AMD has not announced all of Vega's architectural improvements

[AngryGoldfish]

It is possible that Vega 10 was deliberately shown in its weaker 'form' so that they don't suffer the same defeat as the previous generation where the Fury X was supplanted by the 980ti. If AMD puts their weaker foot forward, Nvidia might lower the performance of their 1080ti (to increase profits) or raise the price of their 1080ti (again, to raise profits). Then when actual Vega 10 comes out, and it's far beyond the original CES teaser, Nvidia is in the awkward position of having an $800 GPU that only performs around the same as a $600-700 GPU.

 

[Shingara]

It is possible that Vega 10 was deliberately shown in its weaker 'form' so that they don't suffer the same defeat as the previous generation where the Fury X was supplanted by the 980ti. If AMD puts their weaker foot forward, Nvidia might lower the performance of their 1080ti (to increase profits) or raise the price of their 1080ti (again, to raise profits). Then when actual Vega 10 comes out, and it's far beyond the original CES teaser, Nvidia is in the awkward position of having an $800 GPU that only performs around the same as a $600-700 GPU.

Well the fury x seems to be getting kicks up the bum out of thin air just via driver optimisation, same with the 480's. And something that amd flashed up with there simple render could be something they are putting into vega to combat what i can only see as the planned obsolescence that nvidia have within the generations of cards.

Most the performance hits we see in games are simply from how the work is pushed and the shear number of polys that stuff is trying to show, or not show via generating polys in areas you cant even see.

One thing i noticed though is that the vegas that were running were in cases that were taped up, the gpus had gaffer tape on them also so if someone did take the sides off that you couldnt see the power delivery.

And as cards these days run on thermal throttling then i dont think taping everything up helps a great deal.

Edit, what i was leaning towards with the simple render is that if it can do it but it doesnt have the drivers to do it then it might not have been working at ces.

 

[A7ibaba]

Im so excited to see whats new. I prepared 500 $ in cash and credit card. I prefer 24 month payments for the rest )

 

[AngryGoldfish]

Well the fury x seems to be getting kicks up the bum out of thin air just via driver optimisation, same with the 480's. And something that amd flashed up with there simple render could be something they are putting into vega to combat what i can only see as the planned obsolescence that nvidia have within the generations of cards.

Most the performance hits we see in games are simply from how the work is pushed and the shear number of polys that stuff is trying to show, or not show via generating polys in areas you cant even see.

One thing i noticed though is that the vegas that were running were in cases that were taped up, the gpus had gaffer tape on them also so if someone did take the sides off that you couldnt see the power delivery.

And as cards these days run on thermal throttling then i dont think taping everything up helps a great deal.

Edit, what i was leaning towards with the simple render is that if it can do it but it doesnt have the drivers to do it then it might not have been working at ces.

The Fury X and 480 cards have received big boosts in performance from drivers, but that doesn't change how poorly they performed at launch. Well, the RX 480 was fine, but the Fury X wasn't. AMD have been known to mature with age, going back to the 7000 series and likely before. But the problem with this is, all those day-one reviews that applaud Pascal will not be shared with Vega. That will mire the launch and forces those who waited patiently for Vega to finally cave in and buy the now cheaper Nvidia counterparts. Vega 10 has to be good out of the gate and become great over time. It can't be mediocre out of the gate and become good over time. The GPU will already be a year later than the competition.

 

[Shingara]

The Fury X and 480 cards have received big boosts in performance from drivers, but that doesn't change how poorly they performed at launch. Well, the RX 480 was fine, but the Fury X wasn't. AMD have been known to mature with age, going back to the 7000 series and likely before. But the problem with this is, all those day-one reviews that applaud Pascal will not be shared with Vega. That will mire the launch and forces those who waited patiently for Vega to finally cave in and buy the now cheaper Nvidia counterparts. Vega 10 has to be good out of the gate and become great over time. It can't be mediocre out of the gate and become good over time. The GPU will already be a year later than the competition.

Well when vega drops /shrug, rumors abound that its june and that nvidia are gonna drop the ti at the same time as a fire with fire tactic. Point i was making is that in the past amd cards suffered from the way games run things like tesselation, throwing the work at the cpu instead of gpu and rendering stuff that has no need tobe rendered, looking at you oceans under the floor.

