RX 480 just hit USA stores, initial supply 20x bigger than one from 1080

[LAwLz]

A Swedish store has the RX 480 8GB model listed now. I really hope it's just that particular store trying to screw people over with the price, because it is 2700 SEK. That's about 320 USD.

Why does Sweden seem to get screwed these days? The prices were fine 1-2 generations ago.

[DXMember]

11 minutes ago, Trixanity said:

Volkswagen owns a lot of brands:

  Hide contents

 

did I ever say they don't?

 

[xAcid9]

7 minutes ago, LAwLz said:

A Swedish store has the RX 480 8GB model listed now. I really hope it's just that particular store trying to screw people over with the price, because it is 2700 SEK. That's about 320 USD.

Why does Sweden seem to get screwed these days? The prices were fine 1-2 generations ago.

Reference XFX RX-480 with a slight factory overclock to 1328mhz + backplate cost $340 here in Malaysia.  

[Trixanity]

4 minutes ago, DXMember said:

did I ever say they don't?

 

No, was merely pointing out that VW owns a lot of brands; more brands than what most would think and most of them relatively successful despite product overlap. That and I pointed out all the brands they own.

 

I mean who would look at a Lamborghini and think VW?

 

 

[DXMember]

21 minutes ago, Trixanity said:

No, was merely pointing out that VW owns a lot of brands; more brands than what most would think and most of them relatively successful despite product overlap. That and I pointed out all the brands they own.

 

I mean who would look at a Lamborghini and think VW?

 

 

well to be honest Lamborghini is itallian and VW just bought bought them...

plus this is a tech forums, I don't even own a car...

[daniellearmouth]

On 25/06/2016 at 9:39 AM, Colt_0pz said:

Let's see if it compares to the 1070

Comparing a Ferrari to a Vauxhall Insignia here.

[IGJoe2192]

..... just a couple more days.... I cannot wait to get one of these cards. 

[Ryoutarou97]

On 6/25/2016 at 0:46 PM, mariushm said:

-SNIP-

So as a TL;DR would you say:

The problem is that the 1080 and 1070 are much larger chips than the 480 and production couldn't quite keep up with demand whereas the 480 has much larger supply (not launched too early? There was thought that Nvidia tried to beat AMD to market and launched early, though I have my doubts.) and is much faster to make because there are less failed chips.

[mariushm]

Short story, no. I think nVidia on purpose ordered a limited number of wafers with GP104 chips because they weren't sure how well TSMC's manufacturing process was, and didn't want to risk ordering too many and basically lose a huge amount of money. It costs about $5000-7000 to make a wafer holding something like 100-200 chips the size of GP104 and I guess they probably ordered only 50-100 wafers...  It takes months from the order to having the end result in your hands so in that time, they could only wait or work (and order wafers with) on other designs like the smaller GP106 chip which is meant to compete with RX480 and similar AMD cards.

Let's say the wafers held in total 15.000 chips in total and nVidia expected maybe 12.000 chips of various amounts of quality and 3000 unrecoverable chips but the manufacturing process produced wafers worse than expected, maybe only 10000 chips were actually functional. Let's say they managed to recover only about 1500 chips worthy of being labeled 1080 and maybe 6500 chips worthy of 1070 name, and the rest of about 2000 chips they could recover and make functional but only by disabling a lot of units and memory (basically they're under 1070 performance levels)

Before ordering a lot more wafers, they wanted to see if they needed to tweak the design of this GP104 so that more 1080 worthy chips would come out, because that would be more profitable. Their cost is fixed, $500k for 10k chips, plus maybe $20-50k in testing and packaging the dies, so we're looking at around $50 per chip. It's one thing to expect to sell 3-5000 1080 chips at 150-200$ a piece and another thing to end up with only 1500 chips in your hand. It's also not cost effective to get the few thousand crap chips (worse than 1070 in performance) and label them at as a potential 1060 because the profit per chip would be too small... they just either throw them away or keep them, hoping that in time they collect enough of them to sell something like 1065 or 1060 pro.

 

My guess is the stocks are limited for 1080 and 1070 because there's no chips out there, my guess is that after the initial order of 50-100 wafers nVidia ordered wafers with GP106 which is probably going to show up in a few weeks to compete with RX480 and then they waited to actually get the 1080 and 1070 chips in their hands from the factory to see how well they came out, and decide if they want to make a second revision of the chips to increase the number of 1080 chips coming out from each wafer (or hope that between these months TSMC gets their act together and improves the efficiency and quality just by optimizing their manufacturing process). Next batch is 3-4 months after a batch comes out, so hopefully since the first 1080 and 1070 chips came out the factory around April-May, you'd see a new batch around August-September and then you'd know if it's a new chip revision or same old thing, just a new batch of wafers.

