Retailers are Buying AMD RX Vega 64 at $675 Each

[Rahmat Sofyan]

As much as I'd love to be proven wrong, I don't think there will be a consumer-segment Volta card with HBM2, only expensive Tesla or Quadro. NVIDIA will pull through using GDDR5X or GDDR6 for GeForce.

Same thought..

Thats why I wonder.. Some rumors said volta will with HBM, some said with GDDR. If Vega used GDDR, maybe the price not this high, despite all miners craziness..

actually, what was AMD purpose with HBM ?

 

[EarthDog]

actually, what was AMD purpose with HBM ?

high bandwidth (check, but relatively useless below 4k), and lost cost to produce (nerp).

 

[Evildead666]

Same thought..

Thats why I wonder.. Some rumors said volta will with HBM, some said with GDDR. If Vega used GDDR, maybe the price not this high, despite all miners craziness..

actually, what was AMD purpose with HBM ?

It was a real good idea at the time, but then the HBM delays and cost increases (they expected volume production by now).
You can't design a chip with a 2048bit HBM interface and just switch it to GDDR.
The lower Vega will prob all be GDDR based.

Consumer Volta is most probably going to be GDDR5X, leaving the HBM versions for the Pro Crowd.

 

[Totally]

Moar likes from AMD fanboys. Keep them coming. Meanwhile in the real world:

AMD Radeon RX 470 0.34%
AMD Radeon RX 480 0.92%

That's less than NVIDIA GeForce 840M which is a mobile only GPU which has sold less than 100K units.

RX 4XX series has been on the market for more than a year already.
RX 5XX series has been on the market for four months.

RX 570 and 580 are nowhere to be seen. Keep deluding yourself that people use modern AMD GPUs for gaming. I'd love to see this dying breed.



Logistics and post-sale support (warranty, returns, etc). AIBs don't have money, people and resources for that.

Where'd you pull that data from? And of course you are aware how unbelievably Ill informed it is to compare laptop gpus to desktop discrete gpus. Where are the figures for the GTX 1050/1060 for a more rational arguement? Why aren't you squaking about how intel igps outsell nearly everything and therefore must be better than everything else.

 

[Hood]

http://wccftech.com/radeon-rx-vega-64-msrp-pricing-update/

This statement from "a source close to AMD who is very knowledgeable about the matter" is the biggest pack of lies I've ever heard. "tens of thousands of units were shipped, but ran into logistical delays" - so I guess every truck they shipped them in broke down, or all the bridges were washed out - what a coincidence! "This problem will be quickly remedied and the entire stock should be up for grabs within this week. In fact, the card has already been restocked at Newegg and Amazon, albeit at the $599 price point (Amazon’s dynamic pricing algo has already taken it above the $700 mark)" - another lie, they've never been $499 or $599, always $689 and higher, as they are right now (I've checked every day since pre-orders started). AMD is as bad at lying as they are at marketing. If you take these statements at face value, they make AMD look stupid, disbelieve them and they look like liars - ether way, they lose, especially since Vega 64 is at most a $400 card.

 

[Captain_Tom]

"There is another option AMD can try, in our opinion. It can re-launch RX Vega 64 and RX Vega 56 as new SKUs which come with crippled cypto-currency mining abilities (a special BIOS or something driver-level, or even something at the silicon-level), and discontinue the RX Vega 56 and RX Vega 64. "


OMG how dumb are some of these tech editors. Jesus it's like some of you fundamentally do not understand mining or even basic business!

1. Vega's current gaming performance is definitely about 30% lower than it should be. Maybe it will get better, heck it almost definitely will; but right now it isn't good enough. AMD needs to move inventory to make money while they ramp up HBM2 production. The crypto-boom is a god send for AMD while they fix up their drivers!
2. You fundamentally cannot nuke mining performance at the hardware level without destroying gaming/rendering performance. Crypto currencies are designed from the ground up to utilize GPU architectures in order to decentralize computing to the masses. Even if they found a way to "Slightly" nerf Ethereum mining performance, it wouldn't matter when you can mine 100 other f***ing coins. LOL this is a $150,000,000,000 market AMD is tapping into. Why would they not?!
3. Crypto-currency mining isn't going anywhere! EVER! It's been 10 years lol (The internet is here to stay too btw lmao). Every time this has happened in the past AMD has made ZERO extra money even though their cards were in insane demand. It's about time they made some money. Do you guys want any R&D budget for Navi? Heck they could use money for their driver team alone!
4. I agree AMD should do a relaunch - but it should be of AIB-cooled cards when they had Primitive Discard enabled in like 2-4 months. Right now would be insane. They can't even make enough cards to meet supply, so yeah prices are high for now.

