We have 14nm Vega until the mid-2019 arrival of Navi on 7nm - How about a "Gigahertz Edition" redo?

Jedibeeftrix

Limp Gawd
Joined
Dec 1, 2016
Messages
153
Remember the 7970?
Remember the 7970 Gigahertz Edition that cam about a year later?

https://www.hardocp.com/article/2012/06/21/amd_radeon_hd_7970_ghz_edition_video_card_review

Effectively the same chip, same pcb, same vrm setup, simply some better core/mem clocks resulting from improved yield and better binning.

Do you think AMD can afford [not] to do something like this, given that it is all they have to offer for the next 18 months?
How do you think they could improve Vega? Is is memory bound?
 
Remember the 7970?
Remember the 7970 Gigahertz Edition that cam about a year later?

https://www.hardocp.com/article/2012/06/21/amd_radeon_hd_7970_ghz_edition_video_card_review

Effectively the same chip, same pcb, same vrm setup, simply some better core/mem clocks resulting from improved yield and better binning.

Do you think AMD can afford [not] to do something like this, given that it is all they have to offer for the next 18 months?
How do you think they could improve Vega? Is is memory bound?


It's bound by its comparatively high power consumption, therefore, high heat output. I think the VRAM is plenty fast at this point, it's the core clock where its problem mostly is.
NVidia kinda predicted that they might run into a similar problem, so they disabled or completely removed some fairly advanced shit that the cards/games were not using anyway,
thus making the 1000-series GPUs lean and fast, and pretty hugely more power-efficient, when compared to the RX/Vega.
 
Is it memory bound?

Two immediate problems to solve:

1. small batch production (~5k units) is utterly unsustainable for the next 18 months
> At some point they have to put this into full production.
2. a redux product achieving the peak efficiency the architecture is capable off
> Does this include more memory bandwidth? It is the easiest fix I can envision.

This cannot go on for another 18 months:

https://www.overclockers.co.uk/pc-components/graphics-cards/amd/radeon-rx-vega-64
 
It's bound by its comparatively high power consumption, therefore, high heat output. I think the VRAM is plenty fast at this point, it's the core clock where its problem mostly is.
NVidia kinda predicted that they might run into a similar problem, so they disabled or completely removed some fairly advanced shit that the cards/games were not using anyway,
thus making the 1000-series GPUs lean and fast, and pretty hugely more power-efficient, when compared to the RX/Vega.

Not memory bound at all and last I checked, who cares what Nvidia is doing here in this AMD thread? However, production should could use a boost. Yes dude, you are AMD bashing in every thread I have seen you in so far. Otherwise, why the near constant mention of your Nvidia lovefest?
 
Even 12 nm is 14nm+, just the smallest feature is 12nm, most features are 14nm at smallest. There will be stepping revisions and process improvement. Process improvement at 10nm may improve for a 14nm shrink and port within the next year.
 
Not memory bound at all and last I checked, who cares what Nvidia is doing here in this AMD thread? However, production should could use a boost. Yes dude, you are AMD bashing in every thread I have seen you in so far. Otherwise, why the near constant mention of your Nvidia lovefest?
I'm not fucking bashing AMD. I owe AMD stock, and I own 8 AMD GPUs, from R9 270 to RX 580. I support AMD and I really really fucking want them to do well on the stock market and GPU retail market. I am just super bitter about the GPU designs they made as of late, that's why I always put comparisons to nVidia in my posts.

I fucking owned no nVidia GPUs since the ATI Radeon 9700 days. So buzz off.
 
I'm not fucking bashing AMD. I owe AMD stock, and I own 8 AMD GPUs, from R9 270 to RX 580. I support AMD and I really really fucking want them to do well on the stock market and GPU retail market. I am just super bitter about the GPU designs they made as of late, that's why I always put comparisons to nVidia in my posts.

I fucking owned no nVidia GPUs since the ATI Radeon 9700 days. So buzz off.

That is nice, if you say so. I do not care if you owned 1000's of shares of stock and every card since the beginning of time, your posts still reek of bashing. I have a friend who has used Microsoft products for decades and yet he still cannot stand them and bashes them whenever he has a chance.
 
