Pricing and OC headroom are going to be critical with the AIB 480s. Although with availability being limited they can probably get away with whatever pricing they want.
The 4GB models are the stars of the show. Sapphire's 4GB Nitro looked to be very close if not equal to a 1060, but at $220. This could do even better with 8Gbps memory.
As an owner of a 980, I'll be glad to switch back to the AMD camp on my main PC soon. The number of issues I've had with drivers and games I play with friends, and media playback, were too much of a pain.
My 7970 running on my old second machine trots along just fine, just with lower gaming performance. All this "nVidia drivers are better" has been a lie for a long time now. I'd say they're the same, both sides having issues here and there. You just have to pick what's better with what you are going to do.
The 980 uses much more power, and is much more expensive, so it's not really a useful comparison. The 970 would be a better comparison.
Around here anyhow, those two are fairly closely matched in price, though the really cheap 480's are still cheaper. And the 480 uses less power, and it's probably more future-proof too, even though perf at the moment is comparable.
In short: unless there's some specific game you feel the RX480 isn't going to run well enough, I'd pick that.
And if you think it's not fast enough, I'd be really really cautious picking last-years models. They're just not cheap enough, and the extra power use means more noise and typically earlier fan-death. And of course there are feature differences too.
So I'd never pick a 970 or 980 (unless you get an *extremely* good deal); if rx480 isn't fast enough in your games of choice, pick a 1060, 1070 or 1080 (and check benchmarks for that game first).
Oh please. The difference in power between a 980 and Rx480 is 30 watts at load. Its 100 freaking watts against a 390X. Since when did AMD GPU people give a damn about power usage? LOL.
Yeah. Those Fermi times when AMD people didn't give a damn about power usage, but nVidia people bought 450/460 etc which were SLOWER and consumed MORE and also more EXPENSIVE.
To be honest, we don't know if it is GloFo or Samsung 14nm process vs TSMC 16nm in general. Although what I've got from toms where they compared apple's chips from both factories and figured Samsung was actually better (although, there were some donws too, it did consume less power).
The main reason AIBs will be faster is that ref 480... throttles. So gains from 5% OC will be more like 10%-ish.
Depends on who you ask. Toms says Sammy is marginally better. Just about everywhere else says TSMC to one degree or another. I'd guess TSMC is better on average, but perhaps by less the normal binning variation of the chips (which is to say, insignificant). How this translates into larger chips with compounding effects on clock and power distribution and unpublished defect rates remains to be seen.
TSMC's 16nm aren't good either and is mainly desinged for SoC in mind. It hitting very high MHz doesn't mean it is very good. GloFo's 32nm SOI can also hit very high MHz.
i havent seen one negative thing about TSMC's 16nm process here lately. When you have a GTX 1060 pulling so little power, but giving you gtx 980 performance, i'd say it's a pretty good node shrink.
@Geranium: "TSMC's 16nm aren't good either and is mainly desinged for SoC in mind."
You'll have to describe what you mean by not very good. Though I think you are correct in saying the current process was initially designed for SoCs, like most foundries, TSMC designs multiple processes at a node with differing power and performance characteristics to address different markets.
@Geranium: "It hitting very high MHz doesn't mean it is very good."
So tell us what makes it bad. It certainly isn't clock rates. Power consumption and leakage characteristics look good as well. Switching thresholds, rise times, and fall times have to be good to achieve high clocks. Capacitance characteristics have to be reasonably good or they would negatively effect rise and fall times (clocks). You could perhaps call it inferior in the sense that the transistor size / density doesn't meet Intel's definition of 16nm/14nm, but neither does Samsung or Global Foundries. The only legitimate issue that I know the process has is defect rates. Given the same number of defects per wafer, smaller chips will have better yields. Low yields for relatively large (compared to SoCs) chips could be contributing to the current lack of supply. That said, this is not an unexpected factor when moving to a new process and, thus far, there is nothing to indicate abnormally low yields this early on in the process life cycle.
@Geranium: "GloFo's 32nm SOI can also hit very high MHz."
