Support NeoGAF

[Deleted member 465307]

1080/1070 are a 330mm^2 die. Polaris 10 is a 230mm^2 die. Some expect more out of them since they are on a slightly smaller node than Nvidia and typically have more dense designs. Rumors are pegging it at up to 390X tier performance at $300 and under.

This confuses me because I thought that the 390X was almost equivalent to a GTX 970 and the 970 launched at $329. I admit I'm very ignorant, though. I see that the 390X launched at $429. Even though I know it's foolish to compare prices across brands and use that to assume power, is the 390X more powerful than I thought?

 

[Zojirushi]

This confuses me because I thought that the 390X was almost equivalent to a GTX 970 and the 970 launched at $329. I admit I'm very ignorant, though. I see that the 390X launched at $429. Even though I know it's foolish to compare prices across brands and use that to assume power, is the 390X more powerful than I thought?

I think the 390 shoud be the direct competitor of the 970.

 

[Rushersauce]

I think the 390 shoud be the direct competitor of the 970.

Yes. It's the 390 the direct competitor, not the 390X

 

[chaosblade]

390 competes with the 970, 390X competes with the 980 (not Ti).

If the 1070 falls short of Nvidia's claims about it being in the ballpark of a 980Ti and the full/top P10 chip exceeds the 390X at the rumored prices, that could have interesting results - P10 slightly short of the 1070 in performance while costing nearly $100 less. I'd definitely be interested in P10 then, but I'm not going to set myself up for disappointment and expect it.

 

[Deleted member 465307]

I think the 390 shoud be the direct competitor of the 970.

Yes. It's the 390 the direct competitor, not the 390X

390 competes with the 970, 390X competes with the 980 (not Ti).

If the 1070 falls short of Nvidia's claims about it being in the ballpark of a 980Ti and the full/top P10 chip exceeds the 390X at the rumored prices, that could have interesting results - P10 slightly short of the 1070 in performance while costing nearly $100 less. I'd definitely be interested in P10 then, but I'm not going to set myself up for disappointment and expect it.

Oh, excellent! Thanks for the responses! A 390X+ with a lower TDP and at less than $300 sounds like it would be perfect for my needs, so I hope that's what we get. (I mean, obviously a Fury X at less than $300 would be great, too.)

 

[roadrunn3r]

"The first AMD Radeon graphics cards based on the company's "Polaris" architecture are slated for a late-May launch, according to Thai tech-site Zolkorn. The company is reportedly planning an elaborate launch event in Macau, China, days ahead of the 2016 Computex trade-show in Taipei. AMD has reportedly already sent invites to media outlets, although to a very limited number (in comparison to, say, NVIDIA's GTX 1080 event in Austin, US). The event could see a paper-launch of the first Radeon R9 400 series graphics cards based on the 14 nm "Ellesmere" and "Baffin" chips, with AIB-branded cards being exhibited at Computex, and market-availability following shortly after."

TechPowerUp link.

 

[LordCiego]

"The first AMD Radeon graphics cards based on the company's "Polaris" architecture are slated for a late-May launch, according to Thai tech-site Zolkorn. The company is reportedly planning an elaborate launch event in Macau, China, days ahead of the 2016 Computex trade-show in Taipei. AMD has reportedly already sent invites to media outlets, although to a very limited number (in comparison to, say, NVIDIA's GTX 1080 event in Austin, US). The event could see a paper-launch of the first Radeon R9 400 series graphics cards based on the 14 nm "Ellesmere" and "Baffin" chips, with AIB-branded cards being exhibited at Computex, and market-availability following shortly after."

TechPowerUp link.

If thats true its good news that AMD its ready to do an unveiling before the expected Computex trade show.

 

[Ozorov]

NVIDIA’s GeForce GTX 10 Series announcement has put some doubts on how AMD is planning to address such a powerful GPUs with Polaris architecture. However according to some unconfirmed whispers AMD might have just changed its plans for (probably most awaited) enthusiast GPU called VEGA 10.

http://videocardz.com/59808/amd-vega-gpu-allegedly-pushed-forward-to-october

 

[Papacheeks]

What do you mean? Nvidia had Vulkan support in their regular drivers since very shortly after its release, and it's not really "custom" in any way.