 

AMD delayed the Polaris release by a month even though they had chips ready because they wanted to give stores time to empty their old stock, they knew the Polaris 10 and 11 are easy to make being smaller size, they had enough quantity of them, the OEM manufacturers already ordered a ton of them so the factory was already busy producing chips going straight to OEM companies, they know RX480 and RX470 would come out at $200 (and $150 for 470) and basically make a lot of cards obsolete.

nVidia took advantage of this and rushed to launch so that they could be the first to release cards made on new manufacturing processes, for the hype, for medals, for marketing.  It's almost a paper launch.

 

You have hot chips (reference cards can't stay at that 1666 mhz for more than a few minutes before throttling down), cheap cut down VRMs on 1070, small potential for overclocking on reference designs... maybe also a lot of companies held the better chips back for their non-reference designs...

 

[Briggsy]

2 hours ago, mariushm said:

You have hot chips (reference cards can't stay at that 1666 mhz for more than a few minutes before throttling down), cheap cut down VRMs on 1070, small potential for overclocking on reference designs... maybe also a lot of companies held the better chips back for their non-reference designs...

 

Not true actually. I would speculate a botched application of thermal paste or a defective chip in such a case.  I can only speak for my own case (although others have confirmed similar results), but my FE 1070 boosts to 1911Mhz out of the box and holds around 1850 indefinitely. With a 24/7 overclock of 2100/9000, My card holds 2000-2050Mhz on the core indefinitely under load, doesn't go past 78C at 65% fan speed, which is barely audible, and its on average 30C outside where I live right now and my A/C can't keep up, so my room is probably 24C on average while I'm gaming. With forced boost I can hold my card at 2126/9300 stable indefinitely with 75% fan speed, so if companies are holding on to the better chips for non-reference designs... that's pretty insane performance on them potentially.

[mariushm]

In your case, you were probably lucky.  You probably got a chip that was perfect from the point of transistor leakage, power consumption, maximum frequencies, but maybe simply couldn't be sold as 1080 due to a manufacturing flaw in the cache memory ram (maybe they had to disable a few hundred KB of cache, so they brought down the cache to 1070 levels and disabled cuda units and renamed it to 1070) or maybe too many CUDA units had faults (a cuda unit has 64 cuda cores, or whatever the number is on nvidia cards) .. if they didn't have the minimum number of cuda units to label it 1080, they labeled it 1070.

 

The stock VRM on the 1070 is not a very capable one, two power phases are cut out completely, it's a four phase design compared to 6 phase for the 1080, but can be enough to do some overclocking.

The reference 1080 averages to about 166 watts in gaming, with a maximum (limited by card) of 186 watts.

The reference 1070 averages to about 145 watts in gaming, with a maximum of 154w (not sure if limited artificially on the reference design).

Keeping in mind that 1070 comes with plain GDDR5 (which uses about 10-15w more compared to GDDR5X), we're looking at a difference of at least 30w between the power consumption of 1070 and the 1080 chips (166w vs 145w - 10-15w to account for more ram power) , so it makes perfect sense 1070 could overclock slightly better than 1080 (as long as you're lucky and got an otherwise good chip)...

30w is a lot when the chip in theory uses a maximum of 120-130 watts, it's about 20-25%. in theory by boosting the power consumption of the chip by 20-25%, you'll get good frequencies and if the fan helps, then the throttling won't be an issue. 

I suspect the 1070 is also helped by the fact it uses GDDR5 at lower frequencies, but powered by a stronger VRM compared to the one used for GDDR5X. The gpu chip may be able to clock better when it doesn't have to do so accurate timings and work with gddr5x, classic gddr5 is more "flexible"

 

When i said thermal issues and throttling, I meant what's happening with reference 1080 cards, as you can see shown here (just one of many) : http://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_1080/29.html

[Briggsy]

7 minutes ago, mariushm said:

In your case, you were probably lucky.  You probably got a chip that was perfect from the point of transistor leakage, power consumption, maximum frequencies, but maybe simply couldn't be sold as 1080 due to a manufacturing flaw in the cache memory ram (maybe they had to disable a few hundred KB of cache, so they brought down the cache to 1070 levels and disabled cuda units and renamed it to 1070) or maybe too many CUDA units had faults (a cuda unit has 64 cuda cores, or whatever the number is on nvidia cards) .. if they didn't have the minimum number of cuda units to label it 1080, they labeled it 1070.