 

[EarthDog]

Its been well over 2 days captain...... still waiting...

 

[Captain_Tom]

Its been well over 2 days captain...... still waiting...

For what?

 

[EarthDog]

Still wating to hear your explanation of...

https://www.techpowerup.com/forums/...-smoke-and-mirrors.236177/page-5#post-3711427

Ive posted some results with the latest beta driver and it puts it around 36 mh/s with overclocked memory. Ive asked for actual ROI numbers and, even though youve been doing this since college, the math doesnt seem to support your claims.

Perhaps we/I am missing something...eagerly waiting your response in a couple days.

You responded with 37mhs result, but never broke down the math for this being the best at whatever the hell you were talking about.

 

[Captain_Tom]

Still wating to hear your explanation of...

https://www.techpowerup.com/forums/...-smoke-and-mirrors.236177/page-5#post-3711427

That's right now too lol. I have been tweaking it with some other early adopters and we should be to 45 MH/s within a month.



Wait how would Vega not at least get like 35 MH/s or higher? I don't believe you have ever mined a day in your life because you clearly do not even know how this works...

 

[btarunr]

actually, what was AMD purpose with HBM ?

With Vega AMD went HBM2 because it [claims to have] solved many of its memory management problems using the HBCC approach, and needed a very fast, very low-latency memory sitting next to the GPU. Memory management was AMD's problem-area with GCN. So they kept working around it by throwing brute memory bandwidth to solve the problem (imagine multiple failed fetch/store ops to correctly fetch/store once). Then they realized they can't go wider than 512-bit GDDR without encountering huge power costs, not to mention running out of PCB. So they did HBM with Fury, hoping that the exotic nature of HBM, coupled with 512 GB/s bandwidth would help sell it (which it did, until NVIDIA launched Pascal).

 

[Captain_Tom]

With Vega AMD went HBM2 because it [claims to have] solved many of its memory management problems using the HBCC approach, and needed a very fast, very low-latency memory sitting next to the GPU. Memory management was AMD's problem-area with GCN. So they kept working around it by throwing brute memory bandwidth to solve the problem (imagine multiple failed fetch/store ops to correctly fetch/store once). Then they realized they can't go wider than 512-bit GDDR without encountering huge power costs, not to mention running out of PCB. So they did HBM with Fury, hoping that the exotic nature of HBM, coupled with 512 GB/s bandwidth would help sell it (which it did, until NVIDIA launched Pascal).

Pretty much exactly this. At least Fury gave them some time to optimize an architecture around HBM. Too bad their driver team can't keep up with their engineers :/

 

[phanbuey]

Pretty much exactly this. At least Fury gave them some time to optimize an architecture around HBM. Too bad their driver team can't keep up with their engineers :/

Seems to be a recurring problem...

 

[Captain_Tom]

Seems to be a recurring problem...

View attachment 91450

Haha I know it is buddy. Although I will say it seems like AMD saw that coming. Even after the update Vega FE is competitive with the Titan that costs $200 more. So yeah the Titan now wins at some professional applications, but Vega wins at others (And these are the terrible launch drivers. They will improve substantially).

 

[phanbuey]

Haha I know it is buddy. Although I will say it seems like AMD saw that coming. Even after the update Vega FE is competitive with the Titan that costs $200 more. So yeah the Titan now wins at some professional applications, but Vega wins at others (And these are the terrible launch drivers. They will improve substantially).

Re: the drivers for sure... i mean NV released a card with incomplete driver set altogether. They had a
"hey Bob that vega is looking pretty good in Maya, did... uh... did you ever add that one maya thing to the package before you sent it?"
- bob: "Ohhhhhh.... shhh... ahh. We gotta call Matt, he will figure out how to make this sound like a good thing." moment.

what in the current price of the currency that you mine is 45 mh/s or 35 mh/s mean in terms dollars per day?