That is nice, if you say so. I do not care if you owned 1000's of shares of stock and every card since the beginning of time, your posts still reek of bashing. I have a friend who has used Microsoft products for decades and yet he still cannot stand them and bashes them whenever he has a chance.
Okay, dude. I'm gonna end my replies to you here. This is where it ends. Have a nice day.
 
In the mean time........ :D I cannot see a higher clocked Vega being released like the 7970 was because of the way in which the clocks are dynamic. Maybe in a shrunk die node but otherwise, probably not. That said, I can easily get higher clocks on my Vega 56 with the 64 bios and a higher power limit, I just have not taken the time at making a serious attempt at down volting the card, yet. :)
 
In the mean time........ :D I cannot see a higher clocked Vega being released like the 7970 was because of the way in which the clocks are dynamic. Maybe in a shrunk die node but otherwise, probably not. That said, I can easily get higher clocks on my Vega 56 with the 64 bios and a higher power limit, I just have not taken the time at making a serious attempt at down volting the card, yet. :)


If they shrink it expect 10% increase in clocks, pretty much what GF/Papermaster from AMD stated for the shrink. Same power draw.
 
In the mean time........ :D I cannot see a higher clocked Vega being released like the 7970 was because of the way in which the clocks are dynamic. Maybe in a shrunk die node but otherwise, probably not. That said, I can easily get higher clocks on my Vega 56 with the 64 bios and a higher power limit, I just have not taken the time at making a serious attempt at down volting the card, yet. :)
Exactly, you cannot see a higher clock (where nVidia can). Enjoy your 300 Watt wonder computer or whatever the fuck.
 
Mid 2019? That is a joke right? So their CPU's are doing really good and hopefully will get better, while their GPU's are ....... somewhere...

This means Nvidia won't need to push it and we don't get a good GPU war. Maybe if they increased production and had cards near MSRP they would sell right now like hot cakes.(ok, even if price was higher right now)
 
Exactly, you cannot see a higher clock (where nVidia can). Enjoy your 300 Watt wonder computer or whatever the fuck.

No, I see a higher clock than what is stock, out of the box. The fact is, they will probably not get beyond where they are NOW without getting node shrink to at least 12+nm although, as Justreason has said, he got 1800Mhz, so it is doable. Also, I do fully enjoy my computer, especially after I resolved the black screen issue I was getting only one is a very great while. (Had to move the card to the second slot because it was conflicting with an NVMe SSD, or at least that appears to have been the issue.)

Just a shame that SLI and Crossfire are really not all that anymore and most games do not support mGPU out of the box. Otherwise, having 2 x Vega 56 would be really good.
 
Not worth it for AMD. It better to run the mining train as long as possible. There is a good chance Vega will be discontinued after Nvidia next chip comes out.

Vega only makes sense to continue to be sold if it can make a profit which is possible with the mining market.

If the mining demand ceases and Nvidia amphere/volta pushes Vega's sell price to 300-400 with the expensive HBM on board, AMD is better off discontinuing it like they did with fury X. Taking a loss on a sale is not sustainable. There is a good chance(if the mining market continues to tank) that Nvidia will release a 2070 at around 400 dollars and performs about 10% better than a gtx 1080 which is a very conservative prediction. This is because the 670/970/1070 performed at the level of the last gens large die flagship. This prediction is undershooting this performance by 20% which makes it conservative.

AMD GPU survival this year is dictated by the mining market staying alive. If it crashes and the market is flooded with cheap AMD GPU and Nvidia GPU's, AMD will not be making any money and it it looks like AMD has bet on this happening. This is why they have been conservative with GPU production. They don't want to sit on a bunch of inventory which will depreciate and be unsellable unless under fire sale prices like the 290x/290.
 
Mid 2019? That is a joke right? So their CPU's are doing really good and hopefully will get better, while their GPU's are ....... somewhere...
I wish it were.
But 12nm Vega has been cancelled.
The first 7nm vega is a HPC product and scheduled for 2018 Q4/Q1 2019.
The next consumer product is Navi (on 7nm), I I don't see anyone suggesting it will arrive mere months after HPC vega.
 
Vega only makes sense to continue to be sold if it can make a profit which is possible with the mining market.
Is that really true?

Isn't the problem with Vega due to a shortage in supply of HBM2?
Fix that, and Vega will found in greater supply, and therefore cheaper.

