Mobile Polaris Discussion

I'm not an engineer nor designer, and don't claim to be... but, what were the engineer and designers at Alienware thinking when building the new laptops?

"How can we make these more unappealing to look at, thoughts Bob?"
"Well we can give the laptop an ass that can also act like a cheese grater."
"Good thinking Bob! Joe I'm not hearing any ideas from you."
"Well, we're going through this whole 80's retro theme lately, we can give it some pink neon lighting."
"Excellent, we can call it the Jennifer... someone who used to be sexy and desirable, but now has not aged well and never left the 80's."

 

The desktop RX 470 is no slouch though in terms of performance.
I would prefer the full 480 in mobile though.
As I said before, when its properly undervolted on both the stock core and VRAM, the 480 actually consumes same if not less than 1060.
Performance-wise, stock 480 (without undervolting- which also raises performance) is already on par with 1060 in most DX11 titles (only a mere handful favour 1060 - and even then the differences between the two are not enough to ruin anyone's gaming experience since both provide ample performance). 480 is faster in DX12, and of course, for the time being it practically obliterates Nvidia in Vulkan.
I don't understand how are people claiming that 1060 is 15% faster than 480, when most actual tested games show they are basically on par with each other in most titles, while conveniently setting aside that 480 also runs on SLOWER clock speeds than 1060.
Sigh... market monopoly with internet shills are doing numbers on AMD.
Mind you, if Nvidia actually had similar or same DX12 performance as well as optimized Pascal for it, included ASYNC and of course implemented what they needed for Vulkan, I'd probably go with them due to current lack of options from AMD side (though this likely has more to do with OEM's than AMD itself), but right now, I'd prefer to wait and see what happens, because I'd like to get AMD due to them actually looking forward and having some kind of hardware implementation for it... while Nvidia does not.


I really don't get alienware though.
What's with the neon flashing stuff?
It's just ending up as a proverbial power hog in an attempt to look 'cool' while in reality it looks quite bad.
I actually had a hard time with my current Acer during the night with some neon flashing blue lights on the mobile when I left it to do something while I slept and the blue light was just an annoyance.
I can already imagine what a pain a PINK light that shines so bright would do.

 

Putting the design itself aside, it's pretty obvious you can change colour and switch off the lighting on the AWs

 

Yeah, the lighting itself (except for the power button) can be turned off by a key combination. You can also change the color (I find light blue-green to be close to the common white while fitting well with the laptop's design). You can even change some zones to black as a way to turn off the lighting.

 

Think the joke flew over some folks... in which case, bad joke.

 

Lol... well, I prefer functionality over looks - though its not bad when both are combined.

Its' as if more expensive laptops have been designed by someone in an attempt to make them 'look' better, but in reality, they do not. They end up looking too flashy (at least they do to me), and of course, the price tag is then artificially bumped upwards to justify this somehow.

The alienware for instance seems way too blocky. All its missing is greebles - it's like trying to make the surface as detailed as possible. Whatever happened with technical efficiency and doing more with less?
You can make some very nice shapes and surfaces without having millions of greebles on the surface to make it look busy (which can be a bad thing).

But I digress... right now I would like to see AMD desktop grade Polaris 480 in mobile for a change.... or at the very least the 470.

If history is any indication, the AMD option was usually cheaper... this might not mean it will remain cheaper, but I'm thinking it will be.

 

Haha, I enjoyed the joke, and I almost gave you a like for it, but you lost me with the last line - poor Jenifer, whoever she may be! Haha

 

Has there been indication on whether the Alienware with Polaris will support Freesync on the internal display? Another thread is developing buzz about a nice 120Hz QHD (2560×1440) display, which appears to be an option.

(sorry, this is going off topic for this thread, it responding to the comments about the 2016 Alienwares)

Keyboard

Keyboard looks nice -- 2.2mm throw! Unlike Dell's current keyboards, it keeps a dedicated Home key and End key, the right Control key isn't too far out of place, and the Arrow keys are full size. And the left Alt key can be reached by thumb without much strain while fingers are over the home keys.

Ridges

The rear ridges might serve functions, not just style.

The heavier copper radiator inside receives the heat from the heat pipes and transfers it to the air. Most heat gets transferred on the inside to the air blown by the fans. The outside can become hot also, so like a heat sink or radiator, using fins and ridges increases the surface area in contact with air, so more heat per second can be transferred to the air.