 

The stock VRM on the 1070 is not a very capable one, two power phases are cut out completely, it's a four phase design compared to 6 phase for the 1080, but can be enough to do some overclocking.

The reference 1080 averages to about 166 watts in gaming, with a maximum (limited by card) of 186 watts.

The reference 1070 averages to about 145 watts in gaming, with a maximum of 154w (not sure if limited artificially on the reference design).

Keeping in mind that 1070 comes with plain GDDR5 which means about 10-15w LESS power consumption for the memory alone, we're looking at a difference of at least 30w between the power consumption of 1070 and the 1080 chips, so it makes perfect sense 1070 could overclock slightly better than 1080 (as long as you're lucky and got an otherwise good chip)..

30w is a lot when the chip in theory uses a maximum of 120-130 watts, it's about 20-25%. in theory by boosting the power consumption of the chip by 20-25%, you'll get good frequencies and if the fan helps, then the throttling won't be an issue. 

I suspect the 1070 is also helped by the fact it uses GDDR5 at lower frequencies, but powered by a stronger VRM compared to the one used for GDDR5X. The gpu chip may be able to clock better when it doesn't have to do so accurate timings and work with gddr5x, classic gddr5 is more "flexible"

 

When i said thermal issues and throttling, I meant what's happening with reference 1080 cards, as you can see shown here (just one of many) : http://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_1080/29.html

I know all about binning. And I definitely agree, no one should buy the reference 1080, the vapor chamber deal is smoke and mirrors imo, as the heat saturation point is hit fairly quickly, and the reference cooler exhausts the same volume of air on both the 1070 and 1080. The reference 1070 on the other hand isn't nearly as bad, and exhausts enough air to keep up.

[captain cactus]

15 minutes ago, mariushm said:

In your case, you were probably lucky.  You probably got a chip that was perfect from the point of transistor leakage, power consumption, maximum frequencies, but maybe simply couldn't be sold as 1080 due to a manufacturing flaw in the cache memory ram (maybe they had to disable a few hundred KB of cache, so they brought down the cache to 1070 levels and disabled cuda units and renamed it to 1070) or maybe too many CUDA units had faults (a cuda unit has 64 cuda cores, or whatever the number is on nvidia cards) .. if they didn't have the minimum number of cuda units to label it 1080, they labeled it 1070.

 

The stock VRM on the 1070 is not a very capable one, two power phases are cut out completely, it's a four phase design compared to 6 phase for the 1080, but can be enough to do some overclocking.

The reference 1080 averages to about 166 watts in gaming, with a maximum (limited by card) of 186 watts.

The reference 1070 averages to about 145 watts in gaming, with a maximum of 154w (not sure if limited artificially on the reference design).

Keeping in mind that 1070 comes with plain GDDR5 (which uses about 10-15w more compared to GDDR5X), we're looking at a difference of at least 30w between the power consumption of 1070 and the 1080 chips (166w vs 145w - 10-15w to account for more ram power) , so it makes perfect sense 1070 could overclock slightly better than 1080 (as long as you're lucky and got an otherwise good chip)...

30w is a lot when the chip in theory uses a maximum of 120-130 watts, it's about 20-25%. in theory by boosting the power consumption of the chip by 20-25%, you'll get good frequencies and if the fan helps, then the throttling won't be an issue. 

I suspect the 1070 is also helped by the fact it uses GDDR5 at lower frequencies, but powered by a stronger VRM compared to the one used for GDDR5X. The gpu chip may be able to clock better when it doesn't have to do so accurate timings and work with gddr5x, classic gddr5 is more "flexible"

 

When i said thermal issues and throttling, I meant what's happening with reference 1080 cards, as you can see shown here (just one of many) : http://www.techpowerup.com/reviews/NVIDIA/GeForce_GTX_1080/29.html

FYI, even the FE 1080 misses a power phase. It has room for 6 but only has 5 installed.

 

"We've put all of their effort and quality in this reference card" sure nVidia, and I'm a dog. They lied, big time. 

[VagabondWraith]

24 minutes ago, lots of unexplainable lag said:

FYI, even the FE 1080 misses a power phase. It has room for 6 but only has 5 installed.

