Like Daniel said. 470 will be one hell of a buy at 149.. for 4gb or even the 8gb for 179. I might be getting the 4gb model myself to through in my AMD computer that my nephew uses, old recycled rig with Athlon x4. Right now it has 7870 it should be a big boost. I am personally looking forward to that card. I think those are going to sell like hot cakes. Nvidia will have nothing to match at that price. That's gonna be just too good of a value to beat.
![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
AMD confirms Power Issue for RX 480, blames memory bus, will patch
- Thread starter pcgeekesq
- Start date
Like Daniel said. 470 will be one hell of a buy at 149.. for 4gb or even the 8gb for 179. I might be getting the 4gb model myself to through in my AMD computer that my nephew uses, old recycled rig with Athlon x4. Right now it has 7870 it should be a big boost. I am personally looking forward to that card. I think those are going to sell like hot cakes. Nvidia will have nothing to match at that price. That's gonna be just too good of a value to beat.
Check if whichever card you are interested in getting will support a non uefi motherboard. I can't get a definite on the 480 yet. Powercolor said no in a roundabout way. I would think there are a lot of people in the same situation as you and would welcome the new card at least in the interim between system rebuilds.
High PCIe Slot Power Draw Costs RX 480 PCI-SIG Integrator Listing
"What this means for AMD is that it cannot display the PCI-Express certification logo on the product or its marketing materials."
"What this means for AMD is that it cannot display the PCI-Express certification logo on the product or its marketing materials."
At least they did it in a timely fashion
CombatChrisNC
[H]ard|Gawd
- Joined
- Apr 3, 2013
- Messages
- 1,140
AIB's which can handle the draw via their 8-pin connectors might be still able to display the logo, according to the article. Maybe AMD will take a page out of their book and release a 480 2.0 with an 8 pin and not a 6 pin.
Shintai
Supreme [H]ardness
- Joined
- Jul 1, 2016
- Messages
- 5,678
They need to do something. As of now, using a reference PCB type 480 with other components as such voids warranty on them in the strict sense. And OEMs will pretty much instantly dump this product.
These kinds of certifications probably matter more to system integrators than the general public, and I imagine AMD makes more $ per card selling their reference boards directly than they do AIBs taking the chips and selling cards. System integrators won't want to use non-compliant parts in their systems which means you'll be less likely to see reference 480's in prebuilt systems, which means less $ for AMD, which AMD can't really afford.
Araxie
Supreme [H]ardness
- Joined
- Feb 11, 2013
- Messages
- 6,463
The true money is in the OEM market not in the typical consumer side, and this card without PCI Signature certification will be easily avoided out of that market by any OEM which generally use reference boards unlike system builders as example maingear which it's the primary AMD system builder which typically use AIB cards....
Boutique builders might still have RX480 in their systems, but major OEMs probably won't go for it when they can have the 1060 that costs around the same, performs better in DX11, has less CPU overhead which allows them to go with cheaper CPUs, and use less power, so cheaper PSUs as well.
D
Deleted member 93354
Guest
I don't expect us to see a firesale on RX480 reference designs any time soon. They are still selling out. However, reference production is likely halted or slowed. I expect they will alter the design and put on both an 8 pin adapter and move a power phase from the PEG to the PCIe. (Yes this will require a new reference board design, not just soldering on a new connector)
I will dare to say AMD has lost the battle with the RX480. I'm not saying this to poke fun at the AMD guys or pour salt in their wounds, but I am being a realist. The RX470 and RX460 series should be able to make AMD some money.
Luckily AMD didn't dump a ton of cash into the 480 either. I mean NVIDIA has to sell a TON of 10xx cards to make back the money they invested. So AMD might break even with the ROI sooner than NVIDIA. And I'm hopeful Vega will be their saving grace.
I will dare to say AMD has lost the battle with the RX480. I'm not saying this to poke fun at the AMD guys or pour salt in their wounds, but I am being a realist. The RX470 and RX460 series should be able to make AMD some money.
Luckily AMD didn't dump a ton of cash into the 480 either. I mean NVIDIA has to sell a TON of 10xx cards to make back the money they invested. So AMD might break even with the ROI sooner than NVIDIA. And I'm hopeful Vega will be their saving grace.
Last edited by a moderator:
Anarchist4000
[H]ard|Gawd
- Joined
- Jun 10, 2001
- Messages
- 1,659
They don't even need that. They just need to ship the cards with an updated BIOS to bias it towards the connector. That's all the driver fix was doing. As for the reference board they should just clock it lower. Doesn't take much of a clock/voltage reduction to save a lot of power. It's also possible that the situation corrects itself as the process improves.
Shintai
Supreme [H]ardness
- Joined
- Jul 1, 2016
- Messages
- 5,678
They still need a new certification. And by the rules it seems it cant be done on the current reference PCB. So they need some kind of second revision PCB that goes back to PCI-SIG testing.
