AMD, where art thou?

Means nothing for a Notebookreview forum though. I'd be more interested if AMD can get their mobile GPUs to function to their full potential and not 50% POS.

 

50% POS? I don't understand. AMD doesn't have an incapability to get mobile GPUs to function at proper power (though Enduro leaves much to be desired)... AMD has no GPUs that can fit in the format and be any sort of worthwhile. Their entire lineup from midrange to top end is 275W+ GPUs.

 

This is another wall of text. But this has been my experience, whether some on this forum think it's deluded or not, but this is my impression with AMD and NVidia with mobile gaming. I'll be honest, I buy Clevo for gaming and massive CPU for compiling/encoding and other fun things I may try. The strengths AMD flaunts in their marketing, I don't use or care.

I don't care about AMD's desktop range right now since it doesn't translate to mobile. For NVidia it's different, when I see performance from GTX 960-970 I have a good idea what to expect on a 980M. With AMD, in the past, when I saw the xx70 series, did not translate. I had perpetual GPU utilization issues, some games as low as 30-40%. Remember Max Payne? When my 7970M died and I got 675MX, I was still screaming at the 7970M, why can't you do this? Why can't you have 90% GPU utilization?

I don't care about GPU compute because when it comes to compute, I'm still a big CPU fan. I love Intel i7 when I do any compiling/encoding. I do zero CAD/3D work. Once NVidia solves the asynchronous issues and fully implements in their drivers, my interest in "compute" ends.

My expectations are slim to none.

 

Even in the desktop space, from everything I have seen AMD (including their new Fury stuff) does OK against NVIDIA running stock. When the overclocking starts, AMD gets spanked and sent to bed with no dinner. That pretty much ruins it as far as I'm concerned. Adding to Mr. @Zymphad list of don't cares... I don't care about stock. If I cannot overclock the living crap out of it and have it run extremely well and survive repeated overclocked and overvolted benching sessions, the product is utterly worthless and irrelevant to me; even if it runs like a Swiss watch at stock clock speeds. If they can get their act together again and execute on that successfully, and beat their NVIDIA counterpart in overclocked benching, then I might consider the brand a viable option once again. It has been a very long time since they were able to do that, and I have very little confidence in their ability to execute on this.

Now... having said all that... I sure do wish they would. I am so sick of the Jolly Green Giant's monopoly and their lackluster performance in the absence of competition. I'm not thrilled with either brand and it would be nice to not have to choose between the lesser of two evils. I miss the good old days, where they were exchanging blows for the victor's crown... back when we had option 1A and 1B, not a clear #1 and a distant trailing #2 that most people don't have any love for.

 

It is not a wall of text. I'm simply correcting your statement that is liable to give unknowing people the wrong idea.
Your statement is: "AMD cannot get mobile GPUs working beyond 50% of their supposed power" (paraphrased).
My counter statement is: "AMD's mobile GPUs work fine. They are just weak."
It does not matter whether you like or dislike or care or don't care about AMD. It does not matter whether nVidrosoft is stronger or weaker. What matters is that your statement is incorrect. IT IS AN ENTIRELY VALID POINT THAT AMD HAS NO WORTHWHILE OFFERINGS IN THE MOBILE MARKET. And I would not counter that statement at all. But their cards do not work at 50% potential.

ENDURO, as I pointed out, caused much issues with utilization for AMD GPUs, and while those utilization problems are mostly fixed, Enduro is still a pain in the behind. It does not help you now, and I get that, but RIGHT NOW if I tossed two 8970M cards in my P370SM3 I would have them working just fine at proper utilization.

nVidia CANNOT "fix" async compute with their current cards and drivers are not going to help. The CARDS THEMSELVES LACK THE HARDWARE. MAXWELL WILL NEVER DO ASYNC COMPUTE FOR DX12 BEYOND 31+1. It is a hardware bound fact. Your 980M will NEVER IN ITS LIFE PERFORM GREATLY UNDER ASYNC COMPUTE IN DX12 ABOVE 31+1. That is pure fact. Your waiting is pointless, unless what you are waiting for is to upgrade to a Pascal GPU next year.

I don't need or want you to like or even care about AMD. What I wanted was that the correct idea is displayed. That's all. Feel free to hate AMD and or nVidrosoft as much as you want. Just be fair and don't say things that aren't applicable or not true, because there's a ton of lurkers and people who aren't actively replying that can get the wrong idea.

I know about AMD's downsides, and AMD has quite a few that prevents me from ever buying their cards unless I plan to use only a single GPU. I'm not denying the overclockability (or lack thereof) that they possess, and I understand why AMD isn't an option for you and many on this forum. I simply wanted to correct his statement that all AMD mobile cards only use 50% of their core.

But like you, I wish they would hurry the hell up and put nVidrosoft in their place.

