AMD Vega Discussion Thread

[cucker tarlson]

Funny thing, Vega's performance plummets a bit more behind 1080 in games like mankind Divided and RoTR after enabling MSAA/SSAA, especially in the first one. If anything I'd suspect Vega 64 close the gap or beat the nvidia's GDDR5X 256-bit card the more memory taxing settings are used. It does so at 4K but using MSAA/SSAA seems to tell a different story. I know MD can eat up to 7.5GB vram without MSAA @1440p and it was this game that AMD showcased their HBCC so that's absolutely bogus.

 

[notb]

Well I pulled the trigger this afternoon on the sapphire Rx vega 64. $780+tax pre ordered and hope the price drops once in stock. Says final price can change when recieved from vendor/supplier.. fingers crossed it don't shoot to the moon.

Is this an option? Can you cancel it after they announce the final price? Do you lose anything?

I have to say... I don't get the idea of preordering, but it seems even more weird now. Couldn't you wait a month, really? We don't know how buggy this GPU is. Latest Ryzen problems are a bad sign, but in that case you're at least getting performance to wipe away the tears.

 

[RejZoR]

I really wanted to grab RX Vega until I was at a point of decision. Reference models are crap so they didn't even come into my selection. Vega56 is still way too far away, same for aftermarket cooled Vega64. Only thing now that was viable was Vega64 Liquid. Only problem there, it costs as much as GTX 1080Ti with really beefy aftermarket air cooler. Now, knowing how GTX 1080Ti performs this exact moment is no secret or speculation.

I was also considering GTX 1080 vanilla, but came to realization that paying like 150€ more now gives me option to skip one 400€+ upgrade in the future. So, I went with 1080Ti instead. The fact 1080Ti is actually doubling the framerate of GTX 980 is also a nice thing to know and have. Buying new graphic card for 20-30% boost is kinda crap. But 100% boost...

Realistically, I can see Vega being interesting for people still on HD6900 or HD7000 series. Or maybe NVIDIA people with Fermi cards and older... But for anyone with R9 Fury or even R9 290X/390X/RX480/RX580, ah, it's a tough order. And GTX 1080 is in a similar position really. It's faster, but just not that much faster. Either you wait out for next update or you go big. Which is kinda why I went big...

I really hope Navi will work out for AMD and won't be late again. RX Vega would be amazing option if it wasn't late...

 

[cucker tarlson]

Buying a 1080Ti now is a no brainer really. Nothing comes close.

They did a good job making Vega louder than the legendary reference 290X's (leaf) blower style coolers

 

[RejZoR]

Buying a 1080Ti now is a no brainer really. Nothing comes close.

They did a good job making Vega louder than the legendary reference 290X's (leaf) blower style coolers

Although to be honest, everyone raving about massive RX Vega's consumption are always presenting the numbers you get when using the most unrestricted Turbo mode where GPU just drinks as much power as you give it. Where wth Balanced or even Power Saver, drop in performance (from Turbo) is minimal, but they save tons of power and there, while a bit more power hungry than GTX 1080, it also scores tiny bit higher as far as I remember TPU numbers (at least for 1080p). I mean, you can get GTX 1080Ti to sip 400W as well. But no one is parading that around as something typical. Where for Vega, everyone is parading around with Turbo numbers. C'mon folks, don't be silly.

People also have to realize that implementation of HBM2 causes additional problem. Where GDDR5(X) chips are offset from GPU cooling plate or often don't even have heatsinks, HBM is always next to GPU core, contributing heat. Knowing they had to overvolt it because of not ideal performance yields by Hynix, this is the major factor. GDDR5X on GTX 1080 doesn't directly contribute to the GPU heat output since it's usually indirectly cooled where HBM does. So, AMD is dealing with a lot of very concentrated heat and that extra head is causing additional leaks in core, contributing to even more heat and even higher draw. It's in a way interesting that AMD managed to keep it this much under control.

 

[cucker tarlson]

Perf/wat is better on pwr saving bios setting but still way behind gtx 1080. Vega's default balanced mode has worse perf/watt than nvidia's 600mm2 28nm GPU and that's just sad considering it's 2.5 years since GM200 came out in march 2015.

while in the power saving bios vega64 essentially loses ~10% performance

To my mind the card is a (tera)flop when compared to 1080Ti and that should be the point of reference. On the other hand V56 is a very nice card, I'd take that over 1070 easily.

 

[RejZoR]

Vega 56 indeed has the best balance and just like difference between Fury and Fury X, it's not really that big compared to Vega 64. But it's a lot cheaper and eats less power.