It also had good power consumption and leakage characteristics compared to other 32nm / 28nm processes. The process is generally considered a good process. It just gets compared with Intel's 22nm and 14nm 3D transistor processes. Even discounting the power and frequency gains, it clearly doesn't have the transistor density to compare. The "doxer" architectures used to show off the process doesn't do it any favors either.
Well, honestly, OC mode on cards like this always only add a few Mhz. You can always overclock it further yourself, and I would be we will see 1400+ Mhz on cards like this.
That cooler looks like overkill for a 150W card that doesn't overclock well.
Personally, I'd like to see a better blower cooler implementation from AIBs - one with a more substantial fin stack, and maybe heatpipes or a vapor chamber. That should cool the card well while cleanly exhausting all heat out the back.
I'd go for a quality blower too, SFF cases don't need to be dealing with all that heat. I doubt that monster PCBs like that MSI will even fit in a SFF case though...
I'm curious to see how that one fares in reviews. Someone was saying that in the past, Asus has used a plain extruded aluminum block in their cheap blower cards. No heatpipes, no fin stacks, just a block of metal, '90s style. Hopefully that's not the case with current cards.
I feel like blower style cards would benefit a lot more from having a 3-slot cooler. There's a lot of back pressure with trying to blow air out when 1-slot of the 2-slots at the exhaust exit is covered with ports, leading to reduced performance.
3-slots would give blower style cards a thicker fan, for more air volume being pushed, a larger volume of heatfins where heat can accumulate before it's transfered into the air, and 2-slots worth of exhaust space, rather than the usual 1-slot (or less when they also decide to shove a DVI-D port there, too).
Blower style cards get a lot of flak, but nobody really makes them right. No, not even the Nvidia reference blowers are great due to the 2-slot width, and therefore compromised performance.
Although, having a 3-slot width basically makes them incompatible with most miniITX cases which would really want to have a good blower style cooler, I still think someone should be making a legitimately no-compromise blower style cooler sometime. We've had 2.5-slot and 3-slot width open-air GPU coolers before, but never 3-slot width blower-style GPU coolers.
The 10 series blowers are a little bit better then in the past. The HDMI and DP outputs are so short that they only take up about half a slot in height. You can see it in the 1080n review here. http://api.viglink.com/api/click?format=go&jso...
It's just that miserable DVI port that's blocking so much airflow. You can't see any fins at all on the right side of that picture.
so the 6- vs 8-pin difference being that the 6-pin can provided 75W of power whereas the 8-pin can provide up to 150w. The article doesn't explain this, but would this be a correct presumption?
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
34 Comments
Back to Article
Stochastic - Tuesday, July 26, 2016 - link
Pricing and OC headroom are going to be critical with the AIB 480s. Although with availability being limited they can probably get away with whatever pricing they want.monkeybars - Tuesday, July 26, 2016 - link
With the 1060 out, they really can't get away with whatever pricing they want.Drumsticks - Tuesday, July 26, 2016 - link
The 4GB models are the stars of the show. Sapphire's 4GB Nitro looked to be very close if not equal to a 1060, but at $220. This could do even better with 8Gbps memory.damianrobertjones - Tuesday, July 26, 2016 - link
Or just buy a second hand 980?LordanSS - Tuesday, July 26, 2016 - link
Second hand 980? Give me a break.As an owner of a 980, I'll be glad to switch back to the AMD camp on my main PC soon. The number of issues I've had with drivers and games I play with friends, and media playback, were too much of a pain.
My 7970 running on my old second machine trots along just fine, just with lower gaming performance. All this "nVidia drivers are better" has been a lie for a long time now. I'd say they're the same, both sides having issues here and there. You just have to pick what's better with what you are going to do.
medi03 - Tuesday, July 26, 2016 - link
nvidia cards age terribly, so, uh, unless there is a great deal, nope.Morawka - Tuesday, July 26, 2016 - link
i'd take a 980 anyday over a rx480. more performance, better drivers (DX11).980's are selling for around $325 tho on ebay, there is a price gap
emn13 - Wednesday, July 27, 2016 - link
The 980 uses much more power, and is much more expensive, so it's not really a useful comparison. The 970 would be a better comparison.Around here anyhow, those two are fairly closely matched in price, though the really cheap 480's are still cheaper. And the 480 uses less power, and it's probably more future-proof too, even though perf at the moment is comparable.