 

[nemesis.ie]

"There is another option AMD can try, in our opinion. It can re-launch RX Vega 64 and RX Vega 56 as new SKUs which come with crippled cypto-currency mining abilities (a special BIOS or something driver-level, or even something at the silicon-level), and discontinue the RX Vega 56 and RX Vega 64. "

I'd like to see them launch the next batch with a UEFI that sets a lower voltage on chips that can do it, as this seems to reduce the power draw a lot.

Likewise the AIB partners to put one up for d/l for existing owners given it seems to be locked at the moment so folks can't currently do it themselves.

On the other topic, regarding the 3x improvement on the TitanXp, that was nothing to do with development, they already had that performance available and used on the higher grade cards and just enabled it because of competition from Vega.

 

[efikkan]

Vega's current gaming performance is definitely about 30% lower than it should be. Maybe it will get better, heck it almost definitely will; but right now it isn't good enough. AMD needs to move inventory to make money while they ramp up HBM2 production. The crypto-boom is a god send for AMD while they fix up their drivers!

I see you are still in denial about Vega. This is just the same old excuse we hear every time with AMD hardware; wait a while and it will improve, but it never does. Keeping in mind that RX Vega was three months overdue, the drivers were more mature than previous generations at launch, so all the low hanging fruit is pretty much picked already…

Cryptocurrencies are pretty much irrelevant for Vega10 due to energy efficiency. Just like with datacenters, "serious" miners primarily care about energy efficiency because cooling always becomes the problem.

 

[R0H1T]

I see you are still in denial about Vega. This is just the same old excuse we hear every time with AMD hardware; wait a while and it will improve, but it never does. Keeping in mind that RX Vega was three months overdue, the drivers were more mature than previous generations at launch, so all the low hanging fruit is pretty much picked already…

Cryptocurrencies is pretty much irrelevant for Vega10 due to energy efficiency. Just like with datacenters, "serious" miners primarily care about energy efficiency because cooling always becomes the problem.

Is it?

Quick note on primitive shaders from my end: I had a chat with AMD PR a bit ago to clear up the earlier confusion. Primitive shaders are definitely, absolutely, 100% not enabled in any current public drivers.

The manual developer API is not ready, and the automatic feature to have the driver invoke them on its own is not enabled.

https://forum.beyond3d.com/posts/1997699/

http://radeon.com/_downloads/vega-whitepaper-11.6.17.pdf

Also pseudo channel is not working as intended, maybe needs a driver or firmware update ~

 

[efikkan]

Is it?

Yes, the low hanging fruit are all picked. The Vega you see now, is the Vega you get.
If a hardware feature is disabled, then it's because it doesn't work as intended. This is quite normal. But some of you will always interpret this as future gains, and always keep waiting for the next mythical driver to finally solve everything. If you keep waiting, we'll be buying the successor of Volta long before you get to "unleash" your Vega…

 

[R0H1T]

Yes, the low hanging fruit are all picked. The Vega you see now, is the Vega you get.
If a hardware feature is disabled, then it's because it doesn't work as intended. This is quite normal. But some of you will always interpret this as future gains, and always keep waiting for the next mythical driver to finally solve everything. If you keep waiting, we'll be buying the successor of Volta long before you get to "unleash" your Vega…

And you're saying you know these features are disabled in hardware, how?

 

[Vayra86]

And you're saying you know these features are disabled in hardware, how?

Because its not active and the card is on the market right now. People are buying a card with a price and a performance metric that they know is there. The pricing and positioning is now set, which means its general performance is also set. There are exceptions to this rule, but they are so rare, I think I know of only one, and that was when Nvidia activated Shader Cache on Kepler. But even that was what, +5-7% in select titles.

Go look at the fine wine that is any older GCN card today, such as the Fury X. It has gained some performance, but again, only in select titles. The general Fury X perf level is still where it was at launch, +-5% depending on game.

Hell, not even AMD itself is saying Vega will go much faster, yet somehow some people still believe it will.

 

[efikkan]

And you're saying you know these features are disabled in hardware, how?