Is it being used currently with current cards? If not I want to see real world retail bench marks from retail cards, and component setups.

Let's see if those same frames hold true on a actually retail built system.

 

[Kezen]

http://videocardz.com/59808/amd-vega-gpu-allegedly-pushed-forward-to-october

Fantastic.

 

[DieH@rd]

http://videocardz.com/59808/amd-vega-gpu-allegedly-pushed-forward-to-october

I would rather AMD releases Polaris chip of the size of 1080 [~400mm2]. My budget does not allow purchase of full enthusiast cards.

 

[Zojirushi]

http://videocardz.com/59808/amd-vega-gpu-allegedly-pushed-forward-to-october

I don't see it. It's not like AMD could possibly be caught off guard by Nvidia's new cards.

Nothing beyond what anybody was already expecting was announced so why change plans now?

 

[Sora the Cat]

After being forced to switch from gtx 970 to an r9 390, I can say that I will support AMD with their new series, none of those old amd facts are true anymore, card isn't more noisy (although I think that would change within brands, this one is an MSI), it has the same exact temps my old 970 used to have (70° tops, again, MSI version) and power consumption isn't that much, surely not a deal breaker. Drivers are fine, last amd card I had was an 5770, and boy things have changed since then, amd software is a lot better now, I honestly find it just as easy to use as nvidia.

 

[JohnnyFootball]

I just want to know when Polaris is getting released and how it will perform.

I really dont want to buy another nvidia GPU.

I have long craved AMD making a competitive comeback.

 

[Adry9]

http://videocardz.com/59808/amd-vega-gpu-allegedly-pushed-forward-to-october

HBM2 in 2016? That's gonna be expensive.

I would rather AMD releases Polaris chip of the size of 1080 [~400mm2]. My budget does not allow purchase of full enthusiast cards.

Same, 300-350€ is the sweet point for me.

 

[McHuj]

If I were AMD, I would announce that Vega is coming this year even if it will be July.

That would at least hold me off from purchasing a 1070/1080 until Vega comes out.

 

[Tacklemacain]

I wonder if the smaller Vega 10 which should be relaesed as the 490(x) will use HBM 2.
Maybe it will use GD5x and only the new Furys will use HBM 2.

If not, there would be a huge gap in performance and price beteewn Vega and Poalris.

Or Polaris is way faster then the 390x or we get Vega 10 for 350€.

Either way, it doesnt realy add up right now.

 

[Adry9]

I wonder if the smaller Vega 10 which should be relaesed as the 490(x) will use HBM 2.
Maybe it will use GD5x and only the new Furys will use HBM 2.

If not, there would be a huge gap in performance and price beteewn Vega and Poalris.

A ""cheap""" Vega with GD5X would be great. Otherwise like you said there would be a huge gap.

 

[dgrdsv]

I would rather AMD releases Polaris chip of the size of 1080 [~400mm2]. My budget does not allow purchase of full enthusiast cards.

1080 is 333mm^2 and since one Vega should contain the same amount of SPs as Fiji you can be pretty certain that it will be in 350mm^2 range as well. I doubt that we'll see anything bigger than 400mm^2 on the new processes this year.

 

[Irobot82]

1080 is 333mm^2 and since one Vega should contain the same amount of SPs as Fiji you can be pretty certain that it will be in 350mm^2 range as well. I doubt that we'll see anything bigger than 400mm^2 on the new processes this year.

How could AMD fit 4096 SP's in under 400 but Nvidia could only fit 2560 in 333?

 

[icecold1983]

How could AMD fit 4096 SP's in under 400 but Nvidia could only fit 2560 in 333?

Theres more to a chip than just compute units. Those are actually rather small

 

[dgrdsv]

How could AMD fit 4096 SP's in under 400 but Nvidia could only fit 2560 in 333?