 

Speaking about high TDPs, it's been a while now, a lot of 980 laptops should be on the market. Can we realistically say that there is an improvement over the last year, or they just bumped the TDP and everyone started passing-out by the sheer "awesomeness". Yes, I'm speaking about the VERY popular performance per watt, much used against AMDs. A quick glimpse here and there tells me that it is actually worse than last year.

 

Here here D2 Ultima. AMD are still a release generation behind anyways. Their M395X is comparable to the 880M. Fairly powerful but inefficient.

Let's hope AMD release a new gen card next year and "skip" a generation to be on par with Pascal. Doubtful but possible.....

 

New move from AMD - Radeon Crimson, This is looking good atleast a bit to end the nGreedia's filthy business lately, given this is not just a smoke and mirrors cosmetic show.

 

Wish they add DX triple buffering too.

 

Framerate target? Multiple AA formats? Tessellation reduction features? Specifying OpenGL triple buffering instead of having people think they're getting it in DX titles like NCP does? Per-title overclock settings?

Yes. Goooood. Good AMD. GOOOOOOOOOOOOOD.

I think that's a limit in DirectX and not AMD/nVidia

 

D3DOverrider and RadeonPro can enable triple buffering in Direct3D

 

Is that a DX hack?

 

It is not. The function is present in D3D, AMD and Nvidia drivers don't include it for whatever reason.

 

I see.

Well then. Screw whoever is responsible.

 

I remember talking to DSR developer that hence he included not just custom resolutions but AA and other features too he could add triple buffering too.

He said that D3DOverrifer's triple buffering is actually not real triple buffering as it was designed but called "add frame to queue" or so. It still WORKS by limiting tearing. It's side effect is the latency of 1 frame.
So that is why AMD/Nvidia don't put this feature. But I think they should. Just call it different in your AMD Software Settings and that's it! We should send this feedback to them.

 

So... apparently... somebody managed to get OVER 100% SCALING in Crossfire. http://www.3dmark.com/compare/fs/6525252/fs/6525107

I'm stupefied. @Mr. Fox @octiceps @n=1 you got any light to shine on this O_O

 

I think I mentioned it before, I've seen this but only with CrossFire.

 

2.5% is within margins of error

 

OK, I knew Crossfire scaled well but...

 

My best guess is that Fury X was somehow getting underutilized in single card config.

Also please grab Rebel Galaxy because it's awesome kthx

 

LOL ok. I see you actually playing it these days. I almost never see NBR users actually gaming on steam, funnily enough

 

Yeah 123 hours sunk into it already. For a $20 game (10% off first week so actually got it for $18), it was worth every penny I paid for it. Hell had I known it was this addictive and was done by a 2 man band, I'd have gladly paid the full price. Oh and a modding guide was just released recently, and already people are modding the hell out of it.

 

A little update on the situation with Arctic Islands.

http://wccftech.com/amd-1416nm-arctic-islands-launching-summer-2016/

I know it's nothing we didn't expect, but it's nice to have something to go on for Team Red.

 

Aha! Some legitimately good news. Looks like AMD is bringing FreeSync to HDMI. While it doesn't appear that any current monitors will be using it, AMD has plans to unveil displays for both desktops and laptops that will do FreeSync over both DisplayPort and HDMI early next year.

http://www.pcworld.com/article/3012...-support-for-hdmi-mobile-displayport-hdr.html

http://m.hexus.net/tech/news/graphics/88694-amd-announces-freesync-hdmi/

 

Let's hope that FreeSync wont require BIOS cookies or whatever so one can upgrade to FreeSync panel down the road. Not that it would ever happen to 16:10, but for the rest, and I can always hope, so looking forward to it.

 

Ohh... 16:10... It just so happened that I am using 16:10 T410 right now and I must say 2 things:
1. Using 16:9 for years kinda spoiled me because now it's distraction for me to see black bars on TV shows and first 2 days I couldn't get why I saw those bars at all
2. I still like it and it's more cofortable and I want 16:10 panels! I guess it will never happen though.

 

Well, the R9 M295X and R9 M395X have all but disappeared from Alienware's lineup. AMD has entirely exited the high end mobile graphics arena. When I pressed an AMD rep about the fabled Tonga XT R9 E8950 MXM, he stated it would not be coming to consumer computers, and would only remain for embedded devices and equipment. AMD is dead fellas, long live nGreedia

 

Shockingly sad yet not surprising if true. Arguably they already had exited.

The M295X was hardly that common. While the M290X is simply a well binned 7970M with 4GB vram.

 

I'm not giving-up on them just yet. There's still W7170M.

@James D - That's what I like about 16:10, you can do whatever you want (especially productivity stuff) and to be honest, I don't mind the black bars (actually I like them, feels roomier, so to say), but that's personal preference.

 

16:10 FTW!

