100 might not make it into a geforce card, but if that turns out to be the case it will probably be because Nvidia released a specialized gaming chip without HPC stuff like double-precision and NVLink (GP102, presumably).

 

[dzoni2k2]

I did say "up to". So far I'm not wrong. I certainly won't count a paper launch.

I wasn't expecting AMD to abandon the high end all together. I guess they need to focus on consoles.

I don't think they are abandoning high end at all, they are just refocusing on the mainstream, where most of the volume is. It's not the 980Ti that shifted the market share, it's the 960 and 970.

 

[Game_dev]

I don't think they are abandoning high end at all, they are just refocusing on the mainstream, where most of the volume is. It's not the 980Ti that shifted the market share, it's the 960 and 970.

They are targeting last year's performance. They won't be up against NVIDIA's old cards.

 

[3DVagabond]

I did say "up to". So far I'm not wrong. I certainly won't count a paper launch.

I wasn't expecting AMD to abandon the high end all together. I guess they need to focus on consoles.

So, define "up to". Would you consider that to cover you if say, AMD released 1st? or withing a few weeks? Would "up to six months" still make your prediction accurate?

That's just squirming to avoid eating crow.

 

[3DVagabond]

They are targeting last year's performance. They won't be up against NVIDIA's old cards.

They are comparing to current performance.

As far as targeting last years performance, isn't that exactly what nVidia did with GM104 but with lower power usage?

And I'd say they failed seeing as how Hawaii is starting to slap it around fairly regularly lately. All they did was deceive the masses with lies about specs and inflated benchmarks due to Gameworks.

 

[JDG1980]

The question here is what is AMD trying to achieve with Polaris?

I think Polaris 11 is aimed at OEM design wins (especially in mobile and AIO), while Polaris 10 will focus on perf/watt and perf/$ for the mainstream desktop market (the kind of people currently buying 960s and 970s).

Pricing predictions:
R7 470 (cut Polaris 11 w/ 4GB GDDR5): $149
R9 480 (cut Polaris 10 w/ 8GB GDDR5): $249
R9 480X (full Polaris 10 w/ 8GB GDDR5[X?]): $349

Rumors don't mention a 470X, so I think there won't be a R7 470X at first because the best dies will all be going to Apple for the 15" rMBP. The 27" Retina iMac should offer a choice between full Polaris 11 and Polaris 10 (they currently offer Pitcairn and Tonga). The new Mac Pro may only offer dual Polaris 10, and potentially will get a minor refresh offering Vega as an option in 9-12 months.

 

[AtenRa]

I did say "up to". So far I'm not wrong. I certainly won't count a paper launch.

I wasn't expecting AMD to abandon the high end all together. I guess they need to focus on consoles.

I wasnt expecting NVIDIA to abandon the High-End all together, GP104 will not be the fastest card on the market when it will release (Pro Duo will). I guess they need to focus on cars.

:sneaky:

 

[airfathaaaaa]

i really dont understand why people keep doubting news about polaris 10 being close to 980ti/furyx

this is their roadmap we knew this since the cringe worthy event amd did this is what we all expect
its natural for the cards to evolve its not like 90ti was so far ahead that amd cant reach it with those high end cards....

once vega arrives(remember vega is the true fury line successor)then we can talk

 

[R0H1T]

i really dont understand why people keep doubting news about polaris 10 being close to 980ti/furyx

this is their roadmap we knew this since the cringe worthy event amd did this is what we all expect
its natural for the cards to evolve its not like 90ti was so far ahead that amd cant reach it with those high end cards....

once vega arrives(remember vega is the true fury line successor)then we can talk

I don't think that was ever in question, the only point worth debating was that could it beat the two (or just Fury) with regular GDDR5, especially at higher resolutions, & how the bandwidth would be a major roadblock in achieving that objective. Let's be clear, if the Polaris' are clocked high enough, it could even beat the Titan X but it'd need the GDDR5x to lead at all resolutions.

 

[airfathaaaaa]

why?they downclocked the entire 7xxx series and still was in front in most cases gddr5x wont help them i have yet to see a game to tank a 512bit bus

 

[rainy]

AMD's Bulldozer (and successive chips) desktop chips and APUs also lagged behind their own K10 chips. Here's a comparison between Phenom X4 (K10) and one of their Piledriver CPUs which was fabbed with a 32 nm process, whereas the older K10 was on a 45 nm node.

I know it is off-topic but in all of those tests FX-8350 is faster (sometimes by big margin) vs X4 980.

Otherwise it's a pretty big stretch to think a ~230 mm^2 GPU is going to compete head to head with a 601 mm^2 GPU, especially when you consider that GCN still has a lot of compute hardware, whereas NVidia basically stripped out anything that wasn't useful for gaming with Maxwell.

It's interesting how people are missing quite an obvious thing: if those 232 mm2 of Polaris 10 are correct, it looks indeed smallish in comparison to Grenada/Hawaii (438 mm2) and especially Fiji (596 mm2).
However we are comparing 14 and 28nm - the same Grenada/Hawaii made at 14nm would be circa 220 mm2 which means that Polaris is slightly bigger.

I'm not expecting that top Polaris would be on 980 Ti level, nevertheless I have no problem to imagine that it could be very close to Fiji in 1080p and quite close in 1440p at much lower TDP.

 

[dzoni2k2]

They are targeting last year's performance. They won't be up against NVIDIA's old cards.

What the heck are you talking about? What last year's performance?

 

[rainy]

What the heck are you talking about? What last year's performance?

Just ignore him - he's a pure troll (history of his posts speaking for itself).

 

[Glo.]

I did say "up to". So far I'm not wrong. I certainly won't count a paper launch.

I wasn't expecting AMD to abandon the high end all together. I guess they need to focus on consoles.

AMD is bringing high-end performance to mainstream. They do not even talk yet about high-end solutions, apart from Fiji.

 

[jpiniero]

I guess the statement about making the entry requirements for VR more affordable is a trivial one. Explain how it applies to mobile in a very significant way.

AMD does have a history of exaggerating things, y'know. If the entry point for VR is 970ish performance, then I suppose he could be talking about a decently cut P10 that's slightly worse than 970 performance for something less than what the 970 goes for.