They are different architectures. GM200 has 3072 SPs while Fiji has 4096 SPs, the chip sizes are comparable and GM200 is faster. So I'd expect AMD to fit more SPs into even the same size let alone a bigger, ~400mm^2 one. But this would put that Vega GPU somewhere around Fury X performance wise and this may not be enough to beat even 1080, less that what's coming after it. So I'm hoping that Polaris will deliver some performance improvements from its architecture updates.

 

[ZOONAMI]

They are different architectures. GM200 has 3072 SPs while Fiji has 4096 SPs, the chip sizes are comparable and GM200 is faster. So I'd expect AMD to fit more SPs into even the same size let alone a bigger, ~400mm^2 one. But this would put that Vega GPU somewhere around Fury X performance wise and this may not be enough to beat even 1080, less that what's coming after it. So I'm hoping that Polaris will deliver some performance improvements from its architecture updates.

Oh come on dr rus, Fury x beats the 980 ti in quite a few bencmarks. To suggest AMD's next enthusiast gpu will be no better/somewhere around its current enthusiast gpu is silly.

I hope that a full polaris 10 chip will be close to or better than fury x / 980 ti, as some rumors have suggested. Otherwise amd has nothing to compete with the first pascal chips, which isn't a good thing for the market or consumers.

 

[RaijinFY]

imo the base Polaris 10 will compete with the 980 (non Ti) and have good overclocking capabilities that will allow it to come close to the 1070 performance.

 

[MuchoMalo]

They are different architectures. GM200 has 3072 SPs while Fiji has 4096 SPs, the chip sizes are comparable and GM200 is faster. So I'd expect AMD to fit more SPs into even the same size let alone a bigger, ~400mm^2 one. But this would put that Vega GPU somewhere around Fury X performance wise and this may not be enough to beat even 1080, less that what's coming after it. So I'm hoping that Polaris will deliver some performance improvements from its architecture updates.

That's a bold prediction to make without knowing the full specs or even just the clock speeds. Extremely bold. No way AMD is that far behind.

 

[chaosblade]

That's a bold prediction to make without knowing the full specs or even just the clock speeds. Extremely bold. No way AMD is that far behind.

AMD is so far behind they are even behind themselves!

 

[ZOONAMI]

AMD is so far behind they are even behind themselves!

Right. Lol. Next enthusiast chip same as current enthusiast chip. They aren't that incompetent. Not to mention nvidia is supposedly doing more performance with less shaders on pascal vs maxwell, and AMD will apparently not be able to do more performance with the same amount or more shaders on a new node. Vega will assuredly be a big step forward over fiji.

 

[Renekton]

http://videocardz.com/59808/amd-vega-gpu-allegedly-pushed-forward-to-october

I smell baby-wipes-soft launch

Will definitely buy it if at least 70% faster than my 290X

 

[icecold1983]

They are different architectures. GM200 has 3072 SPs while Fiji has 4096 SPs, the chip sizes are comparable and GM200 is faster. So I'd expect AMD to fit more SPs into even the same size let alone a bigger, ~400mm^2 one. But this would put that Vega GPU somewhere around Fury X performance wise and this may not be enough to beat even 1080, less that what's coming after it. So I'm hoping that Polaris will deliver some performance improvements from its architecture updates.

youre reaching pretty far dude

 

[Aztechnology]

http://videocardz.com/59808/amd-vega-gpu-allegedly-pushed-forward-to-october

Ohhhhh I'm getting excited. I hope this is true.

 

[grumble]

imo the base Polaris 10 will compete with the 980 (non Ti) and have good overclocking capabilities that will allow it to come close to the 1070 performance.

That is a fair assumption from what I've read. Not sure how well it will overclock but it being the equivalent of a 980 is reasonable.

 

[dgrdsv]

Oh come on dr rus, Fury x beats the 980 ti in quite a few bencmarks. To suggest AMD's next enthusiast gpu will be no better/somewhere around its current enthusiast gpu is silly.

