Are Bummy Item Flippers Going to Make the 480 Hard to Get?

[PigSkinWetDog]

I've never gotten a GPU at launch, and just started following the gtx 1080 and rx 480 release. I see dead end job people typing about buying more than they need to supplement their low pay-checks like they do on SlickDeals. Anyone think these cards will be hard to get?

 

[NTMBK]

I've never gotten a GPU at launch, and just started following the gtx 1080 and rx 480 release. I see dead end job people typing about buying more than they need to supplement their low pay-checks like they do on SlickDeals. Anyone think these cards will be hard to get?

Maybe for a couple of weeks, while they milk the more impatient people. Just wait until the price normalizes, then buy.

 

[Mopetar]

I've never gotten a GPU at launch, and just started following the gtx 1080 and rx 480 release. I see dead end job people typing about buying more than they need to supplement their low pay-checks like they do on SlickDeals. Anyone think these cards will be hard to get?

Depends on how much supply AMD can push in the launch period and how much demand there is for the cards. If they can keep retailers stocked on a regular basis, then there's no market for trying to scalp cards because you can just wait a few days to buy one.

With a smaller die it's reasonable to think AMD will have more supply, but the lower price probably means more demand as well. The only real way to know is to wait and see unless there's some insider info.

 

[Qwertilot]

I doubt the 'more demand' bit - if it goes well then more net sales over the cards lifetime than the 1080 yes, but the 'burst' demand for the top NV card is enormous
(And very impatient, in a way I can't really see 480 buyers being.).

Anyway you want the better coolers, which might well not be there at launch. So wait a little bit for reviews of those, sane pricing and get one then if you want. A month perhaps.

 

[Stuka87]

Most retailers limit sales to one per customer. So its not like some guy is going to buy up a ton of them to flip on ebay. Sure they can have multiple accounts, but I doubt anybody is going to buy up any sort of large quantity, its just way too much work to have an account per card.

I bought my 7950 day one, and I had to wait three days for it to get back in stock. but NewEgg shipped it soon as it was in stock.

 

[n0x1ous]

I think card will have decent availability until the mining performance is determined. If its great they will start flying off the shelf and if not, availability should be fine.

 

[poofyhairguy]

I think the first batch will sell out early to the F5 crowd, but by August you can go to a store and get one.

 

[sirmo]

I think card will have decent availability until the mining performance is determined. If its great they will start flying off the shelf and if not, availability should be fine.

The mining performance should be well known based on the 5tflops+ figure. Mining relies heavily on integer ALU and that hasn't changed, if anything there might be some slight IPC improvement due to cache improvements.

380 was already a good miner, so 256-bit VRAM shouldn't be an issue. So you get a significantly more efficient miner than the 380 for the same price. I think it will be a significant upgrade for miners.

 

[poofyhairguy]

I think it will be a significant upgrade for miners.

If the software/drivers get fixed maybe. The problem is that recent 16.x AMD drivers have been much worse than old 15.x drivers for mining, but with old cards that isn't a problem because you can use the old driver for mining. The new cards will need the new driver, so unless AMD fixes whatever they did to the driver post 16.3 then Polaris cards will underperform with mining.

 

[sirmo]

If the software/drivers get fixed maybe. The problem is that recent 16.x AMD drivers have been much worse than old 15.x drivers for mining, but with old cards that isn't a problem because you can use the old driver for mining. The new cards will need the new driver, so unless AMD fixes whatever they did to the driver post 16.3 then Polaris cards will underperform with mining.

What's the issue with the driver? I wouldn't think a driver would play a major role in primitive compute performance like mining.

Found this thread: https://forum.ethereum.org/discussion/6108/radeon-software-crimson-16-4-1-performance

Looks like plenty of folks don't have any issues with the latest drivers and mining.

 

[boozzer]

most places would have limits per customer. you can always pre order if you know you will get it 100% maybe it will up your chance of getting one of the early stocks.

 

[nerp]

Find one with a good price. Preorder or sign up for auto-notify and be done with it. It if takes a while, cancel. In a few months they'll just be out there like any video card.

 

[amenx]

The main attraction of the 480 is its low selling price. Not sure how scalpers can make much money out of those drawn to the 480's low price. The 1080 is the top performing single card, I think many are willing to be scalped on it for that reason alone. But as stated, if initial inventories are limited, it can have an effect on price.

 

[poofyhairguy]

What's the issue with the driver?

Some minor OpenCL tweak. Mining is such an edge case it's hard to know exactly.

Found this thread: https://forum.ethereum.org/discussion/6108/radeon-software-crimson-16-4-1-performance

Looks like plenty of folks don't have any issues with the latest drivers and mining.