And Vega on paper negates this, it shouldnt render stuff that you cant see, it shouldnt run into situations where it doesnt have enough vram, card gets capped, it should be able to use the vram on another card for itself or m.2 memory.

And vega isnt being released in a vacumn, its paired with ryzen, so its not just on how powerful vega is, but how it will play with ryzen.

 

[medi01]

Nah, just more architecture stuff that won't be supported properly until the next, next, next gen cards come out and devs decide it's worth utilising.. looks good on paper but at the end of the day unless it's supported and fully functional it's just idle hardware.

Yeah, like it was with 480 vs 1060... Oh wait it took 480 only 4 month to take over.

 

[pixell]

My first post on OC3D forum so hello to everyone.

My opinion on Vega is that AMD has to make it look better and not just faster than 1080 Ti. Getting a card year later in 1080 performance range would be just ridiculous. So I'm thinking they are buying some more time and sandbagging to shape up the product and avoid any bad scenario they somehow managed to hit over the last few years.

Points we know and don't know so far:

Die size - The only info we have is from the pictures of Raja's holding the Vega stack at CES. No idea about the model (smaller, bigger one) albeit looking at 2 HBM modules it should be the 8GB version presented and tested with DOOM 4K Vulcan. Size estimated from 470 to 530mm2.

Transistor No. - twice as Polaris?

Power consumption - some rumors indicate that VEGA test machines at CES had 2x8 pin power connectors but since this is still early silicon and power tweaking is ongoing this is most likely not the final configuration (stability before performance approach).

CU and SP - allegedly CU count is 64 but the number of SPs within is still unknown. It's been almost a year since the 4096 SP Vega count was leaked but no info about "powerful next gen V9 GCN cores" and their performance.

Frequency - IF the card has ~ 12 TF and IF the core count is 4096 clocks should be around 1.5 GHz which is a step forward comparing to Polaris but not enough IMHO. That would end in 50% Fury X performance increase (which DOOM demo is implying) but still in the 1080 courtyard.

This that leaked are not very promising so far. So AMD please better start pulling rabbits from the hats soon. Oh, and I hope reference cards fiasco they have with Hawaii and Polaris 20 will be avoided this time

 

[NeverBackDown]

Disagree with basically everything you said.
It'll be 16GB of HBM2.
Die size is useless. In addition when AMD measures it, It includes the interposer, which means it'll be way over inflated to what it really is. Also in comparison to the picture from the shot that was taken to a Fury X die, it looks smaller. A more advanced nm and smaller interposer is why it's smaller.

No point in speculation for the transistor count. We honestly have no clue...
Power consumption will probably be 225~250. High? Sorta. But power efficiency is way up. That's what important here.

CU and SP, well it's not really the same as before now is it? They have never really been comparable. Even if it's 4096, it'll still blow the Fury X out of the water.

1500mhz is pretty damn good for this architecture. GCN has always been a power brute, allowing it to clock much higher is just going to further help in all areas, especially for removing any bottleneck from within the architecture itself(HBM2 also helps here)

All the things that have been leaked is exactly what AMD needed to do in the GPU market. Instead of brute forcing it's way through everything, it's now working smarter and using its brute strength to do the things it actually needs to do much faster. I'm very excited for what they got to offer.

 

[Shingara]

Disagree with basically everything you said.
It'll be 16GB of HBM2.
Die size is useless. In addition when AMD measures it, It includes the interposer, which means it'll be way over inflated to what it really is. Also in comparison to the picture from the shot that was taken to a Fury X die, it looks smaller. A more advanced nm and smaller interposer is why it's smaller.

No point in speculation for the transistor count. We honestly have no clue...
Power consumption will probably be 225~250. High? Sorta. But power efficiency is way up. That's what important here.

CU and SP, well it's not really the same as before now is it? They have never really been comparable. Even if it's 4096, it'll still blow the Fury X out of the water.