When the outside gets hot, it could be hot to touch. The edges of the fins or ridges should be cooler than the base. In addition, the ridges reduce the reduce area in contact with your hand, so less heat per second is transferred to your hand, increasing the chance you can feel it and remove your hand before it does any damage.

(Ridges can also add strength without adding much weight. But if ridges were strictly for strength, they could be on the inside.)

Trapeziods

I'm not a fan of the trapezoidal, duckboat/aircraftcarrier look, but I imagine that might partly be function over style. Most laptops have a rounder bottom edge, and as laptops get thinner, if your fingers are too thick they might be difficult to pick up. The flat sides angled in toward the bottom might help people get their fingers under it and pick it up, and it will be the same no matter how thick gloves they are wearing. (Their target audience includes people outside the comfort and convenience of their family homes, such as people who move frequently between posts in the military or geoengineering.)

Corners

The sharp corners may be simply style. I think rounder corners would be easier to insert and remove from a bag or case, and would distribute stress better if it were to hit the floor. It may look more historically mechanical in a steelpunk sense, or construction/farm equipment sense, (or even a utility server rack sense), less biological or sporty in a luxury sportscar/motorcycle/jet/bullettrain sense. While curves can be a sign of aliens, it's also harder to design wide-appealing alien curves (see their desktops).

(Speaking of bags and cases, beware the sizes are a little larger to accommodate the rear radiator for higher GPU heat dissapation, so they might not fit in the same bags and cases as previous versions, and their accessory cases need to be enlarged.)

 

No mention of a Free-Sync display on the laptop (doesn't Lenovo have a laptop that has one, or was it HP?), but I would think the mini DP can support it. The 120Hz displays are G-Sync.

 

Ignoring nvidia for the time being, the 470 offer twice the performance, if not a bit more, compared to amd's previous 7970m. It's good to finally have such a jump from AMD's side.

The good side is that you can have cheaper gaming laptops from 15inch onwards. The bad side is that you can't get higher with AMD on laptops for the time being. It should be like an overclocked 980m anyways.

I just hope it is sufficiently cheaper than a 1060, to catter to the current 970m sweet spot in price, being 1k or sometimes even lower than that.

 

Isn't the mobile 1060 on the level of 980m performance-wise or just above it?
If that's the case, then the mobile 470 and mobile 1060 should be more or less on par with each other.

The desktop 480 in a laptop could offer a step above the 470 (mobile) and still offer some very valid solutions.
I would imagine AMD would still be cheaper. And while it wouldn't have anything to go up against 1070 and 1080 mobile directly for now... that would likely wait until Vega gets out, but it would also give AMD some kind of options in the mobile space as far as Polaris goes (and people like me who don't want to spend a lot for just a better GPU in a laptop - nvidia is really gouging the prices on them).

 

Mobile 1060, when working correctly (note that I've only seen about six laptops "work correctly" with one so far) is 25-30% or so above a 980M. It is a desktop 1060 with a 7% clockspeed shave. Desktop 1060s are ballpark (but generally less) performance than 980s. Considering that the 980 was around 30% faster than a 980M (due to 25% more cores, more clockspeed on core, and much faster memory) and that desktop and mobile 1060s' performance is largely dictated by a huge boost clock (the boost range is somewhere in the vicinity of 400MHz), and a working mobile 1060 can boost to the same actual speed that a desktop card may boost at, the mobile 1060 should cream a 980M, specs-wise. Also, despite having a 192-bit memory bus, its GDDR5 is clocked at 8GHz on mobile (as opposed to 5GHz on 980M) and its bandwidth improvements have achieved I believe somewhere around a +7-10% boost from Maxwell (I can't remember the exact number), the 1060 actually has far faster memory than 980Ms do; something like 30% faster effective bandwidth.

Of course, again, this requires a working 1060. So most laptops don't have that.

 

Ok... so what do most laptops that come with 1060 get performance-wise?

 

Within the range of 5% better to 5% worse than a 980M. Due to power/heat/etc issues. So far very few non-clevo models have handled 1060s well. The cards are extremely temperature dependent, and are extremely thermally dense on the core and thus extremely hot (TDP is an irrelevant stat for determining their heat output). Since there is no 960M replacement, and maxwell cards appear to not be created anymore, the whole laptop scene is basically turned on its head. All the OEMs are trying to shove 1060s into the models that barely handled 970Ms, and it's failing badly. Some have added a bunch of heatpipes that do more harm than good, like the MSI GS43VR where heatpipes run directly on top of (as in, touching or almost touching) the wifi card, slightly above the passively-cooled PCH, and touching the HDMI out port. Others haven't changed cooling at all and just exist perpetually at thermal throttle, like the Razer Blade 2016.

