Support NeoGAF

[E-Cat]

So, assuming that Volta comes out sometime in Spring 2018, how is AMD even going to compete with that? I mean, they kind of already aren't in the highest-end, but it's only going to get worse with Volta. NVIDIA have at least a new architecture boasting 50% better power efficiency over Pascal, so they can scale the 12nm (16FF++) die up to 600mm^2+ quite easily; whereas Navi won't be out until 7nm most likely, and the 64 CU Vegas are already hitting over 300W TDP on a 484 mm^2 die! There's nowhere to go, at least not without a drastic overhaul.

 

[joesiv]

So, assuming that Volta comes out sometime in Spring 2018, how is AMD even going to compete with that? I mean, they kind of already aren't in the highest-end, but it's only going to get worse with Volta. NVIDIA have at least a new architecture boasting 50% better power efficiency over Pascal, so they can scale the 12nm (16FF++) die up to 600mm^2+ quite easily; whereas Navi won't be out until 7nm most likely, and the 64 CU Vegas are already hitting over 300W TDP on a 484 mm^2 die! There's nowhere to go, at least not without a drastic overhaul.

infinity fabric lol

 

[TC McQueen]

I've never been a fan of the default GPUs, they just look like big metal boxes. I'll probably wait and get a custom one which I assume will be clocked higher. The 56 will be a nice upgrade to my GTX 970.

Honestly, if there was a custom Vega 56 for $375 or lower on sale during Black Friday or post-Christmas sale, it would be compelling a compelling 970 replacement, but if Nvidia brings out an 1170/2070 at the same price as the 970, Vega 65 gets blown out of the water.

 

[DieH@rd]

Honestly, if there was a custom Vega 56 for $375 or lower on sale during Black Friday or post-Christmas sale, it would be compelling a compelling 970 replacement, but if Nvidia brings out an 1170/2070 at the same price as the 970, Vega 65 gets blown out of the water.

Nvidia will have nothing new this year. They can only drop the price of existing offering.

 

[SoundLad]

Nvidia will have nothing new this year. They can only drop the price of existing offering.

Why is that? Out of the loop so excuse if this has been asked previously.

 

[Sir Abacus]

So, assuming that Volta comes out sometime in Spring 2018, how is AMD even going to compete with that?

This is basically the elephant in the room. It took them two years to drag a GCN point release out the door and its only FP32 improvement is a mild boost in clock speed. All of that sweet Vega R&D went to not-really-gaming things so AMD can go get stomped by Nvidia in the datacenter and deep learning.

Their plan to compete in the gaming market appears to be buy off developers to replace as many FP32 shaders with FP16 versions for the Vega code path as possible.

 

[ViciousDS]

This doesn't appear (I'll reserve final judgement until reviews) to be AMD's greatest GPU release, but as usual, it's the same coterie of Nvidia fans going way OTT with their negative assessment of RX Vega before it has even released.

Let's remember these same 2-3 posters tried to convince everyone that Ryzen wasn't a great architecture at launch (remember that nonsense about CCX latency?). Final and seemingly total judgement made before it was released despite myself and others sensibly suggesting to wait and see. Fast forward 3 months and Ryzen is almost universally recommended. The R5 1600 the greatest bang-for-buck CPU out there aside from the super budget Intel G4560.

Vega isn't another Ryzen, but this thread is like someone died in here because of certain gatekeepers of the discussion.

Oh im not dismissing Vega.......its all about the actual field results obviously and Ryzen when it first launched was ok out the gate, but after a few BIOS updates my current setup smoked my 7600k setup by far. It's overall just an outstanding platform and bang for buck isn't even close to being questioned with Zen as the clear as day winner overall.

I'm just going to wait and see when more cards hit the market for how good they are in real life tests.

 

[Sir Abacus]

Why is that? Out of the loop so excuse if this has been asked previously.

Nvidia chips have settled on a pretty reliable and sensible cadence for the release strategy. First customers who get new toys are the big iron guys who pay 100K per dev board. Yields are low, working through yield issues takes time, they buy basically everything Nvidia can produce. After that come the mass production variants. Usually codenamed some variant of Gxyy4, these are separate cutdown versions of the full dies which have the fully enabled variant (x80) and the ones with disabled hardware to improve yields (x70). As yields improve with full dies you see the big variants of chips ramp up. First it goes to Titans. Then, later, the x80 Ti variants.

Right now, extrapolating based on previous chips, and the traditional timing of each Nvidia GPU segment you can make an educated guess that the earliest we're going to see any sort of GV104 variant is early next year. Maybe straight into the new year and launch at CES. If I was a betting man that would be where my $20 would go for the 1170 and 1180 (or whatever they end up calling them).