In short: unless there's some specific game you feel the RX480 isn't going to run well enough, I'd pick that.
And if you think it's not fast enough, I'd be really really cautious picking last-years models. They're just not cheap enough, and the extra power use means more noise and typically earlier fan-death. And of course there are feature differences too.
So I'd never pick a 970 or 980 (unless you get an *extremely* good deal); if rx480 isn't fast enough in your games of choice, pick a 1060, 1070 or 1080 (and check benchmarks for that game first).
Chaser - Thursday, July 28, 2016 - link
Oh please. The difference in power between a 980 and Rx480 is 30 watts at load. Its 100 freaking watts against a 390X. Since when did AMD GPU people give a damn about power usage? LOL.medi03 - Tuesday, August 2, 2016 - link
Yeah. Those Fermi times when AMD people didn't give a damn about power usage, but nVidia people bought 450/460 etc which were SLOWER and consumed MORE and also more EXPENSIVE.Hypocrits.
xthetenth - Wednesday, July 27, 2016 - link
A new card with a warranty has some significant advantages, it's why used cards get priced lower than new ones of the same model.prisonerX - Tuesday, July 26, 2016 - link
With GameWorks shenanigans, a 980 with have the performance of Intel HD Graphics within a few months.medi03 - Tuesday, July 26, 2016 - link
Given pricing and availability of 1060, they sure can.FATCamaro - Tuesday, July 26, 2016 - link
OC mode is a joke at 5% OC. What's the point even... GloFo has bungled AMD's chances for a great card.medi03 - Tuesday, July 26, 2016 - link
To be honest, we don't know if it is GloFo or Samsung 14nm process vs TSMC 16nm in general.Although what I've got from toms where they compared apple's chips from both factories and figured Samsung was actually better (although, there were some donws too, it did consume less power).
The main reason AIBs will be faster is that ref 480... throttles. So gains from 5% OC will be more like 10%-ish.
hojnikb - Tuesday, July 26, 2016 - link
Exactly, It could just aswell be the architecture not being designed to run at high frequencies.Same story in the past with athlon xp vs p4
The_Assimilator - Tuesday, July 26, 2016 - link
There's nothing wrong with Samsung's 14nm process. There's plenty wrong with how GloFo has botched it.extide - Tuesday, July 26, 2016 - link
Oh yeah, like what?andrewaggb - Tuesday, July 26, 2016 - link
Too bad apple doesn't use glofo as well and we could compare tmsc to samsung to goflo.So far wasn't the consensus that tmsc is the better process?
BurntMyBacon - Wednesday, July 27, 2016 - link
Depends on who you ask. Toms says Sammy is marginally better. Just about everywhere else says TSMC to one degree or another. I'd guess TSMC is better on average, but perhaps by less the normal binning variation of the chips (which is to say, insignificant). How this translates into larger chips with compounding effects on clock and power distribution and unpublished defect rates remains to be seen.Geranium - Tuesday, July 26, 2016 - link
TSMC's 16nm aren't good either and is mainly desinged for SoC in mind. It hitting very high MHz doesn't mean it is very good. GloFo's 32nm SOI can also hit very high MHz.Morawka - Tuesday, July 26, 2016 - link
i havent seen one negative thing about TSMC's 16nm process here lately. When you have a GTX 1060 pulling so little power, but giving you gtx 980 performance, i'd say it's a pretty good node shrink.BurntMyBacon - Wednesday, July 27, 2016 - link
Defect rates?BurntMyBacon - Wednesday, July 27, 2016 - link
@Geranium: "TSMC's 16nm aren't good either and is mainly desinged for SoC in mind."You'll have to describe what you mean by not very good. Though I think you are correct in saying the current process was initially designed for SoCs, like most foundries, TSMC designs multiple processes at a node with differing power and performance characteristics to address different markets.
@Geranium: "It hitting very high MHz doesn't mean it is very good."