 

Which is why this would be a great time for AMD to release the 35W Polaris that is roughly equivalent to the Maxwell 960M... but it's nowhere to be seen.

 

That'd probably do well, if it was actually 35W, and actually held 960M performance in the majority of DX11 titles. This is where the AMD driver overhead issue comes hard into play. Their cards need to be 30% faster than the competition, so that when they lose that "30% faster" in DX11, they perform "equal". DX12/Vulkan game tests kill their CPU overhead from DX11 titles, granting their cards a ~30% performance boost over DX11 in general. They cannot be counted. Synthetic benchmarks like Firestrike etc too, are extremely well-optimized because they're basically constants. They don't count either. We need non-AMD-favouring (like Hitman, or Deus Ex: MD), non-Gameworks (like Witcher 3) DirectX 11 titles to test in. You could grab Killing Floor 2, Dying Light, GTA V, Black Ops 3, Battlefield 1, Evolve Stage 2, Crysis 3, etc to test. Those would show the real deal.

Also, Enduro will need to be at least as problem-free and effective as Optimus currently is. Optimus isn't perfect, but both it and Enduro are much improved since launch... however I've not seen any recent enduro-using laptops, or heard anything about it, so I don't know if they've actually improved it that much.

I don't know if they can do that. If they can, then more power to them, and by all means let them take over things.

 

Probably because it's true?

Here is a good sampling of 28 games where @ 1080p the 1060 averages 12% faster, and the RX480 is only faster (14%) in 3 out of 28 games. It says 25 games in title, but I count 28 games in the graph.

In 24 out of 28 games tested @ 1080p the 1060 is faster, with 16 games being >=12% faster, as much as 40% faster in one game, and a tie in performance in 1 game.



@1440p it gets a little better for the RX480, it's faster in 5 games out of 28 games, and the 1060 is still 12% faster on average, with 14 games >=12% faster, and the top 1060 performer is now 49% faster than the RX480.

 

^ Higher resolutions close the gap between cards more because the further you force a GPU bottleneck the less the CPU matters, so AMD's CPU driver overhead becomes less apparent.

And higher resolutions are more GPU load and less CPU load.

 

So, this has been on the back of my head. Alienware and AMD jointly created a VR backpack (it's pretty small), with it using a Polaris GPU. If it's supposed to be VR ready, does that mean it has an RX 480 in it? Or does it have an RX 470 like the new 15 and 17 do?

 

we have to wait and see... hopefully, it will be a 480... however, we might end up getting 2x 470.

 

Strange... on the current Alienware site retail pages showing the 2015 design (yellow accents on AW15, orange accents on AW17, 60° chiseled corners below screen, no Tobii), all the images have the word "polaris" in the name.

- Is "polaris" just the coincidentally also name of this design and has been there for years?

- or has AW been doing some early search-engine-optimization?

Makes one wonder if lower-end models with AMD's Polaris GPU will have the previous design,
rather than the 2016 design with more expensive larger radiator and Tobii-sensor screen (and fushia accents in current images, and square corners below screen).

The previous design handled the 980m which is reported to be around 125W TDP.
The RX470 is reported to be around 120W TDP also, so in theory the cooling might be sufficient if mobile chips are binned to stay below that limit.

(I think it just happens to use the same name, polaris. A brief search pulled up articles from August 2015 with polaris in the image name, but I didn't see any from earlier, even though there are articles from January 2015.)

 

VR doesn't like multi-GPU, so it won't be 2 x 470s.

 

The 1060 being too hot is good news for AMD. The chip performs what..around 980M levels +-5%? Pathetic.

A RX470 performing 10% above a 980M while running COOL and efficiently would be a winner and OEM's if pissed at Nvidia might just happily make the switch.

Then a Vega part down the line to compete with 1070.

 

Sure... though it performs properly in the Clevo lines with mostly low temperatures at max fans.