There is nothing magically expensive about the rest of the product.
Yes, its a big chip, but no rumours of poor yields and a £500 SKU can absorb the cost of a big chip easily.
 
Would it be creazy to think that AMD is supplying VEGA cards to big mining farms directly. I have trouble believing that the HBM2 shortages are creating this low supply. Selling bulk . All of the 3rd party manufactures are in on it. ?
Fuck the gamers. We will make our lion share one way or the other. The mining craze is a fucking god sent opportunity.
 
Would it be creazy to think that AMD is supplying VEGA cards to big mining farms directly. I have trouble believing that the HBM2 shortages are creating this low supply. Selling bulk . All of the 3rd party manufactures are in on it. ?
Fuck the gamers. We will make our lion share one way or the other. The mining craze is a fucking god sent opportunity.

Vega isn’t good for a big mining operation, draws too much power.
 
Vega isn’t good for a big mining operation, draws too much power.

depends on where you are on this planet. some places have virtually free power. and mines buy whatever the heck they can get there hands on as curently profit is WAYYYYYY over electric costs no matter what cards you run. There is a reason people still scope up the 290's and 280's dispide the horrable prefomance/watt
 
It may not be so great in ETH, Zcash, but anything based on CryptoNote/Night like monero/electronueme etc...
 
It may not be so great in ETH, Zcash, but anything based on CryptoNote/Night like monero/electronueme etc...

vega is pretty decent for zcash, and there are MUCH better alternitives for cryptonote. my cryptonote mine is made of phi 7250's (2700h/s on 200w), sky 900's (1000h/s on 340w), and some other secret sauce.
 
vega is pretty decent for zcash, and there are MUCH better alternitives for cryptonote. my cryptonote mine is made of phi 7250's (2700h/s on 200w), sky 900's (1000h/s on 340w), and some other secret sauce.

I am doing 2.1K/s on ETN using 224W of power. I would imagine I can get it down a bit under 215W if I really wanted to keep tweaking it. Granted this is with a card that never sees a GPU temp north of 35C even after mining for 10+days straight.
 
  • Like
Reactions: noko
like this
I am doing 2.1K/s on ETN using 224W of power. I would imagine I can get it down a bit under 215W if I really wanted to keep tweaking it. Granted this is with a card that never sees a GPU temp north of 35C even after mining for 10+days straight.

yea but cryptonote is normally alot less profitabel for each card (vega may be an exception) on my sky 900's they do 600h/s on zcash or 1050 on cryptonote. as for my claim of better alternitives i got the phis for much cheaper then one can pickup a verga card and 2700>2100
 
It may not be so great in ETH, Zcash, but anything based on CryptoNote/Night like monero/electronueme etc...
Vega does real good in ETH, second only to the Titan V and power levels around a 1080 Ti. Ether brand works for ETH.
 
...Which means it’s going to be massively uncompetitive with anything Nvidia rolls out next.


Yeah, its going to be worse that what we saw with the hd 3xxx series to the gtx 9x00 series, much worse.
 
My personal opinion, the GCN architecture has had its time, well over its time in fact. If it were me, I'd put Vega to bed, and focus completely on getting the next generation out asap, ahead of schedule. Make it grand, and wow gamer's once again with a competitive product. Trying to focus and put any time and money into Vega is like feeding a zombie, you're just shoving your resources into a dead thing. That is my personal opinion, no offense. Leave it to add-in-board partners to build some robust cards around Vega with some air-cooled and liquid-cooled cards, and leave it at that for 2018.
 
My personal opinion, the GCN architecture has had its time, well over its time in fact. If it were me, I'd put Vega to bed, and focus completely on getting the next generation out asap, ahead of schedule. Make it grand, and wow gamer's once again with a competitive product. Trying to focus and put any time and money into Vega is like feeding a zombie, you're just shoving your resources into a dead thing. That is my personal opinion, no offense. Leave it to add-in-board partners to build some robust cards around Vega with some air-cooled and liquid-cooled cards, and leave it at that for 2018.