 

[HD64G]

More people tend to forget that Pascal GPUs are hard-locked on BIOS to not allow more consumption. That's why their oc is limited. Vega RX on the other side needed to allow higher clocks to win over 1070 and 1080 on reference stock settings with early drivers, so they made those BIOS turbo settings to do this. That caused higher power consumption but you can make it much more efficient with just a click in the driver setting when the game isn't too heavy on GPU usage. $10-20 per year isn't much for a casual gamer. Vega 56 is the best vfm gamer GPU and will become much better when the custom ones get on sale in less than a month from now. Vega 64's success depends on driver evolution to take advantage of new features on most recent and recent games. Some quirks like on Civ6 will be ironed out in 1-2 drivers. It will be faster than custom 1080s that consume 220-240W now (https://www.techpowerup.com/reviews/MSI/GTX_1080_Gaming_X_Plus_11_Gbps/28.html) in 3 months on balanced power profile with not much more power consumption than them. So, it is not as bad as it seems. Fury X was bashed badly when released and on 4K it equals 1070 now. For gamers that keep their GPUs at least for 2-3 years it is a sound choice if they don't need or can afford a 1080Ti which atm is a faultless choice if you have the money.

 

[Vlada011]

I'm sorry because AMD fail to compete with NVIDIA in GPU section.
If someone try to ignore that fact should know that AMD didn't success to launch same product as NVIDIA even 12 months later with 100W more power consumption.
How many time need to AMD launch product as NVIDIA with same performance within same power consumption, probably more than 2 years.
That's a lot of time in IT industry.

What need to happen to someone pay 599$ for RX Vega when GTX1080 cost 550$.
She is faster, she have PhysX, she need 120W less. If someone build SLI that's almost 250W less.
That's significant less power.
I see RX VEGA 64 Liquid Edition is 699, GIGABYTE GTX1080Ti cost 720$.
Difference is much bigger. It would be nice to someone compare RX Vega 64 with TITAN X Maxwell in different resolution and different games to see difference,
Q1 2015. AMD Q3 2017 vs NVIDIA Q1 2015

 

[the54thvoid]

More people tend to forget that Pascal GPUs are hard-locked on BIOS to not allow more consumption. That's why their oc is limited. Vega RX on the other side needed to allow higher clocks to win over 1070 and 1080 on reference stock settings with early drivers, so they made those BIOS turbo settings to do this. That caused higher power consumption but you can make it much more efficient with just a click in the driver setting when the game isn't too heavy on GPU usage. $10-20 per year isn't much for a casual gamer. Vega 56 is the best vfm gamer GPU and will become much better when the custom ones get on sale in less than a month from now. Vega 64's success depends on driver evolution to take advantage of new features on most recent and recent games. Some quirks like on Civ6 will be ironed out in 1-2 drivers. It will be faster than custom 1080s that consume 220-240W now (https://www.techpowerup.com/reviews/MSI/GTX_1080_Gaming_X_Plus_11_Gbps/28.html) in 3 months on balanced power profile with not much more power consumption than them. So, it is not as bad as it seems. Fury X was bashed badly when released and on 4K it equals 1070 now. For gamers that keep their GPUs at least for 2-3 years it is a sound choice if they don't need or can afford a 1080Ti which atm is a faultless choice if you have the money.

Power limits perhaps but still a solid 15% overclock. My 1080ti runs at 2038Ghz, it's not there to reduce power consumption, it's there because the process can't deal with excessive voltage without extreme cooling to enable stability.

 

[RejZoR]

GTX 1080Ti is not "hardlocked". It's power limited because of connectors available. The original design has 1x8pin. That's 150W of power and 75W of power from PCIe slot. So, 225W max. You can up the limit, but that's about it. It's why I picked AORUS (though others have it too) with 2x8pin. Which bumps the power to total 375W.

 

[yogurt_21]

I really hope Navi will work out for AMD and won't be late again. RX Vega would be amazing option if it wasn't late...

yeah you look at Vega and can't help but think this would have been awesome last year... but to line up the 64 perfectly with the 1080 and the 56 perfectly with the 1070 which are now both a year old...utter disappointment.

they had a full year of reviews of the competition to beat, Volta is getting closer by the minute, and they still targeted the exact performance of 1 year old cards? Not even 10-15% higher?

You can't help but think about the old ATI vs Nvidia rivalry and how fast releases used to be. Now we wait 2 years since Fury X(a disappointment itself) and all we get is the same performance as the older competition models yet higher power draw and heat?