So tell us what makes it bad. It certainly isn't clock rates. Power consumption and leakage characteristics look good as well. Switching thresholds, rise times, and fall times have to be good to achieve high clocks. Capacitance characteristics have to be reasonably good or they would negatively effect rise and fall times (clocks). You could perhaps call it inferior in the sense that the transistor size / density doesn't meet Intel's definition of 16nm/14nm, but neither does Samsung or Global Foundries. The only legitimate issue that I know the process has is defect rates. Given the same number of defects per wafer, smaller chips will have better yields. Low yields for relatively large (compared to SoCs) chips could be contributing to the current lack of supply. That said, this is not an unexpected factor when moving to a new process and, thus far, there is nothing to indicate abnormally low yields this early on in the process life cycle.
@Geranium: "GloFo's 32nm SOI can also hit very high MHz."
It also had good power consumption and leakage characteristics compared to other 32nm / 28nm processes. The process is generally considered a good process. It just gets compared with Intel's 22nm and 14nm 3D transistor processes. Even discounting the power and frequency gains, it clearly doesn't have the transistor density to compare. The "doxer" architectures used to show off the process doesn't do it any favors either.
extide - Tuesday, July 26, 2016 - link
Well, honestly, OC mode on cards like this always only add a few Mhz. You can always overclock it further yourself, and I would be we will see 1400+ Mhz on cards like this.chlamchowder - Tuesday, July 26, 2016 - link
That cooler looks like overkill for a 150W card that doesn't overclock well.Personally, I'd like to see a better blower cooler implementation from AIBs - one with a more substantial fin stack, and maybe heatpipes or a vapor chamber. That should cool the card well while cleanly exhausting all heat out the back.
Mr Perfect - Tuesday, July 26, 2016 - link
I'd go for a quality blower too, SFF cases don't need to be dealing with all that heat. I doubt that monster PCBs like that MSI will even fit in a SFF case though...Death666Angel - Tuesday, July 26, 2016 - link
"card that doesn't overclock well."So how many of these unreleased cards have you had a chance to test in OC scenarios, yet?
extide - Tuesday, July 26, 2016 - link
I think ASUS has a blower-fan version of the RX480 coming out. Not sure how good it is, but def worth a look.Mr Perfect - Tuesday, July 26, 2016 - link
I'm curious to see how that one fares in reviews. Someone was saying that in the past, Asus has used a plain extruded aluminum block in their cheap blower cards. No heatpipes, no fin stacks, just a block of metal, '90s style. Hopefully that's not the case with current cards.JoeyJoJo123 - Tuesday, July 26, 2016 - link
I feel like blower style cards would benefit a lot more from having a 3-slot cooler. There's a lot of back pressure with trying to blow air out when 1-slot of the 2-slots at the exhaust exit is covered with ports, leading to reduced performance.3-slots would give blower style cards a thicker fan, for more air volume being pushed, a larger volume of heatfins where heat can accumulate before it's transfered into the air, and 2-slots worth of exhaust space, rather than the usual 1-slot (or less when they also decide to shove a DVI-D port there, too).
Blower style cards get a lot of flak, but nobody really makes them right. No, not even the Nvidia reference blowers are great due to the 2-slot width, and therefore compromised performance.
Although, having a 3-slot width basically makes them incompatible with most miniITX cases which would really want to have a good blower style cooler, I still think someone should be making a legitimately no-compromise blower style cooler sometime. We've had 2.5-slot and 3-slot width open-air GPU coolers before, but never 3-slot width blower-style GPU coolers.
Mr Perfect - Wednesday, July 27, 2016 - link
The 10 series blowers are a little bit better then in the past. The HDMI and DP outputs are so short that they only take up about half a slot in height. You can see it in the 1080n review here.http://api.viglink.com/api/click?format=go&jso...
It's just that miserable DVI port that's blocking so much airflow. You can't see any fins at all on the right side of that picture.
sonicmerlin - Tuesday, July 26, 2016 - link
This card will be pulling 220W minimum.angrysand - Monday, August 1, 2016 - link
so the 6- vs 8-pin difference being that the 6-pin can provided 75W of power whereas the 8-pin can provide up to 150w. The article doesn't explain this, but would this be a correct presumption?