In that very thread there are people complaining about the new driver seriously dropping their hashrate:

With 2 R9 290X, I fall to around 40-45 mh/s from 54mh/s with this driver

I tried installing 16.x and it dropped my hashrate bigtime!!! From 91Mh/s to 60Mh/s.

I'm running 5 x XFX R9 390's and was getting 102 Mh/s on the 16.4 drivers.

I downgraded to 15.12 and am now running OC @ 147 Mh/s. In my village, we call that a 40% gain for free!

I know from my own testing it is around a 20-30% drop after 16.1.

There is a way with the genoil miner to use the old driver dlls to get around the issues with the new drivers, but who knows if that will work with brand new GPUs.

Of course that isn't the whole story for mining, it comes down to how much you can mine per watt. The mining software doesn't do a very good job using all the resources of Fiji (often mines very close to Hawaii) but people still buy Nano Furys for mining because the MH/Watt is lower than with Hawaii. If Polaris is "good enough" people with lots of capital who are doing mining as a serious business endeavour will buy them up to use in large mining farms, while smaller miner might instead jump on the 290s and 390s gamers are selling to get money for an upgrade. That might be the best case scenario for gamers who want a 480 for gaming and who plan to sell an old AMD card to buy that 480.

 

[sirmo]

In that very thread there are people complaining about the new driver seriously dropping their hashrate:

I realize that, but plenty of those with no issues whatsoever. So who knows what's going on. It's hardly a scientific method.

I can benchmark it when I get my rx480.

 

[poofyhairguy]

I realize that, but plenty of those with no issues whatsoever. So who knows what's going on. It's hardly a scientific method.

I can benchmark it when I get my rx480.

Me too, I plan to buy one as soon as I can get it just in case. Fortune favors the bold!

 

[swilli89]

What would be really awesome is if they are able to come out with a massive quantity of RX 480's. GloFo can do 30,000 wafers a month on 14nm. How many 232mm^2 Polaris 10 dies could that theoretically be? As far as I know the ONLY chips GloFo is currently building is P11 and P10, no CPU or APUs.

That's about 800 full Polaris 10 dies from each wafer. Around 2,000 P11 dies if die size rumors are correct, 130mm^2? If Glofo split its production evenly between P10 and P11 that would be..

12 Million Polaris 10 chips.
~32 Million Polaris 11 chips.

This is JUST with GlobalFoundries. There are definitely rumors about extending capacity with Samsung's own ~10,000 14nm wafer per month capacity but I'll leave that one alone for now.

I find it pretty amusing all the doomsday prophets harping on about how Apple will get all of the uncut Polaris 10 chips or that AMD is using all of their capacity on the console design wins.

I understand my numbers have a fair amount of assumptions in them but even with 20% less die harvests you still get a sense of scale about that GloFo can pump out. They would saturate 6 months of demand in 1 month of supply production..

 

[MrTeal]

What would be really awesome is if they are able to come out with a massive quantity of RX 480's. GloFo can do 30,000 wafers a month on 14nm. How many 232mm^2 Polaris 10 dies could that theoretically be? As far as I know the ONLY chips GloFo is currently building is P11 and P10, no CPU or APUs.

That's about 800 full Polaris 10 dies from each wafer. Around 2,000 P11 dies if die size rumors are correct, 130mm^2? If Glofo split its production evenly between P10 and P11 that would be..

12 Million Polaris 10 chips.
~32 Million Polaris 11 chips.

This is JUST with GlobalFoundries. There are definitely rumors about extending capacity with Samsung's own ~10,000 14nm wafer per month capacity but I'll leave that one alone for now.

I find it pretty amusing all the doomsday prophets harping on about how Apple will get all of the uncut Polaris 10 chips or that AMD is using all of their capacity on the console design wins.

I understand my numbers have a fair amount of assumptions in them but even with 20% less die harvests you still get a sense of scale about that GloFo can pump out. They would saturate 6 months of demand in 1 month of supply production..

Do you have a source on GloFo having 30k wafer/month capacity for 14nm LPP? It seems pretty unlikely that they have that kind of capacity deployed and that it has been sitting idle waiting for Polaris to ramp to volume production.

BTW - I don't think you can get 800 full P10 dies per wafer. Even ignoring defects, the number of 232mm² dies on a 300mm wafer is <250.

 

[swilli89]

Do you have a source on GloFo having 30k wafer/month capacity for 14nm LPP? It seems pretty unlikely that they have that kind of capacity deployed and that it has been sitting idle waiting for Polaris to ramp to volume production.

BTW - I don't think you can get 800 full P10 dies per wafer. Even ignoring defects, the number of 232mm² dies on a 300mm wafer is <250.