 

"We've put all of their effort and quality in this reference card" sure nVidia, and I'm a dog. They lied, big time. 

It's a 5+1 phase, so technically it has 6 phases.

[goodtofufriday]

12 minutes ago, VagabondWraith said:

It's a 5+1 phase, so technically it has 6 phases.

Same way the 3.5 +.5 was 4gb on the 980

[Trixanity]

10 minutes ago, goodtofufriday said:

Same way the 3.5 +.5 was 4gb on the 980

 

[VagabondWraith]

8 minutes ago, goodtofufriday said:

Same way the 3.5 +.5 was 4gb on the 980

*970. But VRM are very much different than VRAM.

[Kinda Bottlenecked]

18 minutes ago, goodtofufriday said:

Same way the 3.5 +.5 was 4gb on the 980

 

No this isn't the same at all. 5 of those phases give power to the core. The other one goes to the vram chips and other microcontrollers. Definitely not the same way as the 3.5+5. 

[mariushm]

As far as I know , the 1080 and 1070 cards have a dedicated VRM for the core (6 phase UPI chip) and a separate controller for the memory. The 1080 only uses 5 phases and 1070 goes even lower with just 4 phases. However, the 1070 memory vrm is beefed up a bit, because gddr5 uses more power compared to gddr5x

 

 

[goodtofufriday]

21 minutes ago, Pohernori said:

 

No this isn't the same at all. 5 of those phases give power to the core. The other one goes to the vram chips and other microcontrollers. Definitely not the same way as the 3.5+5. 

Its more how the entire 4gb was dedicated. the entire 6 phases arent dedicated

[Kinda Bottlenecked]

1 minute ago, goodtofufriday said:

Its more how the entire 4gb was dedicated. the entire 6 phases arent dedicated

 

nonono. Different. Some may just tell you the total phases on a board but its pretty common knowledge(to an enthusiast atleast) that not all the phases are dedicated to the GPU itself. 

 

The 3.5+5 was a marketing hoha. Nvidia doesn't advertise the phases in their cards. You'll mostly see phases labelled like that 5+1, 7+1+1, 10+2+1, 12+2+1.

[goodtofufriday]

1 minute ago, Pohernori said:

 

nonono. Different. Some may just tell you the total phases on a board but its pretty common knowledge(to an enthusiast atleast) that not all the phases are dedicated to the GPU itself. 

 

The 3.5+5 was a marketing hoha. Nvidia doesn't advertise the phases in their cards. You'll mostly see phases labelled like that 5+1, 7+1+1, 10+2+1, 12+2+1.

Thats true I suppose. I was more trying to jab at Nvidia. Phases as far as GPU are concerned Ive never concerned myself with exactly how it works. I suppose I shouldnt comment on it aha

[Briggsy]

45 minutes ago, mariushm said:

As far as I know , the 1080 and 1070 cards have a dedicated VRM for the core (6 phase UPI chip) and a separate controller for the memory. The 1080 only uses 5 phases and 1070 goes even lower with just 4 phases. However, the 1070 memory vrm is beefed up a bit, because gddr5 uses more power compared to gddr5x

 

 

The 1080ti will probably use all 6 spots, making it 6+1. That's my guess anyway.

[SLAYR]

1 hour ago, Briggsy said:

The 1080ti will probably use all 6 spots, making it 6+1. That's my guess anyway.

 

1 hour ago, goodtofufriday said:

Thats true I suppose. I was more trying to jab at Nvidia. Phases as far as GPU are concerned Ive never concerned myself with exactly how it works. I suppose I shouldnt comment on it aha

 

1 hour ago, Pohernori said:

 

nonono. Different. Some may just tell you the total phases on a board but its pretty common knowledge(to an enthusiast atleast) that not all the phases are dedicated to the GPU itself. 

 

The 3.5+5 was a marketing hoha. Nvidia doesn't advertise the phases in their cards. You'll mostly see phases labelled like that 5+1, 7+1+1, 10+2+1, 12+2+1.

I came here for rx480 hype and people are only talking about nvidia 1080 VRMs ?

[Briggsy]

2 hours ago, SLAYR said:

 

 

I came here for rx480 hype and people are only talking about nvidia 1080 VRMs ?

as soon as any 480 hype or new rumor is linked people get up in arms and say "wait for drivers" or "propably fake" so people got tired of trying.