Its like they can only compete on 1 front at a time. So the instant they decided they were going to re-enter the cpu wars, the gpu wars were lost.

Now I find myself wanting to wait for volta and see what the 400$ option is there. If I go 1070 or Vega 56 now, it will only irk me all the more if the 400$ option is significantly faster... guess I'm riding this 290 till it quits, its not doing well atm.

 

[EarthDog]

GTX 1080Ti is not "hardlocked". It's power limited because of connectors available. The original design has 1x8pin. That's 150W of power and 75W of power from PCIe slot. So, 225W max. You can up the limit, but that's about it. It's why I picked AORUS (though others have it too) with 2x8pin. Which bumps the power to total 375W.

Sort of.....

There are cards with 2 8-pins on it which have their power limit WELL below 375W... in fact, I would venture to say MOST that have 2 8-pins are well below 375W. Typical power limits are 5-15% increases over its 250W design spec. There is a correlation, but, the reality is most are limited well below what the spec is for 8-pin and PCIe slot.

 

[the54thvoid]

Newsflash!

AMD pricing lies! The initial launch price was for early adopters. Once that preorder drive was fulfilled, pricing will increase, in line with AMD suggested retail pricing. 450 was launch day only pretty much, 550 is the 'new' price, according to Gibbo from OcUK.

Oooh. Bad AMD.

source
horses mouth

Now the good and bad news, the good news is AMD are rebating early launch sales to allow us to hit £449.99 on the stand alone black card which has no games. This is a launch only price which AMD at present are saying will be withdrawn in the near future, when if it happens is unknown, but remember do not be shocked if the price jumps nearly £100 in a few days.

 

[fullinfusion]

Is this an option? Can you cancel it after they announce the final price? Do you lose anything?

I have to say... I don't get the idea of preordering, but it seems even more weird now. Couldn't you wait a month, really? We don't know how buggy this GPU is. Latest Ryzen problems are a bad sign, but in that case you're at least getting performance to wipe away the tears.

Yes it's an option. I just gave a $500 deposit and being local they have a few suppliers they deal with. Xbox and places like that I'll have to wait in the Q but the local shop actually takes good care of me

 

[notb]

People also have to realize that implementation of HBM2 causes additional problem. Where GDDR5(X) chips are offset from GPU cooling plate or often don't even have heatsinks, HBM is always next to GPU core, contributing heat. Knowing they had to overvolt it because of not ideal performance yields by Hynix, this is the major factor. GDDR5X on GTX 1080 doesn't directly contribute to the GPU heat output since it's usually indirectly cooled where HBM does. So, AMD is dealing with a lot of very concentrated heat and that extra head is causing additional leaks in core, contributing to even more heat and even higher draw. It's in a way interesting that AMD managed to keep it this much under control.

No, it's not interesting. It was a stupid choice, they're paying the price. And the users will as well.
Everyone should bash them for using HBM. Instead, some are praising them for making sacrifices and compromises that make it work.

A writer famous in Poland once said that "socialism heroically overcomes difficulties unknown in any other system".
I always think about this when I see people praising AMD for struggling with these self-inflicted wounds...

AMD pricing lies! The initial launch price was for early adopters. Once that preorder drive was fulfilled, pricing will increase, in line with AMD suggested retail pricing. 450 was launch day only pretty much, 550 is the 'new' price, according to Gibbo from OcUK.

ROTFL if true.
You know why they did it, right? Reviews are already out. Each of them says it's not a performance leader and rubbish in efficiency, but good value. This is what people will be reading for the following 2 years when they consider Vega.
I don't think many review sites are solid enough to modify a review and add "rubbish".

Vega RX on the other side needed to allow higher clocks to win over 1070 and 1080 on reference stock settings with early drivers, so they made those BIOS turbo settings to do this. That caused higher power consumption but you can make it much more efficient with just a click in the driver setting when the game isn't too heavy on GPU usage. $10-20 per year isn't much for a casual gamer. Vega 56 is the best vfm gamer GPU and will become much better when the custom ones get on sale in less than a month from now. Vega 64's success depends on driver evolution to take advantage of new features on most recent and recent games. Some quirks like on Civ6 will be ironed out in 1-2 drivers. It will be faster than custom 1080s that consume 220-240W now (https://www.techpowerup.com/reviews/MSI/GTX_1080_Gaming_X_Plus_11_Gbps/28.html) in 3 months on balanced power profile with not much more power consumption than them. So, it is not as bad as it seems. Fury X was bashed badly when released and on 4K it equals 1070 now. For gamers that keep their GPUs at least for 2-3 years it is a sound choice if they don't need or can afford a 1080Ti which atm is a faultless choice if you have the money.