If the RX 470 desktop keeps its 1200MHz boost at 120W (I doubt) then it performs within 5% of a 980M in most DX11 titles (worse in OGL titles). If that translates to laptops, and enduro is as problem-free as Optimus is or better, then sure. Its price would need to be under the 1060 though. But this is a big "if". A very big one. Considering one needs give or take 180W+ for a RX 480 to hold its boost clocks at reference speeds, cutting off 12% of the cores isn't going to cut off 30% of the power draw if the boost clocks are the same. The RX 470 is a card I can see being undervolted etc to go down to the 120W range on laptops, sure. 230W bricks can handle it... not with a decently OC'd 6820HK, but they can handle it. This means it'll easily fit into machines like a P650RP6. But as with a mobile RX 480, AMD's problem is was and will continue to be the power draw it needs to provide the performance. The performance and even the heat isn't the problem.

If Vega is GCN, it won't ever hit laptops. Period. The RX 480 by itself draws too much power to be in a laptop without being in the same format the 1080 is. It needs 160W minimum for its reference clocks. A vega that's as strong as or stronger than the 1070 would be clean over 200W in laptop format. It is never hitting those smaller notebooks like the P670RS because it'd need a 330W PSU at the minimum, even if it ran cooler.

 

Actually, i've seen 480's with undervoltage with really good numbers. My guess is that amd had to really squeeze that chip to be any competition to pascal. When undervolted (or the correct voltage they should have been launched) they are quite ok.

A 470 undervolted would be a good chip for laptops if it had a good price.

 

The RX470 option would have to be a hell of a price/performance deal (including Alienware brand tax) for the Jennifer to look good.

 

Something I meant to post much earlier but couldn't find it, but I did just find it. This article kind of talks about throttling, power limiting, and undervolting on the RX 480 before the "power fix" driver came.
I say "kind of" because the images on the article are down, so it's a bit harder to read.

 

This is the issue I was talking about before. You're seeing undervoltage helping, because the card is at a TDP limit and doesn't hold its boost. You need to undervolt and such to get the TDP it's drawing to stretch. Think of it like how people handle HQ haswell/broadwell CPUs with the 47W TDP lock. Undervolt = less watts drawn for X performance = can clock higher at TDP limit. This is exactly what the desktop card is doing.

Reference RX 480 with all power limits removed has pulled 192W playing Witcher 3 at 1440p before. 177W in BF4. Undervolting at the 150W TDP limit makes it clock higher before it gets TDP limited. This is the problem with getting it into a low power format for the size of notebooks where it would have to compete with a 1060; the power draw would be too high.

The RX 470 on the other hand, might have something. It'll probably be still a 110W or so GPU with undervolted core & memory that comes with laptops, but it'll fit in the machines with 230W bricks. It won't be far overclockable either so it'll handle similar performance as a desktop. This means, give or take, it'll be around 980M performance. If it consistently holds its performance better than 1060s do, doesn't overheat easily, manages to SOMEHOW get into a 100W envelope, and costs less than a 1060 does, and comes in MXM-B format (for all the kepler-using people out there who want more performance and didn't pay for Maxwell) then they'll have something on their hands.

Though they also would need more CPU strength, so 6700HQs will be even more of a bottleneck than they currently are... hmm. It's all the TDP. That's all of AMD's issues. TDP.

 

Yes, it was said a few times before by a handful of people.

Basically what I thought, and what I said ever since the official release, also @Deks said it quite a few times as well. The release settings were in the inefficient zone, because raw numbers. Undervolt helps, undervolt and clock-down helps even more. They got 100W off (!!!) @-50%. Of course the performance is terrible, but it obviously can be set at whatever one wants. So since it's done for everything else (especially grIntel mobile chips, where I'm yet to see a single chip that performs at its full potential (stated boost clocks)), I don't see why it shouldn't be used for AMD as well? Would the performance be hurt? Maybe, depending on the chip quality and the settings they go with. Especially the later, since OEMs tend to "play-along" with AMD. I wonder which would be worse - not getting RX 480 at all, or getting a massively castrated RX 480? I repeated it multiple times before - I don't see Polaris as totaly unable to fit into a laptop, neither performance/power, nor heat/performance, nor whatever else you have in mind, like many want us to believe. I'm certain that binned RX 480 @120W would perform EQUAL with the desktop part. The OEMs fitted up to 2 1080s, which are 180W EACH (!!!) while outputting even more heat, costs as much as a good used car and the cooling is still not up to the task, but they are unwilling to make a few sacrifices for the sake of getting AMD in there. And by sacrifices I mean just different heat-sink and of course an AMD board. Everything else is there.