 

What I wouldn't do have a 1920x1200 RGBLED in my M18x....one can always dream

 

Yeah, I should've made a shootout between my brother's M18x-R1, M17x-R2 RGB LED and my DreamColor, as well as the run-of-the-mill stock 8740w WUXGA panel. It's not like one can really feel the difference as you add camera and the viewer's screen into account, but I guess the difference can still be felt. Arranged by awesomeness they go - DreamColor, RGB LED, M18x and the stock WUXGA, which is nothing to write home about. That's for the beauty of them, for usefulness, M18x is instantly right-off the list. Yeah, these missing 120 pixels are damn obvious, and looks a wee bit grainy in my eyes, not to mention that it just doesn't feel right, but again, that's personal preference. As for the top performers - the DreamColor is no match in terms of viewing angles, colors, and color space selection, the RGB LED is faster and the colors are a bit more natural (stock for stock, the DreamColor is a bit more reddish, but that's how they are tuned from factory, everything is fixed after calibration). All in all I WANT 16:10 BACK! Till then the DreamColor and M17x-R2 RGB LED would do (maybe, but most likely wont get it).

 

I wonder it's because AMD has failed to overcome the hurdle with switchable graphics in high performance use. AMD really should be partnering with Clevo/Asus rather than Alienware to introduce purely dGPU solutions and tell Intel to pound sand with their switchable graphics.

 

No, enduro works fine now. It was rough in its initial implementation but it has really come out to be ok, but optimus is better. I think it is more along the lines of they are hemorrhaging money and don't have the ability to hire proper engineers to revamp/redesign GPU architecture.

 

Raidriar most likely touched upon the closer cause of the problem.
Money issues.
AMD is currently focusing on other things such as desktop market share I think, and possibly preparing themselves for 2016.

So, if no mobile products will arrive before the end of this year... it is possible they are focusing on 2016 (at least I hope so).
I would love to see a mobile HBM gpu based on Fiji such as R9 Nano, or better yet, Arctic Islands (which would make more sense if they are waiting to go with the new manuf. process - meaning HBM2 might be in the works for mobile).

 

Its probably very easy: AMD have no platform to take on Maxwell.
Maxwell was first of all very efficient which is everything in mobile.
But it was also just as cheap as AMD this time, making any move from AMD moot.

2016 will undoubtly be a year where AMD will return to notebooks again since they too are aiming for 2x the performance/watt like Pascal. Rumor has it that they will use GDDR5X on some cards like Nvidia, so expect a mobile MXM GPU from the 400 series with GDDR5X as well.

 

Again, there's still W7170M. I'm eager to see SPECviewperf 12 comparison between W7170M and M5000M. I'm pretty sure it would game well as well. So not all is lost this year, just focusing on winning designs. Not that pros wont get the Quadro for no obvious reason (OK, there are a couple of apps that favor Quadros, but most get it just because it's nGreedia and it's more expensive (false understanding of quality), just like with the consumer market). Then there are students on budget, as well as a few others open minded people, that would get the W7170M, because they know that it would get the job done, for far less money.

 

Everybody says "Maxwell". I gonna say Kepler. It's the exact architecture which jumped from AMD's products one step ahead andonly because Nvidia always hides best generation chips for later we didn't notice how superior it was. 7970M was nice competitor... to 680M. But when Nvidia brought full 780M with overclocking potential AMD offered rebrand of the already choaking card which died from overheating and overcklocking.

Maxwell isn't actually power efficient at all. It's same step in efficiency as Kepler did from Fermi albeit Nvidia optimized some low power alghorithms which strike often enough (and some say it's too often). AMD had nothing to compete with Kepler mobile and that was the first sign that it won't be able to compete with Nvidia Desktop.

 

The R9 m395x can be found on Alienware Canada... and that's about it.
Welp, once stock depletes, the only AMD gaming laptop left is the Lenovo Y700 (with Free Sync). That... that just doesn't sound good.

 

Oh well, atleast they have more reasons to release a mobile chip as soon as possible if they have one in the pipeline

 

http://www.dell.com/us/p/alienware-...4&ven2=,&ven3=772503095648278021&isredir=true
Found it on Dell US. I have no idea why it's hidden, though. Searching its code through Dell doesn't even find it. Seems like Dell doesn't want AMD in their laptops for Christmas.

 

Good catch. It definitely isn't out in the open.

 

To be fair, AMD is correct to bet on 2016, and not try to introduce many new chips in the near term. Any chips introduced this winter would have less than a year's shelf life before being superseded - likely in a major way - by 14nm and 16nm GPUs, and it thus doesn't make sense for them to make a significant investment in 28nm GPUs now when there would be limited time to recoup that cost, and it would have meant fewer resources focused on 16nm or smaller GPUs. True, things would have looked a little better in the short term had they introduced more refreshed 28nm chips, but it's the right call not to do so.