You're definitely right. When I calculated the area of a circle with a diameter of 300 I accidentally used the radius

So this puts us at closer to the 250 you mentioned which puts revised estimates at:

3.75M Polaris 10 chips/month
8M Poalris 11 chips/month.

For the 30k wspm figure:

Another foundry vendor, GlobalFoundries, licensed Samsung&#8217;s 14nm finFET process some time ago. In its New York fab, GlobalFoudries is capable of ramping up around 30,000 wspm of 14nm finFET capacity, according to Pacific Crest Securities

http://semiengineering.com/finfet-rollout-slower-than-expected/

Article doesn't mention if its LPE or LPP but I believe Samsung had LPP ramped up before this article was written.

 

[IEC]

Proof is in the pudding, but I'd expect a small die mass market part to have very high inventory on launch.

At least it better, given the price!

 

[Yakk]

As a card marketed to the mass market, TRG better have built up inventory quantities!

 

[IEC]

As a card marketed to the mass market, TRG better have built up inventory quantities!

Just a small nitpick, it's RTG (Radeon Technologies Group), not TRG. I was scratching my head for a second wondering what "TRG" stood for.

Source:
http://www.amd.com/en-us/press-releases/Pages/radeon-technologies-group-2015sept09.aspx

Which is still technically part of AMD. Even if Raja Koduri appears to have a lot of power over RTG.

 

[Slaughterem]

You're definitely right. When I calculated the area of a circle with a diameter of 300 I accidentally used the radius

So this puts us at closer to the 250 you mentioned which puts revised estimates at:

3.75M Polaris 10 chips/month
8M Poalris 11 chips/month.

For the 30k wspm figure:


http://semiengineering.com/finfet-rollout-slower-than-expected/

Article doesn't mention if its LPE or LPP but I believe Samsung had LPP ramped up before this article was written.

Actually a chip that is 14.1 X 16.4 on a 300 mm wafer with .2 defect will yield 239 candidate dies with 153 being good. Which calculates to 4.59 million P10 chips per month. But it takes 90 to 120 days to to complete a wafer. If P10 started ramp up in March they would have this amount available at end of June. A P11 chip with 10.2 X 12.6 would give 447 candidate dies with 348 good dies.

 

[Yakk]

Just a small nitpick, it's RTG (Radeon Technologies Group), not TRG. I was scratching my head for a second wondering what "TRG" stood for.

Source:
http://www.amd.com/en-us/press-releases/Pages/radeon-technologies-group-2015sept09.aspx

Which is still technically part of AMD. Even if Raja Koduri appears to have a lot of power over RTG.

LOL I keep reversing the T & R, and even when I don't my phone now does it for me. Should be fixed now, thanks.

Found it interesting that both Lisa Sue and Raja Koduri presented their respective products independently. Looks like they are both responsible for their respective products.

 

[Madpacket]

I realize that, but plenty of those with no issues whatsoever. So who knows what's going on. It's hardly a scientific method.

I can benchmark it when I get my rx480.

No scientific testing necessary. Just visit the Ethereum thread if you need evidence of newer drivers gimping performance or one of the many AMD driver release threads. There's workarounds that let you use newer drivers but not using the standards Ethash or qtminer. You have to move to Genoil or Claymore. 15.11 is still the best driver overall but 16.2.1 is a good compromise for Windows 10.

Also your assumptions about performance are probably based off Bitcoin or Litecoin mining.

The DAG / Ethash algorithm ethereum works on is completely different and requires a fat memory bus and lots of memory to get the best performance mining ether. No simple SHA256 going on here.

Number of shaders be dammned.

Example, my Fury X's vs 290/390's, They mine within 2Mh of eachother even though the Fury X has a what 40% more shaders?

Even with several different OpenCL mining ether apps in development this trend holds true.

Anyway Polaris mining speed is unknown but given the relatively poor performance of the 380/X and the same bus speeds I'm not expecting more than 22-23Mh per card without a serious core bump.

That being said if with a major core bump they hash over 25Mh @ 150W while undervolted I'll be one of those guys lining up to buy 20 at launch (if I can get my hands on that many).

If these cards end up good for ether mining and the price of ether holds you'll find it very difficult to get your hands on one regardless of availability

AMD was seriously caught off guard with Bitcoin mining but were saved when FPGA's and ASICS arrived. This likely won't happen again with ethereum as DAG very ASIC resistant.

What this all means is unless Proof of Work ends and Proof of Stake is introduced soon (very unlikely) or the price of ether tanks to unprofitable levels (very unlikely) it'll be a long time before you'll be able to buy a 480 at MSRP prices. Places like NewEgg gouged consumers when 290's first arrived due to miners suckimg them up. Hopefully they're prepared this time and take miners seriously.

More on Ethash / DAG here:

https://github.com/ethereum/wiki/wiki/Ethash