D
Deleted member 93354
Guest
That's the equivalent of the VW Emissions scandal. "Sure we can fix the emissions problem. It will only cost you a few horsepower." AMD could do that. BUT, it wouldn't be very smart.
Anarchist4000
[H]ard|Gawd
- Joined
- Jun 10, 2001
- Messages
- 1,659
It might not even cost horsepower. A new metal spin or improvement in the process may simply allow them to lower voltages without adjusting clockspeeds.
A respin will not be as cost effective as having updated PCBs. Besides, they can release OEM-only RX480 with the updated PCB that gets certified, stop the reference design and just go with AIBs. All will be dandy and nothing lost except a few months of OEM sales.
JustReason
razor1 is my Lover
- Joined
- Oct 31, 2015
- Messages
- 2,483
This is likely the outcome to expect. Oems don't need flash just a working card. I actually see the 470 being the better oem card, with the 460 being the more plentiful.
Shintai
Supreme [H]ardness
- Joined
- Jul 1, 2016
- Messages
- 5,678
Adoption of an AIB card(Sapphire?) as a new reference may be the cheapest and fastest solution.
Anarchist4000
[H]ard|Gawd
- Joined
- Jun 10, 2001
- Messages
- 1,659
Problem is there is nothing wrong with the PCB. They can update it, but the power distribution is fine as is. Respin makes more sense, if it could ultimately address issues, as the parts will likely be used for an eventual refresh. (4X5s) Maybe there is a low end Vega coming, but odds are Polaris will be holding down the mid to low market for a while. Possible exception for APUs. There should have been some power saving features on Polaris that don't seem to be enabled. Dynamically adjusting voltages based on the quality of the FINFETs.
D
Deleted member 93354
Guest
I respectfully disagree.
They have two power planes. Some VRMs are on the PEG, some on the PCIe. Respinning IS EXPENSIVE. If you look at the revision number of the chip, it has been respun a ton of times already. If it was as simple as a respin, AMD wouldn't be in the mess it is in now.
Trust me, moving a trace line is a lot cheaper.
Last edited by a moderator:
Anarchist4000
[H]ard|Gawd
- Joined
- Jun 10, 2001
- Messages
- 1,659
But does moving that trace line allow you to increase performance on your refresh? It's not that much of a cost if they ultimately would have taken that action in the future. Again, this all depends on just what the power issues with the chip are. It could be anything from a sloppy process to bad component configuration. It's entirely possible they can't lower voltages because they needed all the performance they could get form the VRAM. If the refresh gets low voltage memory, suddenly the chip issues might be more of a concern. So even if the chip configuration stays the same, fixing the issue with a respin would still be worth it.
They can't just offload the power to the 6pin connector and get recertified. The 480 is a >150W card with only 150W provided. Sure the 6pin can pretty easily do more than 75W but it's only rated for 75W and no certification group will certify something that performs out of spec. They need to either use higher quality chips and lower voltage or provide proper power. As it is the 480 is out of spec no matter where they route the power.
I don't think AMD right now can afford a respin, with Vega looming so close. They still haven't released all their products for this gen, while their competitor already has pretty much all their lineup sorted out except for GP108.
They need to double down and focus on what's important, getting Vega and Zen out the door. After all that's said and done, and they have a little pocket change, then they can worry about next gen refresh.
If Vega turns out to be a hit, they might just drop Polaris entirely, I think.
They need to double down and focus on what's important, getting Vega and Zen out the door. After all that's said and done, and they have a little pocket change, then they can worry about next gen refresh.
If Vega turns out to be a hit, they might just drop Polaris entirely, I think.
Ieldra
I Promise to RTFM
- Joined
- Mar 28, 2016
- Messages
- 3,539
Vega and Polaris should be the same essentially, like Fiji and Tonga really. The only thing leading people to expect radically different characteristics is that leak showing ipblock revision being changed for Vega, we can't accurately gauge extent of changes from revision numbers changing though, but how likely is it they produce radically different uarch and support both eh? Serious question, it would unprecedented for two new architecture releases in same generation of GPUs
Shintai
Supreme [H]ardness
- Joined
- Jul 1, 2016
- Messages
- 5,678
They cant drop Polaris. I bet you Polaris is part of their roadmap until 2020 or so.
Vega is the same as Polaris, just a different product segment.
The difference is that Vega has more work done on it than Polaris (assumption due to the delay between Polaris launch and Vega launch), plus new IP block (apparently). If it works out well there's no reason to continue with Polaris since there are quite glaring bottlenecks on it.
Shintai
Supreme [H]ardness
- Joined
- Jul 1, 2016
- Messages
- 5,678
And how would that replace Polaris chips again?
The delay is because they cant afford it at the same time due to R&D limitations. And Polaris chips will need years of sales to recover its ROI. They simply need to fix this issue one way or another so they can sell RX480 cards and future rebrands with a PCI-SIG certification.
cageymaru
Fully [H]
- Joined
- Apr 10, 2003
- Messages
- 22,080
Alienware doesn't seem very scared. Aka Dell.