I didn't say anything was disabled in hardware.
You were the one referring to a hardware feature which is (supposedly) not enabled.

 

[dyonoctis]

"There is another option AMD can try, in our opinion. It can re-launch RX Vega 64 and RX Vega 56 as new SKUs which come with crippled cypto-currency mining abilities (a special BIOS or something driver-level, or even something at the silicon-level), and discontinue the RX Vega 56 and RX Vega 64. "


OMG how dumb are some of these tech editors. Jesus it's like some of you fundamentally do not understand mining or even basic business!

1. Vega's current gaming performance is definitely about 30% lower than it should be. Maybe it will get better, heck it almost definitely will; but right now it isn't good enough. AMD needs to move inventory to make money while they ramp up HBM2 production. The crypto-boom is a god send for AMD while they fix up their drivers!
2. You fundamentally cannot nuke mining performance at the hardware level without destroying gaming/rendering performance. Crypto currencies are designed from the ground up to utilize GPU architectures in order to decentralize computing to the masses. Even if they found a way to "Slightly" nerf Ethereum mining performance, it wouldn't matter when you can mine 100 other f***ing coins. LOL this is a $150,000,000,000 market AMD is tapping into. Why would they not?!
3. Crypto-currency mining isn't going anywhere! EVER! It's been 10 years lol (The internet is here to stay too btw lmao). Every time this has happened in the past AMD has made ZERO extra money even though their cards were in insane demand. It's about time they made some money. Do you guys want any R&D budget for Navi? Heck they could use money for their driver team alone!
4. I agree AMD should do a relaunch - but it should be of AIB-cooled cards when they had Primitive Discard enabled in like 2-4 months. Right now would be insane. They can't even make enough cards to meet supply, so yeah prices are high for now.

As long as they don't make real dedicated mining compute unit, Navi is going to be another chaotic launch with overpriced gpu. Nvidia already stated that they don't feel threatened by AMD at all. In Europe AMD got NOTHING in the 1080p budget segment, RX 570/580 are just overpriced ghost, GTX 1060 are the only thing that you can actually get in this segment.

Since i'm also doing 3d rendering I used to bet on AMD since their consummer product got more compute power, but even now the developpers find that cuda is just more mature and easy to work with. Octane renderer ? cuda only. Redshift ? same story. Cycle renderer ? the open cl version got less features. Pixar ? They developped a bunch of tools that only works with nvidia hardware. Even vray is faster with cuda than open cl. Apple only using AMD gpu ? it doesn't help that open cl on mac is not only slower than on windows and linux, it also got stability issue. Having a Nvidia gpu for creative content creation give you more options, while AMD is forcing you to look for specific option that are not always the best. Add that to the current state of the market, and that's another thing that makes amd less attractive for conssumers, and will not push developpers to work with open cl.

Even with not so great mining perfomance, Nvidia is still making a huge chunk of money, and manage to keep their brand fidelity. Meanwhile the gaming market not so happy with AMD right now, and if the price keep being inflated because of the crypto currency, AMD might win a market but lose another one.

I'm not saying that crypto should die, but AMD need to rethink their strategy.

 

[_Flare]

1. A GPU-manufacturer needs to have marketshare to give others a reason to optimize for its products.
2. Why is nvidia able to push that much FPS with the 150W GTX1070 having only 3 GPC, where we wait and wait to see AMD doing equal with GCN with 4 Engines with equal Watts?
3. I can´t build a "good in some years"-Card while needing a good and efficient card to compeet, NOW.
4. Same fault like the Bulldozer-idea... meh in some couple of month it could be widely adopted and optimized bla bla. How with inactive/broken engines and nearly no marketshare to get wide adoption and optimization?
5. being not to rigorous: if the 200W state of Vega does anytime in the future (month or years) beat the OC-Variants of the GTX1080 with same Watts, okay, better late then never. But if Vega in that state EVER gets near the FPS of 1080Ti i will be an idiot.

 

[TheMailMan78]

I think AMD was holding on to all the Vegas so they could mine themselves. When the market went down the market all of a sudden had a supply of Vega. AMD already knows they are made for mining so they jacked up the prices.

Either way Vega is a massive piece of S#!T for the price.