I hope that a full polaris 10 chip will be close to or better than fury x / 980 ti, as some rumors have suggested. Otherwise amd has nothing to compete with the first pascal chips, which isn't a good thing for the market or consumers.

Well, a lot is dependent on how good Polaris improvements will be and how high AMD will be able to clock its 14nm chips.

I don't believe that P10 will be close to Fury X / 980Ti anymore because all I've heard recently puts it on a lower tier - which would be Hawaii/390/970. So it's only reasonable to expect Vega with the same amount of units as in Fiji to perform close to Fiji as well.

Fiji is some 10% behind 980Ti stock on average and 1080 should add some 20-25% to that so we're looking at 30%+ gap hence why I kinda doubt that this Vega is what you think it is - it may be a competition for 1080 but it's unlikely to beat it and it will come half a year later if that latest rumor is correct.

Basically it's not the enthusiast chip you think it is as it's die size is closer to what is called "performance" by AMD.

That's unless the last round of info on Polaris wasn't just a smokescreen of course.

 

[MuchoMalo]

Right. Lol. Next enthusiast chip same as current enthusiast chip. They aren't that incompetent. Not to mention nvidia is supposedly doing more performance with less shaders on pascal vs maxwell, and AMD will apparently not be able to do more performance with the same amount or more shaders on a new node. Vega will assuredly be a big step forward over fiji.

Actually, there isn't much of a performance-per-shader increase with Pascal; the clocks are just that much higher. The clocks and extra shaders compared to the 980 make the 1080 75% faster than the 980 on paper. The 1080 is advertised as being as fast as 980 SLI. It matches up almost perfectly, especially after considering the higher bandwidth and the other areas of improvement.

Well, a lot is dependent on how good Polaris improvements will be and how high AMD will be able to clock its 14nm chips.

I don't believe that P10 will be close to Fury X / 980Ti anymore because all I've heard recently puts it on a lower tier - which would be Hawaii/390/970. So it's only reasonable to expect Vega with the same amount of units as in Fiji to perform close to Fiji as well.

Fiji is some 10% behind 980Ti stock on average and 1080 should add some 20-25% to that so we're looking at 30%+ gap hence why I kinda doubt that this Vega is what you think it is - it may be a competition for 1080 but it's unlikely to beat it and it will come half a year later if that latest rumor is correct.

Basically it's not the enthusiast chip you think it is as it's die size is closer to what is called "performance" by AMD.

That's unless the last round of info on Polaris wasn't just a smokescreen of course.

I'm going to let you in on a secret: Pascal's high clocks were more or less entirely due to the node shrink, not because of architectural improvements. On top of that, Vega is supposed to be even more power efficient than Polaris overall, so unless Vega 10 cards are sub-150W, it's really unlikely that it'll just be on-par with Fiji. You're just making a ton of assumptions based on rumors, and then making those rumors seem worse than they are with your implication that a full Polaris 10 will be on-par with a 970. Everything points to it matching a 980 at worst.

 

[ZOONAMI]

Actually, there isn't much of a performance-per-shader increase with Pascal; the clocks are just that much higher. The clocks and extra shaders compared to the 980 make the 1080 75% faster than the 980 on paper. The 1080 is advertised as being as fast as 980 SLI. It matches up almost perfectly, especially after considering the higher bandwidth and the other areas of improvement.

1080 has 500 more shaders than a 980. 980 ti has more shaders. We don't even have any real benchmarks. The 1080 may not even be much better than a 980 ti. You and dr rus need to hold the phone. The suggestion of Vega, an enthusiast class chip, being near the previous generation fiji enthusiast chip, based having the same rumored shader count, is reaching. As a few other posters have pointed out.

Edit: good edit muchomalo.

 

[icecold1983]

Well, a lot is dependent on how good Polaris improvements will be and how high AMD will be able to clock its 14nm chips.

I don't believe that P10 will be close to Fury X / 980Ti anymore because all I've heard recently puts it on a lower tier - which would be Hawaii/390/970. So it's only reasonable to expect Vega with the same amount of units as in Fiji to perform close to Fiji as well.