The Radeon™ RX-Powered Desktops from Alienware Have Landed – Radeon
Shintai
Supreme [H]ardness
- Joined
- Jul 1, 2016
- Messages
- 5,678
You are not going to see over night changes. If they use reference they will most likely just silently remove the 480 and replace it with the 470 over time. Unless they use AIB cards. Its not like these systems only sells with 480 cards either.
It's a tough one to judge. Especially if the leak about the new IP block is true, AMD will have to optimize for both Polaris and Vega, which will be somewhat/quite/very different driver-wise. I can't imagine them supporting Polaris if Vega is up to snuff.
Respin Polaris or cut down Vega.
Ieldra
I Promise to RTFM
- Joined
- Mar 28, 2016
- Messages
- 3,539
Just to be clear, GP102/104/106 are all using a different SM block compared to GP100. I don't imagine Vega and Polaris will be radically different from each other.
Anarchist4000
[H]ard|Gawd
- Joined
- Jun 10, 2001
- Messages
- 1,659
In theory the hardware to fix the issue is already in the chips, but not enabled for whatever reason. That's why I'm still of the mindset they will do a respin at some point to get it working. Unless it's simply that broken. Another example, look at the work Nvidia did to optimize paths on Pascal to get high clocks. Doing so doesn't change how the hardware behaves and is transparent to it's operation. Addressing the issue would allow them to lower voltage or increase clocks which easily fixes the issue. Even low voltage memory, at the same speeds, may address the issue.
Some of the Polaris functionality was backported all the way to Hawaii with the programmable ACEs. I'm increasingly thinking Polaris and Vega will be a bit further apart. Vega could be more tailored to integrated MCMs and such that it isn't practical for low to mid tier parts. It may not be the type of architecture you want to throw GDDR5 on and call it a day. Integrating Zen APUs and utilizing HBM2 could very well be the target there. I've seen some docs on targeting really small SIMD widths, so that flexible SIMD stuff may be in play as well.
Ieldra
I Promise to RTFM
- Joined
- Mar 28, 2016
- Messages
- 3,539
Yeah I remember the power gating and variable simd width papers you posted a while back
interesting for sureif its that broken, a mask respin won't do the trick, changes in the metal layer will take greater than 1q to get back and verify. More like 2q's, and they can't do what nV did with maxwell 2 to pascal, without more changes in the metal layers, doesn't seem likely they will do something like that, timing just doesn't seem to fit with it. AMD seems to be using a more automated process than nV, where nV is hand tuning their architecture and transistor layouts, you have to factor that in. Also doing metal respins will not be cheap added to that with chips in mass production you need to shift them off as you bring the new chip up, which prior to that you have do testing of the process again for the new chip.
Best approach is binned chips that function at lower voltages or downclock, OEM versions of these cards, could be down clocked, different boards with the addition of the 8 pin as well.
Alienware is akin to a boutique builder. They're not the same as Dell or HP or even Lenovo.
ManofGod
[H]F Junkie
- Joined
- Oct 4, 2007
- Messages
- 12,863
Alienware is Dell and has been for many years now.
Ieldra
I Promise to RTFM
- Joined
- Mar 28, 2016
- Messages
- 3,539
The Alienware brand is a boutique brand so to speak, like Apple in a sense. They will sell you subpar hardware for exorbitant prices and try to prevent you from messing with the hardware or upgrading by using custom motherboard designs and the like for some of the fancier systems they sell.
Alienware can suck a bag of dicks
But they're still a boutique builder in terms of market segment and sales. Their annual revenue is in the ballpark of 100 million USD, while Dell's is 55 billion, and HP over 100 billion.
Also, the average unit sale price for Alienware is around $3k, while both Dell and HP hover around $1k. Not even comparable, really.
From a quick look around google Cyberpower has the highest annual revenue of all boutique builders I know of, around $75 million, according to DataFox (which I'm not so sure is accurate, but it's about the only source I can find that's publicly available).
So yeah, Alienware is a boutique builder.
References:
Alienware Achieve 100 Million in Annual Sales Online with Omniture | Omniture | Alienware
HP Inc.
Dell on the Forbes America's Largest Private Companies List
Dell's dilemma--it's about pricing
Cyberpower Company Profile | Financial Information, Competitors and Executives
Report: PCIe 4.0 slots will deliver more power and higher speeds
Seems like AMD really future-proofed their RX480
Seems like AMD really future-proofed their RX480
Report: PCIe 4.0 slots will deliver more power and higher speeds
Seems like AMD really future-proofed their RX480
Interesting, I would think a pc 4.0 motherboard would be MUCH more expensive to make, also maybe different/more power connectors form PSU.
Shintai
Supreme [H]ardness
- Joined
- Jul 1, 2016
- Messages
- 5,678
Report: PCIe 4.0 slots will deliver more power and higher speeds
Seems like AMD really future-proofed their RX480
Haha!
300W+ via the slot is an awful idea tho.