RX Series Cards Are All Sold Out, Why Isn't AMD Making More?

A lot of theories. Let me add my own.

Polaris cards (and the upcoming Vega cards as well) are all based off of the Arctic Islands technology. Polaris was meant to go toe to toe with NVidia but underperformed (see Kyle's article from a while back detailing this, as well as RTG's desire to split and join Intel).

At this point, I think AMD knows what they have, and internally, have no faith in these cards. Therefore, they just want to break even and stay afloat until Navi, which is 100% Raja's card (people can wring their hands all they want about Raja owning Vega, but it doesn't change the fact that at its core, it is Greenland, which was developed pre-Raja).

My thought is AMD just wants to tread water on the graphics end to get out from under a badly designed architecture from the previous graphics administration. Hence if they break even, they don't care if its because of gamers or miners, because they don't foresee an improvement to their reputation with the gaming community either way.

Everything is pointing to Navi. Not to say Navi will be what puts AMD's RTG back on the map. More like, it better, else maybe the Processor side of AMD should give Raja what he wants and sell off the RTG to intel to cut losses, since the chip side is doing great now. Intel will buy because what RTG has is still light-years ahead of what Intel has.

And to anyone who says "what difference can Raja make when its all the same engineers" - replace Raja with Jim Keller, and Ryzen is your answer. The person handling the design can make all the difference in the world. However jury is still out (and will be for a while) if Raja's team is capable of pulling off a Keller-type of architecture for the RTG.
 
Either they are TOO POOR and can not do anything or they are subject to good business decisions and use their capital wisely. You have claimed both. Your first sentence regarding risk is ridiculous. They won't be going to a loan sharking operation for a bridge loan.

Have you SEEN their quarterly losses for the last... decade!?
 
Are you ABLE to actually defend your position or is it just going to be spurious one-liners?

Can't say I follow you. You're the one trying to argue that a company's financial situation has NO BEARING AT ALL in how they can react to a sudden change in market. You are the one who has yet to make a valid point.
 
Can't say I follow you. You're the one trying to argue that a company's financial situation has NO BEARING AT ALL in how they can react to a sudden change in market. You are the one who has yet to make a valid point.

Thanks for reminding me how futile it is to post on the internet. My point . . . . first you first said blithely that they are too poor to adjust their plans. When challenged on that point you acknowledged that they could get access to Capital but it might be risky . . . . is that you are a flip flopper. Pick an argument and stick to it or don't post 1 liner crap at all.
 
Thanks for reminding me how futile it is to post on the internet. My point . . . . first you first said blithely that they are too poor to adjust their plans. When challenged on that point you acknowledged that they could get access to Capital but it might be risky . . . . is that you are a flip flopper. Pick an argument and stick to it or don't post 1 liner crap at all.

Who is flip flopping? My poiny remains. AMD is too poor to react quickly and effectively to the mining craze. My point has never changed, its alwats been that, it has been adressed in every one of my retorts, quote me if it hasn't. Yours, however, is all over the place and vauge, from suggesting that them not having the funding has NOTHING to do with it, then to saying they can get the funding if needed, and apparently now I'm the one is flip flopping.

Take a look in the mirror to see your biggest enemy in this conversation.
 
This thread really is worthless.

As already stated, production is allocated months ahead.

There's a couple possibilities.
Hi
1. Polaris Production is set aside for ryzen or Vega which are the hot products

2. And severely under estimated customer demand

3. The switch over from 480 to 580 might have met with last minute production issues.

4. Last second revisions to Vega might have taken 580 production unexpectedly offline to get capacity up.

Anybody who claims gloom and doom is a fool and spreading FUD because production agreements are in place in advance and foundries will produce per agreement till they don't receive payment. And I haven't heard any word otherwise about lack of payment.

Case in point:fuhu was a great company offering a 7" IPS 1920x1200 screen tablet with tegra 4 (nabi) for $125. That's a pretty banging deal on a nice tablet, especially for kids.

Well foxconn was slow getting them out the door during the Xmas season and fuhu lost a lot of sales as a result. So fuhu said they weren't going to pay. And foxconn went for their jugglar.