I have to say: I like your post because it looks like a summary of the whole Vega wishful thinking I've seen to date.
Early drivers, custom ones will be better, quirks will be ironed, games will start supporting new features and so on.

Bad news for you: there might not be (m)any "custom ones". AIB partners don't have cooler designs that could support this card and since the sales volume will be low, they might keep the reference design.

As for the Fury X - it was bashed when released for poor efficiency. It did, however, match the 4K performance of flagship 980Ti (released a month earlier!) - something that Vega RX failed to deliver. Also: so what if it matches a 1070 now? It's a way cheaper card. And it's still pretty equal to 980Ti performance-wise.

 

[EarthDog]

Bad news for you: there might not be (m)any "custom ones". AIB partners don't have cooler designs that could support this card and since the sales volume will be low, they might keep the reference design.

I agreed with your entire post outside of this passage.

There are plenty of heatsinks able to cool 250W+. Remember the cooler on the old 780Ti/980Ti Lightning? Yeah that triple slot thing was said to cool 500W. 300W+ is nothing compared to that.

 

[RejZoR]

You people can bash HBM all you want, but the reality is, HBM is the future. In 10 years time, there will be nothing but HBM on all cards (except maybe bottom feeder ones that might still have some revision of GDDR). Modular stacking of stuff around GPU core is how graphic cards will exist in the future. Multiple smaller cores, memory modules, video decoder engines, all of this will come modular next to actual GPU core and not as massive monolithic cores.

 

[EarthDog]

You are probably right. But, call me in 10 years....... As of now, its borderline useless.

I understand things take time to develop, etc.. but, why not start bringing it out now where at least a hint of it being useful now is on the horizon? HBM was out when its bandwidth wasn't remotely needed. Arguments can be made... GOOD arguments can be made, it still isn't needed at 4K. Again, look at GDDR5X. I see plenty of "high performing" cards using GDDRX5 smoking AMD Vega and Fury like a freaked Dutch gets smoked in the hood.

Though I have deja vu... I've said that to you plenty of times. You still can't come up with a reason why it's useful now, except to be useful many years later... which is spot on.

EDIT: I still can't believe you have been here for almost 13 years and still don't seem to quote the person you are talking to... your post just seems random as F as you aren't talking to the person above you, and god knows when HBM was mentioned here.... jebus man..

EDIT2: In fact, you were the one that mentioned it first, post # 730. I don't see it from 601 (from 8/3) on forward...'dat straw man is in full effect!!!!!!!

 

[the54thvoid]

HBM is only the future if another better tech doesn't get invented to replace it. Not unheard of.

 

[RejZoR]

You are probably right. But, call me in 10 years....... As of now, its borderline useless.

I understand things take time to develop, etc.. but, why not start bringing it out now where at least a hint of it being useful now is on the horizon? HBM was out when its bandwidth wasn't remotely needed. Arguments can be made... GOOD arguments can be made, it still isn't needed at 4K. Again, look at GDDR5X. I see plenty of "high performing" cards using GDDRX5 smoking AMD Vega and Fury like a freaked Dutch gets smoked in the hood.

Though I have deja vu... I've said that to you plenty of times. You still can't come up with a reason why it's useful now, except to be useful many years later... which is spot on.

EDIT: I still can't believe you have been here for almost 13 years and still don't seem to quote the person you are talking to... your post just seems random as F as you aren't talking to the person above you, and god knows when HBM was mentioned here.... jebus man..

EDIT2: In fact, you were the one that mentioned it first, post # 730. I don't see it from 601 (from 8/3) on forward...'dat straw man is in full effect!!!!!!!

I'm suppose to quote notb's clusterfuck of quotes? Were you talking about HBM? No. Do you really need a special label for every bloody thing or you can read the context of things?

 

[EarthDog]

Yes, you are supposed to forum!

 

[the54thvoid]

Tempers now. Mommy thinks you ought to sleep it off.

2nd sentence is actually a quote from a film in a song by a band. Bonus points and awesome kudos if someone names it without using google.

 

[EarthDog]

I like my thrills with a side of kult.

 

[the54thvoid]

I like my thrills with a side of kult.

Bloody hope you didnt Google that..... If not -we'd get on well.