 

How much overclocking headroom does a GTX 1060 have? After seeing some scores vs my overclocked scores, it is actually slower than my single 980m, and now it looks like 470 might be an okay cheaper alternative. As long as it is cheaper.

 

If you're considering a working laptop 1060, probably with a Prema mod for a fair comparison against your 980M, then in the right notebook I'd suspect overclocking to around 1800MHz somewhat easy, give or take.

A stock 1060 should cream a 980M. The reason you're seeing 1060s being slower is because they're just not working correctly.

 

In case If it's not posted - Full RX470 in that.
http://radeon.com/en-us/alienware-unveils-new-generation-of-radeon-rx-470-laptops/

 

Cool, interesting, that's a full desktop RX470 in that laptop, no shortcuts, exact same specs according to that info - if it doesn't throttle then that's about desktop GTX 970 performance, so that's not bad if it's a good price, GTX 1060 is a little faster though I think from what I've seen.

 

I imagined that it should. My single 980m overclocked gets 12k in firestrike, and I can push it a bit more. I assumed 12k was the stock score of a 1060 but I've seen anywhere from 10k to 11k+/- scores. If they are already throttling like that, no wonder they offer a slight improvement over 980m.

We might need to disable boost 3.0 and deal with the horrible extra power draw

 

It's really more heat/OEM crap than power draw xD. 1060s pull less than 970Ms. Even if you managed to shove it drawing as much power as a 980M, that'd likely be with a solid overclock, surpassing a 980 easily.

 

So a DT RX470 is on par with a 980m. Sounds not that impressive tbh. Though I dunno how much TDP it uses. If significantly less then the 980m, it should run cool. Which in turn Dellienware could state that the new series has superior cooling!

 

Well, a GTX 970 desktop is about 20 odd percent faster than a 980M when comparing Graphics Scores of various 3DMark benchmarks on notebookcheck.net, so the RX470 should also be about 20% better than the 980M. That alienware link that Astrix showed shows the RX470 as 120W, so that sounds manageable, it's on a smaller node than the 980M which is also about 120W I think so it's probably gonna run hotter than the 980M.

 

The RX 470 is 120W rated, but considering its specs and rated boost speed being equal to the RX 480, it should be pulling a lot more. I'd estimate 150W average to hold its boost. Getting it into a 120W package with undervolts and memory undervolts, I can believe. 110W if they really stretch it, but I doubt. However if they make it pretty cheap, it's a good upgrade for Kepler users in MXM format. 980M-level performance in notebooks already housing 680Ms and 780Ms or 880Ms even is not bad, especially if it's very cheap. 970M price for 980M performance is good.

970 desktop shouldn't be 20% faster. If 970 desktop is 20% faster, and 980 is like 20% faster than that, then 980 should be something like 45% faster than 980M, which I know it isn't. Unless you're considering some OC'd 970s (superclocked counts) in which case OC'd 980Ms would need to be taken into account. Polaris won't overclock much, and it certainly can't be OC'd on mobile; the power draw would be exponential. 50MHz pulling 15W extra would not be unheard of.

See, I keep saying this, but I don't think most people here get it. You cannot look at the "rated" TDP for the Polaris cards. Tonga, Hawaii, did not TDP throttle. Polaris does. You WILL NOT hold the rated boost (and thus the performance desired) at the 150W limit for a RX 480, or the 120W limit for the RX 470. And that boost is what is going to be needed to keep its matching strength. This is why "undervolting" the Polaris cards makes it "perform better". It's because it's not holding its clocks, so undervolting makes it hold its clocks higher, and thus performance goes up. It's why it's going to be hard to get their full, unhindered desktop performance into laptops at power draw levels needed for laptops. If you've got a 230W brick and your laptop won't take a 330W brick, a RX 480 undervolted core/memory pulling 140W is not going to fly. It simply will suck way too much power. A RX 470's base is under 1000MHz. The boost is over 1200MHz. That's almost 25% difference in performance. You want to keep the mobile card over at LEAST 1150MHz or so, which is going to be difficult.

As for 980Ms, why do you all think they're 120W? I'm fairly certain they're closer to 106W or so than 120W at stock. If they were 120W stock, the P65x/P67x machines with 180W bricks wouldn't be able to overclock worth an iota, since power efficiency beyond stock goes right out the window on maxwell.