 

[JohnnyFootball]

This doesn't appear (I'll reserve final judgement until reviews) to be AMD's greatest GPU release, but as usual, it's the same coterie of Nvidia fans going way OTT with their negative assessment of RX Vega before it has even released.

Let's remember these same 2-3 posters tried to convince everyone that Ryzen wasn't a great architecture at launch (remember that nonsense about CCX latency?). Final and seemingly total judgement made before it was released despite myself and others sensibly suggesting to wait and see. Fast forward 3 months and Ryzen is almost universally recommended. The R5 1600 the greatest bang-for-buck CPU out there aside from the super budget Intel G4560.

Vega isn't another Ryzen, but this thread is like someone died in here because of certain gatekeepers of the discussion.

At this point, it would be a monumental and epic shock if Vega can outperform the 1080.
Ryzen on the otherhand is the gift that keeps on giving and getting better as more and more software/BIOS optimizations are arriving.

 

[Bolivar687]

Ryzen on the otherhand is the gift that keeps on giving and getting better as more and more software/BIOS optimizations are arriving.

That's been the case for every Radeon graphics card for the last four years.

 

[ethomaz]

That's been the case for every Radeon graphics card for the last four years.

Every hardware you mean... that is always the case with some perf. gains with better drivers.

But it never make a hardware with perf. around 1080 reaches 1080ti.

 

[Bolivar687]

Every hardware you mean... that is always the case with some perf. gains with better drivers

In general, Radeon cards over the last four years have almost consistently overtaken and then extended their leads over their more expensive Nvidia counterparts.

 

[thelastword]

what i really can't wrap my head around is why the vega frontier edition is beating the 1080 handily in basically every rendering or CAD Software out there and even outdo the TitanXP in some by a huge amount while being this shitty in games. the workloads should be quite similar and yet vega can't convert it's flop-advantage in effective performance in games.that doesn't make sense to me.

I think things will improve drastically, the focus of the FE is for PRO apps, but there's no reason why it can't be great for gaming as well, just like the TItan XP. I think gaming drivers are the main issue there, but as it stands the FE is a great card for it's target market....

Drivers... Rx Vega ain't even out and people are making assumptions. The frontier edition is based on precision and is a developers card. It doesn't even have the gaming mode available until rx Vega drops. Relax

Thanks.....No doubt, even AMD engineers have indicated some changes and improvements will be made across all vega cards. They needed to hit a launch target, they did. Now it's all about improving drivers for Rx Vega's launch and thereafter. Funny thing about AMD is they always have powerful hardware, but they don't always have the best drivers day one. Even then they have improved massively in that area and their gpu apps are the best right now bar none.

Since we're on that, I hope that the new version of Open GL (4.6) runs better on AMD hardware, NV always had the advantage there, but with this new vega architecture, things are looking up. I think too that more people are enthused for Vega because of it's features, improved pixel pipeline (engine), rapid packed math and hbcc. If there are more vega's out there, more developers will make games based on it's architecture and that's where it will beat NV handily...As it stands, AMD announced a plan to collaborate with many developers on upcoming games, so I'm really stoked for the hardware and it's capabilities.

This doesn't appear (I'll reserve final judgement until reviews) to be AMD's greatest GPU release, but as usual, it's the same coterie of Nvidia fans going way OTT with their negative assessment of RX Vega before it has even released.

Let's remember these same 2-3 posters tried to convince everyone that Ryzen wasn't a great architecture at launch (remember that nonsense about CCX latency?). Final and seemingly total judgement made before it was released despite myself and others sensibly suggesting to wait and see. Fast forward 3 months and Ryzen is almost universally recommended. The R5 1600 the greatest bang-for-buck CPU out there aside from the super budget Intel G4560.

Vega isn't another Ryzen, but this thread is like someone died in here because of certain gatekeepers of the discussion.

Truer words have not been spoken. It's funny that some people are trying to undermine the Vega 56 to a GTX 1060....I'm also a believer that Ryzen+Vega will be superior to Intel + Vega, especially on the upcoming titles like Farcry 5 and Wolfenstein. Ryzen 1600 + 3600Mhz Ram + Vega should be quite a beastly propositiion for gaming in a few days...