Fiji is some 10% behind 980Ti stock on average and 1080 should add some 20-25% to that so we're looking at 30%+ gap hence why I kinda doubt that this Vega is what you think it is - it may be a competition for 1080 but it's unlikely to beat it and it will come half a year later if that latest rumor is correct.

Basically it's not the enthusiast chip you think it is as it's die size is closer to what is called "performance" by AMD.

That's unless the last round of info on Polaris wasn't just a smokescreen of course.

you really think amds next high end gpu on 14nm is going to be fury x level of performance? that they are just unable to beat it?

furyx is not 10% behind stock 980ti unless your looking at 1080p resolution where it hits cpu limitation.

 

[chaosblade]

It's not unreasonable to think that Vega should be as about as big of an improvement over Polaris as Hawaii is over Pitcairn. Those are the most comparable chips across generations, physically (based on the rumors so far).

 

[Renekton]

you really think amds next high end gpu on 14nm is going to be fury x level of performance? that they are just unable to beat it?

furyx is not 10% behind stock 980ti unless your looking at 1080p resolution where it hits cpu limitation.

It's not unreasonable to think that Vega should be as about as big of an improvement over Polaris as Hawaii is over Pitcairn. Those are the most comparable chips across generations, physically (based on the rumors so far).

Ya predicting 14nm big die Vega to have same performance as Fiji is just insane.

 

[frontieruk]

Oh com'on guys leave Dr_Rus alone, he's using historical evidence of performance gaps to make his predictions.

Unfortunately for him those gaps were driver overhead under dx11 the hardware has never really been that much weaker in performance (performance per watt probably though).


This we know due to the mantle experiment amd ran, which we are also seeing under DX12 removing the driver overhead amd had incurred from lack of development has allowed their cards a new lease on life.

If his prediction of similar performance to the Fury X holds true for Vega Amd dun goofed.

 

[icecold1983]

Oh com'on guys leave Dr_Rus alone, he's using historical evidence of performance gaps to make his predictions.

Unfortunately for him those gaps were driver overhead under dx11 the hardware has never really been that much weaker in performance (performance per watt probably though).


This we know due to the mantle experiment amd ran, which we are also seeing under DX12 removing the driver overhead amd had incurred from lack of development has allowed their cards a new lease on life.

If his prediction of similar performance to the Fury X holds true for Vega Amd dun goofed.

Theres actually no historical evidence supporting it. neither amd nor nvidia have ever released a high end gpu that wasnt faster than their previous high end gpu 2 node transitions larger

 

[Marlenus]

They are different architectures. GM200 has 3072 SPs while Fiji has 4096 SPs, the chip sizes are comparable and GM200 is faster. So I'd expect AMD to fit more SPs into even the same size let alone a bigger, ~400mm^2 one. But this would put that Vega GPU somewhere around Fury X performance wise and this may not be enough to beat even 1080, less that what's coming after it. So I'm hoping that Polaris will deliver some performance improvements from its architecture updates.

Polaris 10 will be at or near Fury X performance.

Scale based on power usage and AMDs released perf/watt improvement metric from hawaii, grenada, pitcairn or tonga gpus and you get Polaris 10 being faster than Fury X, account for a cherry picked number and you are basically exactly where Fury X is.

Scale based on perf/mm improvements from those same chips and you get the same results.

The only way P10 is much slower is if it is far smaller than rumoured at less than 200mm^2 or they decide to go for a really low TDP.

As we have seen from NV and Apple the 14nm process allows for around a 60% clockspeed increase so if a 400mm Vega GPU does have around 4k shaders it is also likely to be clocked at a conservative 1.3Ghz which would give it performance of Fury X + 30% without taking into account architectural improvements. One of the biggest issues with Fiji is that the front end is inadequate to feed the shader array. It has 45% more shaders than the 390X and gobs of bandwidth yet the performance delta tops out at 25% at 4k and is only 20% at 1440p. Increasing utilisation of the shaders could be good for another 10% bump in performance even if no other changes are made. That would mean we are talking Fury X + 40% as a quite conservative estimate which would be as fast as the 1080.

