Vega's Almost Here, But Are AMD's Board Partners Ready?

[kcarbotte]

The wait for AMD’s next high-end gaming cards is almost over. The first allotments of Vega 64 cards should arrive on August 14, but you may not be able to find cards from your favorite brand. Only a handful of AMD's board partners announced Vega cards.

Vega's Almost Here, But Are AMD's Board Partners Ready? : Read more

 

[Deleted member 217926]

I think AMD launched Vega because they had to. With the relatively low performance they had to drop prices at the last minute to have any hope of competing with the 10xx series. HBM is expensive. So is cooling a 350w+ card. Looks like the board partners aren't jumping on the lose money train.

 

[michael_732]

i totally disagree. under what metric are you quoting "relatively low performance"? there are no reputable benchmarks with release drivers to be found. only rumors, speculation, projection, and spin. using "game-mode" drivers on a pro card (where it is an afterthought) have nothing to do with real world benchmarks with real world *gaming* drivers.
i'm not saying it is the second coming either. it is a competitive card in what is now a competitive market. thank god for that. the real deal with vega is not power consumption either. although the power consumption is notable, it is not remarkable in the history of gpus.
the real deal with vega is the architecture. it is scalable ! in the 7nm node (next up), the node itself will take care of the majority of power draw. what they are doing atm is tweaking performance like normal, but also looking at places to gain efficiency on the 7nm node. and scale-ability will bring us multi-core gpus as we tackle the *big science* with sub-5nm processes and euv/x-ray lithography.

 

[redgarl]

I am holding my opinion on the card until I see benches... I think we have more to be surprised than disappointed.

 

[teamninja]

Low performance? against the only other real GPU maker? I am more surprised by the fact that they aren't as behind as I thought they would be, the jump from Maxwell to pascal was HUGE the fact that they still are managing to hang on is what surprises me THE MOST considering the don't have much money to work with. They CAN still compete on the high end impressively...

 

[hannibal]

It is quite normal that there Are first normal basic versions and later custom versions. Like Nvidia founder editions that did come out about one month before first custom versions.

 

[BBBy]

As long as etherium mining frenzy continues, they will milk up a ton of profits from these cards...
No one should ever think of any option with them feasible or not for gaming for now and short term future...
It will be not a suprise that prices will be seen like 150-200$ more than MSRPs...

 

[RedFIveStandingBy]

Vega vs 1080 the only reason to go Vega is if you need a heater-GPU combo

 

[TJ Hooker]

i totally disagree. under what metric are you quoting "relatively low performance"? there are no reputable benchmarks with release drivers to be found. only rumors, speculation, projection, and spin. using "game-mode" drivers on a pro card (where it is an afterthought) have nothing to do with real world benchmarks with real world *gaming* drivers. i'm not saying it is the second coming either. it is a competitive card in what is now a competitive market. thank god for that. the real deal with vega is not power consumption either. although the power consumption is notable, it is not remarkable in the history of gpus.

AMD's own promotional slides for RX Vega 64 have it performing about on par with a GTX 1080 in gaming, which costs the same amount, uses significantly less power, and came out over a year ago. In other words, it's only barely competitive now. With the geforce 20 series expected out next quarter, Vega's relevancy seems like it will be short lived.

the real deal with vega is the architecture. it is scalable ! in the 7nm node (next up), the node itself will take care of the majority of power draw. what they are doing atm is tweaking performance like normal, but also looking at places to gain efficiency on the 7nm node. and scale-ability will bring us multi-core gpus as we tackle the *big science* with sub-5nm processes and euv/x-ray lithography.

Pretty much any architecture should be able to be scaled down to a smaller process with some tinkering. Smaller process nodes will help Nvidia too, so I'm not sure how that's supposed to help AMD gain an advantage.

 

[rnewton8]

I want nothing more than for AMD to succeed with their GPUs, but There's a reason AMD has been so tight lipped with this card and still haven't released proper benchmarks or even showing the card in action with an FPS counter. And there's also a reason is it coming bundled with savings cards. It's because the card doesnt perform well enough, the leaked benches are legit, they show the card barely able to compete with Nvidias year and a half old cards, GTX 1070, 1080. If AMD had a competitive card, they would have been marketing it and showing benches. Instead they are doing everything they can to delay and hide the performance numbers. Add in the fact that this card is going to run hotter than hell, especially with that joke of a reference cooler, and then you have thermal throttling issues and a PC that sounds like a jet engine while playing games. I might pick up an aftermarket card in Sept with a proper cooler, but still I'm pretty dissapointed in Vega, as i think most consumers are going to be. And no way Vega will be used for mining Eth. Its too power hungry, and the value of Eth dropped but the difficulty spiked so the Eth mining craze should already be over.

 

[michael_732]

i totally disagree. under what metric are you quoting "relatively low performance"? there are no reputable benchmarks with release drivers to be found. only rumors, speculation, projection, and spin. using "game-mode" drivers on a pro card (where it is an afterthought) have nothing to do with real world benchmarks with real world *gaming* drivers. i'm not saying it is the second coming either. it is a competitive card in what is now a competitive market. thank god for that. the real deal with vega is not power consumption either. although the power consumption is notable, it is not remarkable in the history of gpus.

AMD's own promotional slides for RX Vega 64 have it performing about on par with a GTX 1080 in gaming, which costs the same amount, uses significantly less power, and came out over a year ago. In other words, it's only barely competitive now. With the geforce 20 series expected out next quarter, Vega's relevancy seems like it will be short lived.

the real deal with vega is the architecture. it is scalable ! in the 7nm node (next up), the node itself will take care of the majority of power draw. what they are doing atm is tweaking performance like normal, but also looking at places to gain efficiency on the 7nm node. and scale-ability will bring us multi-core gpus as we tackle the *big science* with sub-5nm processes and euv/x-ray lithography.