 

[DemonCleaner]

. NVIDIA have at least a new architecture boasting 50% better power efficiency over Pascal

so where is this information coming from? that would be highly unlikely even if it was on half the node size of pascal, which 12nm FF is not in the least (it basically has the same transistor density as before)

 

[dgrdsv]

So, assuming that Volta comes out sometime in Spring 2018, how is AMD even going to compete with that? I mean, they kind of already aren't in the highest-end, but it's only going to get worse with Volta. NVIDIA have at least a new architecture boasting 50% better power efficiency over Pascal, so they can scale the 12nm (16FF++) die up to 600mm^2+ quite easily; whereas Navi won't be out until 7nm most likely, and the 64 CU Vegas are already hitting over 300W TDP on a 484 mm^2 die! There's nowhere to go, at least not without a drastic overhaul.

This is the big elephant in the room here. If AMD won't have Navi till 2019 then they'll have to compete with NV's next update - whenever that will happen but probably sooner than it took Vega to come out after Pascal - with Polaris and Vega, and chances are high that the next GFx70 card will beat Vega 64 pushing it from somewhat okay'ish $500-700 all the way down to $300 or less.

My only hope is that with Ryzen being successfully launched they will be able to re-allocate their R&D budget to GPUs now and finally move to a new architecture in Navi - although I do wonder if it may already be too late for Navi considering that it should launch in about two years.

so where is this information coming from? that would be highly unlikely even if it was on half the node size of pascal, which 12nm FF is not in the least (it basically has the same transistor density as before)

GV100 peak flops are roughly 50% higher than GP100 peak flops in the exact same power consumption envelope.

 

[DemonCleaner]

GV100 peak flops are roughly 50% higher than GP100 peak flops in the exact same power consumption envelope.

that would be 33,3% efficiency gain over Pascal, wouldn't it?

 

[Aztechnology]

That's been the case for every Radeon graphics card for the last four years.

There's no denying long term GCN architecture has been better. But it's usually not worth the long term if you're someone who upgrades fairly frequently.

 

[ethomaz]

that would be 33,3% efficiency gain over Pascal, wouldn't it?

That means 50% efficiency over Pascal.

 

[TSM]

There's no denying long term GCN architecture has been better. But it's usually not worth the long term if you're someone who upgrades fairly frequently.

People always bring up the "fine wine" theory of AMD performance increasing over time, but I've never figured out how not getting the full potential of your hardware on day one is supposed to be a positive. Nvidia on the other hand tends to release hardware performing near max on day one which is then somehow supposed to be a negative?

 

[Aztechnology]

People always bring up the "fine wine" theory of AMD performance increasing over time, but I've never figured out how not getting the full potential of your hardware on day one is supposed to be a positive. Nvidia on the other hand tends to release hardware performing near max on day one which is then somehow supposed to be a negative?

It's definitely good too, especially for those we plan to keep their cards for a longer time. But it doesn't fit with my style. Also it's not so much that they get better with time. They just hold up better in the long term. While Nvidia tends to release devices past two years or so to legacy support essentially.

May just be a factor of the architecture changes with AMD (GCN) not being as diverse so the drivers generally continue to be supported longer.

 

[saskuatch]

People always bring up the "fine wine" theory of AMD performance increasing over time, but I've never figured out how not getting the full potential of your hardware on day one is supposed to be a positive. Nvidia on the other hand tends to release hardware performing near max on day one which is then somehow supposed to be a negative?

The point is if you spend 500 now for Nvidia or 500 now for and. In future that 500 and will be outperforming Nvidia by x%....

 

[TSM]

The point is if you spend 500 now for Nvidia or 500 now for and. In future that 500 and will be outperforming Nvidia by x%....

Yeah, but if you spend $500 now you don't buy the card in the future. You buy the card in the present. Underperforming at launch is being spun as a positive aspect of the hardware.

May just be a factor of the architecture changes with AMD (GCN) not being as diverse so the drivers generally continue to be supported longer.

You eventually get all the performance that was left on the table for the months or years before they finally optimized the drivers for the hardware.

 

[ethomaz]

It's definitely good too, especially for those we plan to keep their cards for a longer time. But it doesn't fit with my style. Also it's not so much that they get better with time. They just hold up better in the long term. While Nvidia tends to release devices past two years or so to legacy support essentially.

May just be a factor of the architecture changes with AMD (GCN) not being as diverse so the drivers generally continue to be supported longer.

Are there any comprehensive test with launch drivers plus 1 year, 2 year, etc drivers for both companies???

 

[kirankara78]

People always bring up the "fine wine" theory of AMD performance increasing over time, but I've never figured out how not getting the full potential of your hardware on day one is supposed to be a positive. Nvidia on the other hand tends to release hardware performing near max on day one which is then somehow supposed to be a negative?