 

To be fair, AMD hasn't introduced anything new to mobile since 7970M.

To be fair, AMD has been absent in CPU for both mobile and desktop since Phenom II in 2010. Llano, Bulldozer and successors have been complete and utter failures.

 

Indeed! All the more reason to make sure they don't mess up in 2016.

They can't afford it (literally) unlike Nvidia! They did release the M295X which like the 780M was a fully unlocked chip vs 7970M/680M. Unfortunately unlike 780M the card was NOT widely available nor did it clock well in mobile...HOT

 

http://wccftech.com/amd-further-embraces-open-source-with-gpuopen-for-games-and-compute/
http://www.anandtech.com/show/9853/amd-gpuopen-linux-open-source
AMD going a bit more open source with their drivers.

 

Their recent APUs have been decent. True, not top CPU performers, but the overall package for Kaveri (desktop) and Carrizo (mobile) are decent. Provided you have a use for the better-for-equivalent-price-compared-to-Intel integrated graphics, at least.

The real problem of course is single-threaded performance, and Intel has done a very good job on capitalizing on their lead in that area. Bulldozer (the architecture, all generations) was too weak in single-threaded performance in general, but AMD also bet on many-core CPUs a few years too early. With each year the case for a highly-multithreaded CPU becomes slightly better as more programs take advantage of them, but AMD was dependent on it while too many programs were still single-threaded.

It's for that reason that I'm actually considering switching to Piledriver in my desktop. I use single-threaded-limited programs that tax modern CPUs a fair amount less than I did when I bought Sandy Bridge, and multi-threaded programs more. And since I'm not sporting a GTX 980 or Radeon Fury, I don't need a Core i7 4790K to push a single-threaded graphics dispatch thread. At $110 for an unlocked octo-core, Vishera is actually a pretty good value if you don't have a real need for that peak single-threaded performance. Though again, the problem for AMD is that you can actually buy one of their octo-cores for $110 new, whereas you'll pay over $300 for an Intel hexa-core and closer to $1000 for an Intel octo-core.

Ultimately though, AMD needs to execute on Zen and 14 nm. If they can fix single-threaded performance, it will be really interesting to see what they can do. Intel's including integrated graphics on all their i5/i7 GPUs, and a lot of their new silicon is going towards that. If AMD releases a new FX with good single-threaded performance and without integrated graphics, but more cores instead... they might not be absent any more.

If not, well... I'd prefer not to see that, since it would be bad for competition (and thus likely innovation) on both the CPU and GPU fronts. And that might happen, since AMD hasn't always been firing on all cylinders with execution, though this time they certainly have a fire lit under them to motivate them to deliver.

And fixing Linux drivers would also be great. There's no denying the fact that currently, unless you are a hardcore open-source-drivers-only user, nVIDIA graphics are the way to go on Linux. If I were really determined to switch to Linux, I would've already swapped my 6870 for an nVIDIA card. I suspect they're doing this more for the GPGPU business than the consumer Linux graphics business, though, since there's probably a lot more money there.

 

Here's more news on AMD's Arctic Islands:
http://wccftech.com/amd-greenland-14nm-production-q2-2016/

So, we're looking at the summer release most likely for desktops, and hopefully mobile versions will follow shortly after (or hopefully at the same time - and it would be nice if mobile version come with HBM, not GDDR5x - leave the outdated memory layout in the past, time to use HBM ).

 

So you wouldn't be satisfied with a doubling of vram bandwidth? Strange I think.

Considering mobile maxwell fares well with GDDR5. A doubling in performance might be feasible with Pascal but probably unlikely which means GDDR5X is more than enough.

If using GDDR5X means we can keep to the standard memory layout and therefore the standard MXM 3.0B standard then it is a good thing.

It is doubtful there ever will be another standard. Everything will be soldered . Then again if HBM 2.0 can be adapted to MXM then I am all for it.

 

Are there any technical hurdles preventing HBM 2 from being adapted to MXM?

Also, what's the big deal in losing MXM and replacing it with a different interface that can become a new standard?
We had different MXM types coming out over the years, so if HBM 2 requires a different (new standard) interface, it wouldn't be a particularly bad thing - especially if it allows for upgrades like MXM did (as opposed to being soldered to the motherboard - however, I don't think we will have too much of a say in this, because OEM's prefer to save money, and whichever options gets them more savings, that's what they will go with - and since HBM allows for space savings, I think the cooling would need to be redesigned to accommodate the physical changes).

Also, Arctic Islands was effectively stated that it will feature similar improvements (over the current Fiji lineup - which is very close in efficiency/performance to Maxwell) like Pascal, so I don't see an issue here, but I'm not sure we will get double the performance from either. It would be great if we do, but a more realistic estimate would be 50% increase over current offerings (at best).