1500mhz is pretty damn good for this architecture. GCN has always been a power brute, allowing it to clock much higher is just going to further help in all areas, especially for removing any bottleneck from within the architecture itself(HBM2 also helps here)

All the things that have been leaked is exactly what AMD needed to do in the GPU market. Instead of brute forcing it's way through everything, it's now working smarter and using its brute strength to do the things it actually needs to do much faster. I'm very excited for what they got to offer.

I agree with this, amd arnt trying to brute force performance, not saying there being weak but they are using there nogging, they are implimenting stuff in vega that negates specific things that have held amd back in the past. I think the main component of that is the primitive shader.

 

[Dark NighT]

My first post on OC3D forum so hello to everyone.

My opinion on Vega is that AMD has to make it look better and not just faster than 1080 Ti. Getting a card year later in 1080 performance range would be just ridiculous. So I'm thinking they are buying some more time and sandbagging to shape up the product and avoid any bad scenario they somehow managed to hit over the last few years.

Points we know and don't know so far:

Die size - The only info we have is from the pictures of Raja's holding the Vega stack at CES. No idea about the model (smaller, bigger one) albeit looking at 2 HBM modules it should be the 8GB version presented and tested with DOOM 4K Vulcan. Size estimated from 470 to 530mm2.

Transistor No. - twice as Polaris?

Power consumption - some rumors indicate that VEGA test machines at CES had 2x8 pin power connectors but since this is still early silicon and power tweaking is ongoing this is most likely not the final configuration (stability before performance approach).

CU and SP - allegedly CU count is 64 but the number of SPs within is still unknown. It's been almost a year since the 4096 SP Vega count was leaked but no info about "powerful next gen V9 GCN cores" and their performance.

Frequency - IF the card has ~ 12 TF and IF the core count is 4096 clocks should be around 1.5 GHz which is a step forward comparing to Polaris but not enough IMHO. That would end in 50% Fury X performance increase (which DOOM demo is implying) but still in the 1080 courtyard.

This that leaked are not very promising so far. So AMD please better start pulling rabbits from the hats soon. Oh, and I hope reference cards fiasco they have with Hawaii and Polaris 20 will be avoided this time

Im actually hoping amd puts the top gpu in the sub 600 euro price bracket and that the performance is close or as good as a 1080 or 1080 ti, i dont need amd to to beat nvidia on performance, we need amd to beat nvidia on price for said performance, because nvidia with their prices and shenanigans with geforce experience, worse driver support and all the while just charging a premium on a cooler that is yeah nice looking and reasonably quiet, it is worse then any of the respected 3rd party coolers out there.. yeah no, stats and specs are nice and we need to know whats under the hood, but amd beating nvidia on performance is not where it should be.

On top of that, its so incredibly unreasonable for the community to expect that amd must outperform nvidia, they need to sell a ton of good value hardware first before they can even begin to make an effort.

 

[The_Governour]

Its funny how all are speculating and dissagreeing about this, even though all of you have NO real reference to base your asumptions on. Its only what every one THINKS on their own opinion.

The only imporant thing for me will be price/performance ONCE 3rd party benches have been done. I hope that its the same or around the GTX 1080 mark. That would.be enough as AMD is famous for selling smoke like saying *Fury X is the overclockers dream*

 

[NeverBackDown]

Its funny how all are speculating and dissagreeing about this, even though all of you have NO real reference to base your asumptions on. Its only what every one THINKS on their own opinion.

The only imporant thing for me will be price/performance ONCE 3rd party benches have been done. I hope that its the same or around the GTX 1080 mark. That would.be enough as AMD is famous for selling smoke like saying *Fury X is the overclockers dream*

I have pretty good sources and they don't come from any tech sites. They come from Stock sites who the people who write those articles are much more inside the industry than any tech site.

In addition I've been right about every launch since the Fury X came out. So I'm pretty confident in myself

And no it needs to be faster than a 1080 or at least as fast as a 1080ti. The cheaper slightly slower alternative is a strategy they have used for a long time and it's not working. They need to take a page from Nvidia and care first about margins to get afloat.

 

[AngryGoldfish]

I have pretty good sources and they don't come from any tech sites. They come from Stock sites who the people who write those articles are much more inside the industry than any tech site.