As a result fuhu had to sell to a larger toy company to stay afloat. And what they sell now is nothing short of overpriced crap. As part of the settlement the nabies were sold for a fire sale. But fuhu stopped supporting them almost immediately.
 
Someone mentioned that Raj is trying to stay afloat long enough to get Navi out the door as this is his design. This was first brought up in Kyle's article about internal rumblings.

The artic isles is an incredibly powerful workhouse. GCN has a lot of flexibility. Problem is that flexibility comes with a price along with all that nice fp32 power. So it runs slower.

Think of it this way...a Dodge demon will burn a straightaway like nothing else at the strip. And a Porsche will slice curves all day at a track.

Well the race we look at is usually pretty straight. So the simpler design pipe with fewer tricks to process wins.

Now Navi I think is going to be a big advance on architecture. It will be tile based with an increased pre render throw away discard of triangles. But more than that it may break up the processor into a command issuer with multiple dedicated but separate engines with a communication layer (infinity fabric) similar to ryzen/thread ripper.

This could have enormous cost savings as smaller cores get exponentially cheaper to produce.

Before you just made one big dye that had cores deactivated and speed limited to the slowest section.

Binning and matching and reduced waste (no deactivated cores) will save you a fortune in production cost.

That said that algorithm for reducing the intercache communication better be damn spiffy. There's a price to pay for multiple cache coherency.
 
My guess is they are completely committed to vega. If the ether prices stay up i can almost garentee vega will sellout almost immediately even before anyone knows how it mines. To put it simply

would you rather have a 4-500 used card that is decent at mining.

or pay another 100 and get a chance at a potentially superior card.

I personally will be preordering a few of them when they come out.
 
Someone mentioned that Raj is trying to stay afloat long enough to get Navi out the door as this is his design. This was first brought up in Kyle's article about internal rumblings.

The artic isles is an incredibly powerful workhouse. GCN has a lot of flexibility. Problem is that flexibility comes with a price along with all that nice fp32 power. So it runs slower.

Think of it this way...a Dodge demon will burn a straightaway like nothing else at the strip. And a Porsche will slice curves all day at a track.

Well the race we look at is usually pretty straight. So the simpler design pipe with fewer tricks to process wins.

Now Navi I think is going to be a big advance on architecture. It will be tile based with an increased pre render throw away discard of triangles. But more than that it may break up the processor into a command issuer with multiple dedicated but separate engines with a communication layer (infinity fabric) similar to ryzen/thread ripper.

This could have enormous cost savings as smaller cores get exponentially cheaper to produce.

Before you just made one big dye that had cores deactivated and speed limited to the slowest section.

Binning and matching and reduced waste (no deactivated cores) will save you a fortune in production cost.

That said that algorithm for reducing the intercache communication better be damn spiffy. There's a price to pay for multiple cache coherency.


I'll go a step further, as this article confirms something I had considered - Vega, while still a Greenland core architecture, will be supplemented with all the technology they are planning to put into Navi, and are using Vega as a test for the tech: http://www.techarp.com/articles/amd-vega-memory-architecture/

The main focus is on memory architecture, but when you look at the Q&A, and look at all the new implementations they are putting around Vega, it makes it seem like Vega is their guinea pig, and any card limitations will be the limitations of Greenland itself.
 
I'll go a step further, as this article confirms something I had considered - Vega, while still a Greenland core architecture, will be supplemented with all the technology they are planning to put into Navi, and are using Vega as a test for the tech: http://www.techarp.com/articles/amd-vega-memory-architecture/

The main focus is on memory architecture, but when you look at the Q&A, and look at all the new implementations they are putting around Vega, it makes it seem like Vega is their guinea pig, and any card limitations will be the limitations of Greenland itself.

The HBC (Hybrid Cache controller) first showed up on the RX 480 series when they were demoing video-editing on the fly. Basically you are bypassing all the overhead of pulling from the m2 bus to the CPU back across the PCIe.

Under DX when you set aside memory for textures (and other objects) you usually don't specify if that memory exist in system memory, or on the video card (unless you specifically tell it so) That way the developer doesn't have to worry about how much memory is left on the video card. That's a function of video card drivers to determine which textures need to be put on system memory (to be pulled later) versus the video card. This bypasses all that and goes, "F it, we are going to put it all on the video card and eliminate all the overhead associated with missed memory pulls") ...which is an interesting concept no doubt given how textures have ballooned over the years.