Sure it's a nice bonus seeing your card performing better several years later.

But that's only great if you a) intend to keep card from more than few years and b) if you're happy to have sub optimum performance for first few years.

The former and latter are definitely not me

 

[kirankara78]

It's definitely good too, especially for those we plan to keep their cards for a longer time. But it doesn't fit with my style. Also it's not so much that they get better with time. They just hold up better in the long term. While Nvidia tends to release devices past two years or so to legacy support essentially.

May just be a factor of the architecture changes with AMD (GCN) not being as diverse so the drivers generally continue to be supported longer.

Think it says more about the budget AMD have for driver support.

They've definitely improved recently, and i don't expect any great developments this time around. We have not really seen the "fine wine" effect in last couple generations of cards.

There was the big dx11 jump but since then it's been more subtle

 

[Rushersauce]

The magical driver dream.

It has worked before (not with Fiji, though)

 

[Shin]

So, assuming that Volta comes out sometime in Spring 2018, how is AMD even going to compete with that? I mean, they kind of already aren't in the highest-end, but it's only going to get worse with Volta.

Vega refresh (7nm), Navi won't be out until 2019 and that's if they get it together unlike the whole Vega mess (I lost count of the delays).
Until then they have to ride this out, if they can't win on performance then it will have to be price/performance.

 

[saskuatch]

Yeah, but if you spend $500 now you don't buy the card in the future. You buy the card in the present. Underperforming at launch is being spun as a positive aspect of the hardware.



You eventually get all the performance that was left on the table for the months or years before they finally optimized the drivers for the hardware.

Except the 500 Nvidia equivalent now would be performing on par not better now. If it can match 1070 price and performance now with possibility of performing better in the future why would you buy a 1070.

 

[ethomaz]

Vega refresh (7nm), Navi won't be out until 2019 and that's if they get it together unlike the whole Vega mess (I lost count of the delays).
Until then they have to ride this out, if they can't win on performance then it will have to be price/performance.

7nm won't be out before 2019.

 

[SRG01]

Truer words have not been spoken. It's funny that some people are trying to undermine the Vega 56 to a GTX 1060....I'm also a believer that Ryzen+Vega will be superior to Intel + Vega, especially on the upcoming titles like Farcry 5 and Wolfenstein. Ryzen 1600 + 3600Mhz Ram + Vega should be quite a beastly propositiion for gaming in a few days...

I don't see how a 1060 could even compare to a Vega 56, if the Vega 56 can outperform the 1070 in leaps and bounds.

At any rate, there are essentially two problems with Vega right now: one is the power consumption and the other is translating compute power into actual gaming performance. We've seen considerable gains with DX12/Vulkan on AMD cards, but AMD needs to do more than just rely on code paths to solve future performance issues.

There is, of course, a third problem, and that is whether choosing HBM was the right choice for AMD.

 

[rtcn63]

I've only seen like one person so far confuse the 56 with the 1060 (performance-wise).

 

[Drakhoran]

7nm won't be out before 2019.

It's theoretically possible:

“Our 7nm FinFET technology development is on track and we are seeing strong customer traction, with multiple product tapeouts planned in 2018,” said Gregg Bartlett, senior vice president of the CMOS Business Unit at GF.

 

[opticalmace]

I think the Vega 56 is basically gonna be 1070 performance +- 5%.

 

[Delusibeta]

Interesting point: AMD's own Battlefield 1 benchmarks vs the Fury X show a 20%-35% performance increase, depending on the resolution. This is despite the RX Vega operating at a 50% higher clock rate.

 

[Unknown Soldier]

The point is if you spend 500 now for Nvidia or 500 now for and. In future that 500 and will be outperforming Nvidia by x%....

But the Nvidia card was available in May 2016 and the AMD card isn't available yet now in August 2017 so if you bought Nvidia you would have been gaming for a whole year now. What's the value of a year of your mortal lifespan waiting instead of gaming?

Actually I bought a 1080, sold it and got most of my money back, and bought a 1080 Ti while the AMD true faithful have been waiting and waiting.

 

[THE:MILKMAN]

It's theoretically possible:

I don't doubt tape outs will happen in 2018, but don't expect an actual product in consumer hands until ~12-18 months later.

This article was pretty spot on for Zen: http://techfrag.com/2015/12/09/conf...aped-out-yet-reveals-chief-financial-officer/

Kumar further revealed that it takes around 12-14 months from the time you tape out the products to when you start shipping the products. That means, realistically we are looking at about a couple months to begin taping out, plus another ~14 months, which puts the consumer release of Zen at about April of 2017!