Yeah but, even then, it takes time to make a new product, even if you have billions you could spend on it. No point in throwing away what you already have well at the same time, developing the next new thing. They have to make money after the way Raja managed to drive RTG into the ground. (IMO)
 
My personal opinion, the GCN architecture has had its time, well over its time in fact. If it were me, I'd put Vega to bed, and focus completely on getting the next generation out asap, ahead of schedule. Make it grand, and wow gamer's once again with a competitive product. Trying to focus and put any time and money into Vega is like feeding a zombie, you're just shoving your resources into a dead thing. That is my personal opinion, no offense. Leave it to add-in-board partners to build some robust cards around Vega with some air-cooled and liquid-cooled cards, and leave it at that for 2018.
Are you suggesting they just stop making vega until navi arrives in q2 2019?

Provided hbm2 starts being made in decent volume, why wouldn't Amd simply start producing vega in volume?

A relaunch wouldn't cost much, and volume production would solve many of the cost issues. Better than forcing aib vendors into expensive 5k batch production runs.

What does a 15month holiday on a sunk cost actually gain Amd?
 
We really don't know the BOM of current Vega production. If AMD is truly struggling with producing these cards at a profit they'd be better served by producing in volume products that have a healthier profit margin (note that AMD/RTG does not benefit at all from inflated pricing as that is profit realized by the retailers and not AMD/RTG). However, if the BOM for Vega is low enough that it's profitable to sell in volume then they should continue production and consider increasing it. Whilst Pascal is approaching EoL with Volta on the horizon RTG is only six months in their current Vega product cycle. They need to continue their quest for profits and they can only do that by selling product.
 
They need to continue their quest for profits and they can only do that by selling product.
Indeed.

I don't see that life gets any easier by only producing Polaris, as GDDR5 has gone up in price significantly too, on low/mid-range products that traditionally have slimmer margins than halo products.

Ensure volume availability of HBM2, then produce Vega in volume so cost to produce is minimised.
 
I am not sure why so many hate on Vega? Is it as fast as a 1080ti? No. Does it use more power then an Nvidia gpu? Sure with the insane stock voltage. If you tweak a VEGA GPU then its pretty efficient though. I love the fact that my gpu will make me a millionaire if I make the proper guess with the right crypto blowing up. Right now I am banking ETN @ 30+ coins a day and am HODL on everything for a year or so.

AMD did a lot with VEGA, and having to use it as an "All in One" for both major compute and graphics has definitely hurt. AMD could have stripped VEGA down and probably milked another 2-350Mhz of core clock out of it, but they would have been completely giving up the high margin big data market they are trying to break into. I think VEGA is actually a step in the right direction, it just tried to do too much too fast.

AMD *CAN* complete, and I hope the refresh of VEGA will get them through the year and they continue to invest R&D into NAvi's replacement. If they can get HBM supply flowing, they will sell every single VEGA they can make.
 
i do kind of suspect that a vega refresh will be a separate SKU which a separate bios identity, and that the driver will deliberately exclude the crypto maths improvments.
they will be reserved for existing vega64/56 SKU's, as well as Vega Founders Edition.
i.e. it allows AMD to benefit from the mining premium, rather than giving all that premium to the retailers.
 
i do kind of suspect that a vega refresh will be a separate SKU which a separate bios identity, and that the driver will deliberately exclude the crypto maths improvments.
they will be reserved for existing vega64/56 SKU's, as well as Vega Founders Edition.
i.e. it allows AMD to benefit from the mining premium, rather than giving all that premium to the retailers.

That would be stupid for AMD to do, and completely nuke their whole "open" vision for OpenCL computing.
 
There are not enough gpu's on the market at the moment, so really it could not have been a better time for AMD to run into major performance issues vs their competitor than now.

They can still likely sell all that they can produce, and as long as margins are decent, they will keep on making money in that division as well.

Performance is actually sufficient, Vega 56 is a decent part, they should be able to get away with it during next 18 months, unless Volta is amazing to that extent that low/mid range Volta parts outperform it at significantly lower price, which is something I would not expect.
 
That would be stupid for AMD to do, and completely nuke their whole "open" vision for OpenCL computing.
Why?
They are already explicitly bifurcating their gpu lines into graphics and compute with 7nm vega. And I am not suggesting they rip out opencl/hsa. Merely the block chain compute enhancements they added into the driver last year.

If they want to remain a consumer graphics supplier (rather than a block chain asic supplier), then they need to supply graphics to consumers. At a market competitive price.

How can they do this?
And still get high margin return from the crypto Bois?
 
Back
Top