As for Fury X vs 980Ti, at 4k and 1440p they are about the same and at 1080p the 980Ti is faster. As more DX12 games come out and get added to benchmark suites I can see Fury start to pull away from the 980Ti the same way the 290X is pulling away from the 780Ti. source

 

[Durante]

I'm going to let you in on a secret: Pascal's high clocks were more or less entirely due to the node shrink, not because of architectural improvements.

I'd really love a source for that. These days, you can't just use a smaller process and expect to clock things much higher, you also need to work for it.

I'm very curious about Polaris/Vega clocks, I think there's a pretty broad range of where they could end up (and exactly where they do will directly determine how competitive they are).

 

[Marlenus]

I'd really love a source for that. These days, you can't just use a smaller process and expect to clock things much higher, you also need to work for it.

I'm very curious about Polaris/Vega clocks, I think there's a pretty broad range of where they could end up (and exactly where they do will directly determine how competitive they are).

Well the A7 was 1.3Ghz on 28nm and the A9 is 1.85Ghz on 14nm. This is about the same as the clockspeed uplift from Nvidia so is an indication that it is due to the change in node.

 

[Razgreez]

I'd really love a source for that. These days, you can't just use a smaller process and expect to clock things much higher, you also need to work for it.

Isn't the converse true. You normally do gain the ability to clock "much" higher (I wouldn't use the term over clock) if you were to hypothetically take the exact same architecture and shrink it down considerably - the shift from 28nm to 14nm is quite considerable after all. You do of course reach a point of diminishing returns which is where over-clocking actually comes in.

 

[Durante]

Isn't the converse true. You normally do gain the ability to clock "much" higher (I wouldn't use the term over clock) if you were to hypothetically take the exact same architecture and shrink it down considerably - the shift from 28nm to 14nm is quite considerable after all.

That used to be the case. As I understand it (I'm not a hardware engineer) these days, at these very small processes, if you just shrink down an existing architecture without adjustments you can be lucky if it even works.

Of course, no one would do that, and consequently clock improvements can still be expected. But it's not a matter of course to get huge ones, just look at Intel's clock rate stagnation over their past 3 (!) shrinks.

 

[Inspectah_Deck]

I'd really love a source for that. These days, you can't just use a smaller process and expect to clock things much higher, you also need to work for it.

I'm very curious about Polaris/Vega clocks, I think there's a pretty broad range of where they could end up (and exactly where they do will directly determine how competitive they are).

I think this is pretty much the case here.

Look at this chart by GF:

 

[Inspectah_Deck]

Rumor: AMD won't have partners showing Polaris cards at Computex

Nordic Hardware reached out to us with their report, saying that AMD's partners "won't have any new cards to display at Computex and the only Polaris cards promoted to them from AMD are R9 390/390X performance class but for a mid-range price. Great value but no sign of any GTX 1080 contender". I thought I'd reach out to an AMD partner, and received a quick "don't think so" from a board partner, and then I asked for clarification to which they said "as of now, no information".

 

[SneakyStephan]

I'd really love a source for that. These days, you can't just use a smaller process and expect to clock things much higher, you also need to work for it.

I'm very curious about Polaris/Vega clocks, I think there's a pretty broad range of where they could end up (and exactly where they do will directly determine how competitive they are).

From what I've read it's finfet's doing
much lower leakage allowing it to run at a lower voltage and clock much higher.

I expect amd to get the same clockspeed gains as nvidia. And they're going to need it with that piddly little 200mm² die...

on topic: it's interesting that october rumor for vega. Sounds weird to me because I thought the whole point of splitting polaris into polaris and vega was them having to wait for HBM2 volume production.
So either it's fake or vega will use gddr5x I guess?

Which makes it sound even weirder as how can they just change vega on a whim to use gddr5x instead of HBM, I thought you had to design parts of the chip around it.