Pretty much any architecture should be able to be scaled down to a smaller process with some tinkering. Smaller process nodes will help Nvidia too, so I'm not sure how that's supposed to help AMD gain an advantage.

not true. facile maybe, but not true.
Nvidia's technology (very impressive) has focused on ever larger chip dies with the volta being the largest and most unwieldy. chip size is directly related to yields from silicon wafers. volta has not (and may not) produced the yields with which one can market to the masses. vega while fairly large, has yields upwards of 80% (according to AMD), and even more importantly it is scalable...meaning with Infinity Fabric (which nvidia does not have) AMD can have however many teraflops it wants...like 52 teraflops from four vega 64 cores stitched with Infinity Fabric with higher power efficiency and minimal lag with higher bandwidth. i predict that is precisely what Navi will bring.

 

[alextheblue]

It is quite normal that there Are first normal basic versions and later custom versions. Like Nvidia founder editions that did come out about one month before first custom versions.

Founders edition was the first thought that sprung to my mind while reading the article. I think Kevin forgot about that. Whoops.

 

[Lucky_SLS]

It seems like AMD is foucused only on the high end spectrum with scalable design and what not. but the fact remains that AMD is only playing catchup with nvidia and still not beating them with it. Vega is their first tile based rendering while nvidia did that jump from maxwell to pascal. and Nvidia's GPU r efficient as hell. the tesla is 50% more efficient with the new 12nm process. and that will be the base for Volta. it clear where this will lead on with the volta release.
atleast AMD is making its tech open source so that its atleast competitive in the mid and lower end of the spectrum with freesync and their new radeon pro render.

 

[CraigMiloRogers]

PowerColor's Web site mentions Vega now:
http://www.powercolor.com/us/products_features.asp?id=641
http://www.powercolor.com/us/products_features.asp?id=642
http://www.powercolor.com/us/products_features.asp?id=643

 

[Yarvolino]

Could you please explain me why AMD release a GPU which performs like a 1.5 years old 1080 and in addition it prices her even more than a 1080 ?

That's beyond me

 

[Yarvolino]

BTW, Vega drains power like a freight train

 

[TJ Hooker]

Spoiler

i totally disagree. under what metric are you quoting "relatively low performance"? there are no reputable benchmarks with release drivers to be found. only rumors, speculation, projection, and spin. using "game-mode" drivers on a pro card (where it is an afterthought) have nothing to do with real world benchmarks with real world *gaming* drivers. i'm not saying it is the second coming either. it is a competitive card in what is now a competitive market. thank god for that. the real deal with vega is not power consumption either. although the power consumption is notable, it is not remarkable in the history of gpus.

AMD's own promotional slides for RX Vega 64 have it performing about on par with a GTX 1080 in gaming, which costs the same amount, uses significantly less power, and came out over a year ago. In other words, it's only barely competitive now. With the geforce 20 series expected out next quarter, Vega's relevancy seems like it will be short lived.

the real deal with vega is the architecture. it is scalable ! in the 7nm node (next up), the node itself will take care of the majority of power draw. what they are doing atm is tweaking performance like normal, but also looking at places to gain efficiency on the 7nm node. and scale-ability will bring us multi-core gpus as we tackle the *big science* with sub-5nm processes and euv/x-ray lithography.

Pretty much any architecture should be able to be scaled down to a smaller process with some tinkering. Smaller process nodes will help Nvidia too, so I'm not sure how that's supposed to help AMD gain an advantage.

not true. facile maybe, but not true.
Nvidia's technology (very impressive) has focused on ever larger chip dies with the volta being the largest and most unwieldy. chip size is directly related to yields from silicon wafers. volta has not (and may not) produced the yields with which one can market to the masses. vega while fairly large, has yields upwards of 80% (according to AMD), and even more importantly it is scalable...meaning with Infinity Fabric (which nvidia does not have) AMD can have however many teraflops it wants...like 52 teraflops from four vega 64 cores stitched with Infinity Fabric with higher power efficiency and minimal lag with higher bandwidth. i predict that is precisely what Navi will bring.

Yes, I'm well aware that die size and yield are related. However, given that we're talking about gaming cards here, I don't see how GP100/GV100 are relevant. And if we look at Nvidia's last 3 generations of top tier consumer/gaming chips (GK110, GM200, GP102), die sizes are 550, 600, and 471 square mm. Doesn't seem to support your claim that Nvidia's going for larger and larger dies. And I'm not sure how important that 80% yield figure is without any context (i.e. what are Nvidia's GP102 yields?).

Regarding scalability: yeah, that'd be neat. However, AFAIK there haven't even been any announcement of cards with Infinity Fabric-linked GPUs, and even if there was we don't know what that would look like or how it would perform. And while Nvidia doesn't have Infinity Fabric, it does have NVLink (which has greater bandwidth than Infinity Fabric by the way, although I don't know about latency) which is also capable of GPU-GPU communication. Could just as easily say that Nvidia could stitch together 4 of their GPUs. Either way, it's just speculation.

 

[Radmeister]

There are in depth reviews of the Frontier edition, they all show exactly the same thing as the amd slides. The 64cu will slot squarely between a 1070 and a 1080. Why anyone would of bought the frontier edition especially the 1400$ water cooled boggles my mind, uses 15% more power and is 50% slower than a Titan Xp.....My thoughts are that AMD is delaying the release to build up hype in order to sell units on that hype instead of the hardware's merit. I think benchmarks will show that it is an inferior product. All Nvidia has to do is release an HBM2 Titan based on the quadro gp100, and rebrand current Titan as a 1090 and bye bye Vega. Too slow, too hot, too much power draw, drivers too buggy.