In addition I've been right about every launch since the Fury X came out. So I'm pretty confident in myself

And no it needs to be faster than a 1080 or at least as fast as a 1080ti. The cheaper slightly slower alternative is a strategy they have used for a long time and it's not working. They need to take a page from Nvidia and care first about margins to get afloat.

And these sources provide specifics or are you embellishing on what they are saying? Because with your specifications there are many unanswered questions.

 

[NeverBackDown]

And these sources provide specifics or are you embellishing on what they are saying? Because with your specifications there are many unanswered questions.

They give very reasonable specs. If they gave out bogus specs like TT then they would not be a good source now would it?
There will always be unanswered questions until we hear from AMD. Whether you agree with me or not, that won't change.

 

[AngryGoldfish]

They give very reasonable specs. If they gave out bogus specs like TT then they would not be a good source now would it?
There will always be unanswered questions until we hear from AMD. Whether you agree with me or not, that won't change.

It's not that I disagree. Your word and your sources are better than mine. It's just, 16GB seems a lot for a card that employs High-Bandwidth Caching. 8GB is already sufficient in the gaming space and HBC furthers that. 16GB just seems... gratuitous.

 

[NeverBackDown]

It's not that I disagree. Your word and your sources are better than mine. It's just, 16GB seems a lot for a card that employs High-Bandwidth Caching. 8GB is already sufficient in the gaming space and HBC furthers that. 16GB just seems... gratuitous.

Well the thing is, I am fairly certain they are not directly connected. It's hard to tell based off the leaked slides we've seen because it's not specifically stated nor is it implied by the picture. We don't really know how it's going to workout. 16GB would be for the top end card. Everything else will have 8GB. Just seems silly for them to match a 1080s VRAM when that released so long ago

 

[The_Governour]

I have pretty good sources and they don't come from any tech sites. They come from Stock sites who the people who write those articles are much more inside the industry than any tech site.

In addition I've been right about every launch since the Fury X came out. So I'm pretty confident in myself

And no it needs to be faster than a 1080 or at least as fast as a 1080ti. The cheaper slightly slower alternative is a strategy they have used for a long time and it's not working. They need to take a page from Nvidia and care first about margins to get afloat.

I am the first one that wants amd cards to be good and better than Nvidias, but SO FAR they have been behind quite some time now. And untill we have some solid reviews before us then its ALL speculations, nothing more.

My 2 cents are that it will be equal OR less than a 1080. Hope I am wrong.

 

[RobM]

I agree with NBD and also why would there not be 16 on top gpu?
We know from leaks the the green team are going with 10/12
I really do think AMD have been showcasing something less than the best because its that good.

 

[AngryGoldfish]

Well the thing is, I am fairly certain they are not directly connected. It's hard to tell based off the leaked slides we've seen because it's not specifically stated nor is it implied by the picture. We don't really know how it's going to workout. 16GB would be for the top end card. Everything else will have 8GB. Just seems silly for them to match a 1080s VRAM when that released so long ago

But is Vega 10 really the top-end GPU? Vega 20 is rumoured to be a shrunken Vega 10 GPU with higher clock speeds and 16GB of VRAM. Maybe that's where the 16GB is going. 8GB for Vega 10 does kinda suck because it's the same as the RX 480 which is realistically not even half as fast as a 1080 (and Vega 10 is likely going to be faster than a 1080). But 16GB does seem an awful lot.

In retrospect, I feel both 8GB and 16GB could be possible. Maybe the flagship unlocked Vega 10 GPU will come with 16GB while a cutdown version will come with 8GB. 16GB is a huge amount for a GPU only capable of slightly beating a 1080—at least at the moment.

 

[NeverBackDown]

Vega 20 isn't coming out till next year from what has been reported? It's also 7nm. I'm fairly certain AMD said something about 7nm chips.. I can't remember though. But yes it'll be a smaller version, but if a year apart I'm sure we'll see some minor architecture improvements but not much. More like a refresh than a rebadge.

I do see your point on 16GB being a lot. It is. But also remember that they are aiming for businesses too which require loads of VRAM. Makes sense to make a single GPU that does both rather than 2 different versions. 2 birds one stone kinda thing

Yeah Gov I feel like it'll be as fast if not faster than a 1080. If it's not then it's been a waste of time for everyone.