 

Is it confirmed then that AMD will deliver its Polariscards in MXM 3.0b format?
Would be a nice upgrade for all Clevo EM-Users who don't want to spend an arm and a leg for a 980m. Would be interesting to find out, if they could be made working in HM-Series.

 

They have at least one Prema-confirmed MXM card coming, but no idea which, or its specs.

 

It was posted earlier here, but yes, Prema said that one is coming. He said that it would be an alternative for upgraders. No word on what these GPUs are going to be like though. I would guess that it means 1070 in MXM-B, which means that they should fit an RX 480 as well. We'll see, but no one knows when.

 

I highlighted the bit in bold in your post. Here are the links where I got the info from, I'm not certain if the GTX 970 is reference or factory overclocked:
http://www.notebookcheck.net/NVIDIA-GeForce-GTX-970.146750.0.html
http://www.notebookcheck.net/NVIDIA-GeForce-GTX-980M.126692.0.html
These links show the GTX 970 being 20 odd percent faster than 980M.

 

That might be due to the differences in CPUs tested. The GTX 970 tests had the i7-4790K, while the GTX 970m was all over the place. One could do a comparison between the two with the Clevo P770DM.
The GTX 970 they had was also factory overclocked (MSI Gaming X).
As an aside, I have my monitor, desktop and Graphics Amp with me (again). I can probably do some more tests with my undervolted RX 480.
I did notice that overclocking the VRAM while undervolted didn't present any problems. Will re-test it, but just something I remember.

 

Come on guys, I thought it has always been crystal clear that 970 is at least 10-15% faster than 980M even with desktop CPUs, so full-blown rx 470 should be faster than 980m by the same margin. The only concerns I have is temperature and TDP, which might throttle the card as hell. Also competitive price. Let's hope AMD has dealt with these issues as we really need some competition.

 

These were Graphics Scores (not the overall score), so CPU would not come into it (maybe a difference of 100pts max, something like that, not significant). Yeah, I think you're probably right about the factory overclock.

 

I second all of that.

 

Oh, I was looking at the games, not benchmarks.
Another factor to consider is drivers, and since notebookcheck averages the scores (and uses it or the median), it can be negative if one GPU has been tested on multiple drivers (GTX 980m) and another on like two or three drivers (GTX 970). Still, about a 15 to 20% gap seems to be the trend between the data of the two.

 

I have also looked through comments section in that announcement page and this caught my attention:

Steve S says:

September 2, 2016 at 1:13 pm


What took you so long? The Radeon giant has awaken!


Reply

Jason Evangelho says:
September 2, 2016 at 3:23 pm

Just wait till later this year and 2017. We’ve got some awesome surprises for ya.​

This one has also made me smile

Tom says:
September 4, 2016 at 8:03 am


Maybe there is a God who answers prayers.
I’ve been searching (as a comeback) AMD fan, after a few years in excile, for some products that I can buy to myself and to my job.
High on the wishlist has been an AMD laptop (or at least one with AMD gaphics) and I can’t underscore how frustrated I have been while searching.
The best (of medium class) laptop processors AMD have been able to make is the fx 9830p, which cannot be found in even 1 laptop today, only downscaled
15w 9800p model AT IT’S BEST, often without a GPU or with a lousy old model.
While the Polaris 11 would (laptop) 460 model GPU, slightly slower (5-10%) than the desktop model, would have been enough for me, a fullblown desktop rx 470 (80% faster than rx 460), in the affordable pricing, is just perfect. Finally something AMD for gaming laptops. THX for the info, thx to DELL, ALIENWARE guys, THX to AMD, and GOD for listening to my prayers!​

 

Yeah, I saw that in the comment section too, and it was this one that is the most exciting - sounds like they could have some AMD laptop GPUs that are more powerful than the RX 470 before the end of the year perhaps.

 

Notebookcheck benchmarks for mobile GPUs have always been rather wonky, though. I look at them as the "minimum performance" I'll get with a card, rather than average performance. My 780Ms also far exceeded their benchmarks, even before I used the svl7 vBIOS which superclocks them.

980M should be be only about 10-15% slower than a reference 970 at worst. Superclocked 970s are a different story, of course, but as I said: 980Ms can overclock too. I'd say on average going to 1250MHz should be the standard minimum OC level, even for the newer "bad overclockers".

CPUs might also play a part in the games; particularly CPU hungry ones for higher FPS. But not most of them.