The tech I'm talking about here is intelligently sending off all necessary triangle/texture/object data to a dedicated processor core which is a premise of efficient tile architecture. That is a tricky feat because some post processors and certain shaders need to reference data that is OUTSIDE their tiny viewing fulstrum. This requires syncing of cache controllers between chips and a lot of thread context switching. (Can't pull data from adjoining tile if it isn't ready) And Alpha layers present design issues with tiling architecture which was one of the achillies heels of the original PowerVR as you can't discard a sub triangle that is beneath a transparent texture. Most of these issues have been reduced, but there are still performance penalties.

Can it be done? Sure. It's tricky though.

Note: I am not an expert. I just like to fiddle with architecture designs. It's a fun little puzzle to figure out. :)
 
Newegg does have rx 570 and rx 580 series cards in stock. Only downside is the fuckers are only selling it in bundles. They know the market and taking full advantage of it. Wanna get some rx 570s? Buy some CPUs too lol
 
Just saw a listing on ebay for 6 rx 570s currently at $2200. Seriously that is that 366 per card and its not even over yet. That is called people totally crazy about AMD cards, they could easily grab gtx 1060, which frequently come in stock below 300 and run them at less 75w and hash at same rate of rx 570. The way people are over spending on AMD cards is crazy. Its actually pretty similar to how people will buy nvidia card even if AMD card might be a better value at some point for gaming. AMD has people hung over its GPUs, lol.
 
Or they could get a 1070 for like $50 more, which is even better at mining (at least on Ethereum).
 
yes and now all these companies are being greedy marking the prices up .
supply and demand and hiding behind exchange rates bullshit.
 
yes and now all these companies are being greedy marking the prices up .
supply and demand and hiding behind exchange rates bullshit.

Actually its not really the retailers. I have to give respect where its due. Newegg has not marked up a whole lot, if any. Their prices on 1070s and 1060s are fine. I have seen rx 580s come and go out of stock and they were at or near MSRP until they lasted, like 2 mins. Mining motherboard BTC 250 by biostar and h81 pro be Asrock. People ass raping on those on amazon and ebay. Newegg had them instock yesterday at 89.99 and 79.99 respectively. While resellers have them well over 200. So no companies are not marking them up, its third party resellers.
 
  • Like
Reactions: imre
like this
Yeh i have to say, its always America which has the best prices for gadgets with exception to some Asian countries and maybe Dubai. If you look at the UK for exmaple , their prices are the same price as the US dollar listed price but in GBP which i fine bloody horrid.
Now other countries like mine they mark it up a wack load its such bullshit.
Philippines surprisingly also, is a country where the gadgets are not well priced and some components come from there, its just ridiculous.
I therefore reckon that there is price fixing or some sort of deal between America and the suppliers, i have to say.
Such is the system it seems.
 
Yeh i have to say, its always America which has the best prices for gadgets with exception to some Asian countries and maybe Dubai. If you look at the UK for exmaple , their prices are the same price as the US dollar listed price but in GBP which i fine bloody horrid.
Now other countries like mine they mark it up a wack load its such bullshit.
Philippines surprisingly also, is a country where the gadgets are not well priced and some components come from there, its just ridiculous.
I therefore reckon that there is price fixing or some sort of deal between America and the suppliers, i have to say.
Such is the system it seems.

Tax works differently in the US, and so what you always see are prices without the equivalent VAT, which various by state hence why it isn't shown.
 
This is true. We miners have dozens of cards and you cant just throw a geforce 1070 into a rig with five other RX480s generally. Hell i cant even put an R9 390 or a fury in a rig with 480s.
Hm, I do this daily...why can't you do this? Seems to work just fine for me...
 
I guess i never tried linux. Under windows i had way to many problems getting it to work, keeping it running, or getting maximum speed.
 
I usually just run ubuntu server, load amdgpu-pro driver & nvidia driver. I run 2 instances of genoil's ethminer. One for cuda & one for opencl. The cuda detects my geforce cards and the opencl one detects my amd cards. No issues at all.
 
Back
Top