 

[J-Rzez]

AMD is just too far out time scale wise. They need to be closer in releases. Now I feel nvidia will be ahead of them again with their next cards.

In not very found of their bundles to defeat miners ploy.

Finally, sure, AMD drivers improve over the life of the card, but I'd prefer better performance now with games coming out now.

Looks a bit more power hungry than expected too. Regardless can't wait for the tests.

 

[shouamabane]

So would the Vega 56 future-proof me any more than a 1070?

I like to basically build a PC and then not touch anything for several years.

 

[ethomaz]

It's theoretically possible:

These tape out are more like engineer samples... not product releases.

But let's see they said 12-18 months that fits late 2018 at best.

 

[StrongBlackVine]

So would the Vega 56 future-proof me any more than a 1070?

I like to basically build a PC and then not touch anything for several years.

You probably want a top of line card if that is your goal.

 

[icecold1983]

So would the Vega 56 future-proof me any more than a 1070?

I like to basically build a PC and then not touch anything for several years.

Id assume so. It will probably already be slightly faster at launch and you have the few games that will get a 10%ish boost from fp16 and the ongoing fact that both consoles use gcn. Lower level apis will also continue to see increased usage which will be another benefit for amd. However if money is no issue a titan xp or 1080ti will future proof you the longest

 

[gabbo]

You probably want a top of line card if that is your goal.

As someone who needs to update from a GTX660, I'm wondering if the Vega56 would be worth it since I too don't update often. I'm 1080p so I don't need 4k bells and whistles, but I want to be able to do high/ultra settings at that 1080p and I'm still not sure based on what I've read if Vega56 will do that?

 

[rtcn63]

As someone who needs to update from a GTX660, I'm wondering if the Vega56 would be worth it since I too don't update often. I'm 1080p so I don't need 4k bells and whistles, but I want to be able to do high/ultra settings at that 1080p and I'm still not sure based on what I've read if Vega56 will do that?

Both the 1070 and 56 (if it does perform at the same level) should theoretically be able to do 1080p/60/high at least for some years. If anything, it's likely your CPU that might set you back depending on what you currently have and the types of games you play (open-world and RTS tend to be particularly CPU-demanding).

 

[Sir Abacus]

As someone who needs to update from a GTX660, I'm wondering if the Vega56 would be worth it since I too don't update often. I'm 1080p so I don't need 4k bells and whistles, but I want to be able to do high/ultra settings at that 1080p and I'm still not sure based on what I've read if Vega56 will do that?

A 580 will do that for half the price.

 

[FingerBang]

As someone who needs to update from a GTX660, I'm wondering if the Vega56 would be worth it since I too don't update often. I'm 1080p so I don't need 4k bells and whistles, but I want to be able to do high/ultra settings at that 1080p and I'm still not sure based on what I've read if Vega56 will do that?

I got a 1060 while waiting for Vega/Volta and it performs much better than I thought, for a cheaper price. Wouldn't go for a Vega for 1080p @60fps. 1060 and 580 are going to be enough for a while.

 

[Sami+]

Any word on when reviews and pre-orders will start popping up? I want to get the $599 bundle on Amazon.

 

[gabbo]

I got a 1060 while waiting for Vega/Volta and it performs much better than I thought, for a cheaper price. Wouldn't go for a Vega for 1080p @60fps. 1060 and 580 are going to be enough for a while.

If a 1060/580 can max Witcher3 on Ultra at 1080, I'll definitely look into one, always assumed the 1070 might be the baseline I'd be looking for or the AMD equivalent

 

[icecold1983]

Any word on when reviews and pre-orders will start popping up? I want to get the $599 bundle on Amazon.

Reviews on 14th. Dunno about availability

 

[FingerBang]

If a 1060/580 can max Witcher3 on Ultra at 1080, I'll definitely look into one, always assumed the 1070 might be the baseline I'd be looking for or the AMD equivalent

Absolutely yes, with everything at max and hairworks off. Hairworks on is probably going to bring you in the 50s.

 

[Unknown Soldier]

If a 1060/580 can max Witcher3 on Ultra at 1080, I'll definitely look into one, always assumed the 1070 might be the baseline I'd be looking for or the AMD equivalent

You'll probably want to look at a 1070 if you're hellbent on maxing every setting in TW3 and having Hairworks on too in 1080p while still getting a locked 60 fps.

Unfortunately the miners have been buying up the 1070's in a lot of places lately. Depending on where you are, you might as well just get a 1080 instead because the miners have driven 1070 prices up so much.