Vega releasing with HBM2 in october sounds extremely far fetched to me, and if it happened it would be a paper launch with no meaningful supply till a few months later.

edit: and for why I think getting HBM2 earlier than expected is unlikely is because HBM2 is not suffering from delays anymore atm, it is on track , is about to enter volume production this summer and from there on it's just a matter of waiting for enough supply to build up to be able to release some gfx cards using it.

It's not like the timeframe was based on expected delays and things went better than expected so the timeframe is moved up.

 

[riflen]

That used to be the case. As I understand it (I'm not a hardware engineer) these days, at these very small processes, if you just shrink down an existing architecture without adjustments you can be lucky if it even works.

Of course, no one would do that, and consequently clock improvements can still be expected. But it's not a matter of course to get huge ones, just look at Intel's clock rate stagnation over their past 3 (!) shrinks.

I've not seen a quote from anyone at Nvidia, but AMD have said that the GF 14nm finFET process has given their chips greater consistency, which is allowing them to hit higher clocks.

With regards to Intel, I read that there are fundamental problems with electron flow when clock speeds go beyond a certain rate. To improve this you need more voltage but unless you have sufficient cooling then the heat breaks down the gates. It stands to reason that the kind of cooling systems that consumers have access to, given area and cost limitations, are only capable of dissipating so much heat. So here we are with this 4.5-5 Ghz compromise of power vs heat.
As you can probably tell, I'm not an engineer either.

 

[michaelius]

From what I've read it's finfet's doing
much lower leakage allowing it to run at a lower voltage and clock much higher.

I expect amd to get the same clockspeed gains as nvidia. And they're going to need it with that piddly little 200mm² die...

on topic: it's interesting that october rumor for vega. Sounds weird to me because I thought the whole point of splitting polaris into polaris and vega was them having to wait for HBM2 volume production.

So either it's fake or vega will use gddr5x I guess?

Which makes it sound even weirder as how can they just change vega on a whim to use gddr5x instead of HBM, I thought you had to design parts of the chip around it.

Vega releasing with HBM2 in october sounds extremely far fetched to me, and if it happened it would be a paper launch with no meaningful supply till a few months later.

Nothing suprising - it's usual AMD delay tactic just to put some confusion in buyers in hopes not everyone buys 1080 when it's released.

 

[dgrdsv]

I'm going to let you in on a secret: Pascal's high clocks were more or less entirely due to the node shrink, not because of architectural improvements. On top of that, Vega is supposed to be even more power efficient than Polaris overall, so unless Vega 10 cards are sub-150W, it's really unlikely that it'll just be on-par with Fiji. You're just making a ton of assumptions based on rumors, and then making those rumors seem worse than they are with your implication that a full Polaris 10 will be on-par with a 970. Everything points to it matching a 980 at worst.

Gee thanks. That's not true though and let me let you on a secret as well: Polaris is being built on a wholly different node than Pascal and just because of that you can't simply assume that the same clock gains will happen with Polaris as well. There are also other reasons which precludes this from happening.

Vega's additional power efficiency is coming from usage of HBM over GDDR which means basically nothing for the chip's performance itself. I also think that I was pretty clear that I don't expect the 4096 SP Vega to even use HBM2 - especially if it will come out in October - as that seems rather unnecessary for a part with Fiji's math performance.

Yes, I'm making a ton of assumptions, like everyone here do, including you as well.

you really think amds next high end gpu on 14nm is going to be fury x level of performance? that they are just unable to beat it?

I really think that the first Vega card will be a port of Fiji to 14LPE. That's all I think at the moment.

furyx is not 10% behind stock 980ti unless your looking at 1080p resolution where it hits cpu limitation.

Fury X is about 10% slower than stock 980Ti on average. Go check the benchmarks.

Polaris 10 will be at or near Fury X performance.

Polaris 10 is a part with less SPs than there are in Hawaii, how exactly do you expect it to be on Fury X performance levels and why does every piece of info we have on it puts it in 390/390X range? I mean, it's possible if they'll be able to clock P10 in a similar fashion to how GP104 clocks but considering the issues which GCN in general has with hitting higher clocks I'm not really expecting this. Would love to be wrong.