AMD announces Radeon M300 series notebook video cards

But

 

I know... I know.

 

Frankly, those who are set on buying AMD's Fiji likely won't care what nVidia has to offer and vice versa, and price isn't likely to be too much of a deterrent. Really the only thing a lower price will accomplish is it might give fence buyers some slight pause, but even then it's not a guaranteed sell.

I think what AMD is doing is instead of trying to persuade the fence crowd (likely a minority population to begin with anyway), they just said "fack it, we're gonna release this awesome card that will be faster than Titan X, and it will be at this price. Pay or GTFO." Margin vs volume I suppose.

 

That might be true, and the fence crowd might be small, but see if nVidia were to release a 6GB vRAM version of the Titan X and slap a 980Ti on it and say $700 a pop, that's very likely to get recommended over the AMD solution. If the 980Ti is weaker though (like 780 vs 780Ti) and the R9 390 is a $550-600 GPU we'd be in the same boat.

 

980 Ti is a 2816 core cut down GM200 that's going to come in at the same 700 price the 780 Ti came in at and follow the same trend of increasing prices for custom cooling solutions and out of the box overclocks.

It looks like both AMD and nVidia are going to price gouge.

 

See I'm not quite willing to accept that AMD is completely delusional just yet. So the way I see it, IF Fiji actually does launch at $849, there's no doubt in my mind it will be faster than Titan X, and 980 Ti by extension. So it all boils down to whether that 4GB HBM is such magic and so bandwidth that "4GB means 6GB or even 8GB".

Basically I find it hard to believe that Fiji just matching 980 Ti and with only 4GB HBM will go for $849, halo card or not.

 

Wait... 4GB HBM is all? I mean... I think 4GB is still doing pretty well and all right now, but not for a card like THAT. Even if there was such a huge memory bandwidth that it could swap assets and texture data in and out like a party girl changes clothes, you quite likely would get bottlenecked by storage and/or RAM I/O speeds far before you truly see the effects of only keeping what's "necessary" in vRAM.

 

It's almost confirmed by this point that the card will have 4GB HBM, at least at launch.

And yes another reason to be skeptical of the $849 price.

 

Seriously no new mobile cards? FFS AMD.....

I really really really hope they take R&D seriously now and try to make become performance oriented again instead of budget oriented. In fact, they should completely shutter their CPU division and completely pour everything into graphics and try to salvage that because their processors are going absolutely nowhere with comparison to Intel

 

They're banking on Zen. I don't know if anything can save them in mobile though.

 

I'm all for calling amds next card the party girl

 

8GB versions will come later. Maybe they try to save more money now.

 

And wait for HBM yields to improve

 

4GB is the current limit for HBM isn't it? I thought Gen1 was 4GB, Gen2 will be 8GB.

 

I think the limit is how many dies there can be in a stack, but there's no physical limit on the number of stacks you can put on the interposer. At least that's my (admittedly limited) understanding.

 

They're supposedly using multi-layer 1GB chips on a dual-link interposer to install 8GB. HB2 will eliminate the need for dual-link interposing.

 

As for bandwidth, the titan x has 3xx GBps, the 4GB fiji will have over 500 GBps. I posted a couple articles that discussed the bit interface and why 6 and 8gb cards came out. The vram doesn't need to be that large, but you need that much to get the necessary bandwidth or it becomes the bottleneck.

 

But memory bandwidth past ~160GB/s is more of an incremental increase (sort of like raising CPU speed without a heavy CPU bottleneck) so as far as I can see, HBM is, while a good direction to head toward for GPU memory on the whole, is nowhere near beneficial enough to gamers and multitaskers to justify a top-end card having only 4GB today in a market that's overhyped on 4K displays. Not that high AA + 1080p doesn't usually use more than no AA at 4K, but these GPUs are strong enough that one or two of them should chew through games at 4K, and that includes turning up the AA. So yeah, 4GB shouldn't be a thing for it. I wouldn't mind if the R9 380 and R9 380x cards had 4GB and were cheaper and competed with the 970/980 cards, though.

 

What you didn't address in your statement is how more is needed if the bandwidth is no longer an issue. Without stating there is another bottleneck, you are whining you want more ram without giving a technical reason why you need more. For that reason, wait for the benchmarks. If it shows that the titan x is slower, but you want more ram, then buy the titan x and have less performance. It is that simple!

 

You just made absolutely no sense, dude.

I was saying that cards like the Titan X and R9 390X which are powerful enough to run games at 4K (especially in multi-GPU configs) with all the settings AND anti-aliasing turned up, might start needing more than 4GB in higher resolution configs. I'm not saying we need 8GB or 12GB, but 6GB should be a minimum right now for cards in that power class. It'd be like designing a 2GB GTX 980 as the standard for 1080p gaming. Understand?

Remember, multisample-type AA increases the vRAM usage based on the original resolution you're using. So at 4K, they'd use FAR more extra vRAM than they would at 1080p.

 

I doubt it is smart to use any AA for 4K. Anyone who want it should wait at least 2 yers or better 3, otherwise he will just make pain for himself with SLI lottery.

 

Well, with pre-Titan X tech I'd have agreed with you. But SLI 980s do a better job of Titan Black SLI at 4K, generally SLI 980s or crossfire R9 290X cards were fine for maxing or near-maxing 4K without AA. But with the ~30% boost of Titan X (and assumed R9 390X), AA is no longer a problem. And since 4K 120Hz screens don't exist yet and 60fps is all you're looking for, why not use AA? As I said... a card in the power tier of the R9 390X would be quite hampered by 4GB of vRAM should people wish to take advantage of multiple of them at very high resolutions, due to AA having a scaling hit to vRAM usage depending on resolution.

 

Here is what I've understood from the chief tech officer's explanation from AMD (and their rational for using 4gb for the first Fiji release), the reason they added more ram was not ram utilization, it was for bandwidth. Because of the configuration, you add more ram chips, more connections between the gpu and ram since the bits were limited, but the amount of ram was not. That way you increase bandwidth but do not use all the ram. You have a valid question whether the 4GB of ram will be the choke point now. BUT that doesn't have any relation to whether it is the BEST card on the market on release. Let's say you are right, titan x still would be less amd no one would pay more for less performance over ram if the additional ram of titan x gives no advantage. They may over driver support, etc. But if it performs better (but less than what we think it would with more ram) it DOES NOT MATTER!!! It would still be king! In addition, the reduction in heat translates to smaller form factor AND potentially higher clocks, ESPECIALLY ON WATER COOLED SYSTEMS! Titan has to cool 12GB of ram, 390X (our as other articles suggest it will be its own line like titan breaks away from numbering) has to cool 4GB. That is a HUGE heat saving. The proof is in the pudding, synthetic and game play frame rate.

 

Sidestepping this VRAM debate, I do have some details of the R9 m290 Apple announced.
Clock speed: 975 MHz
Compute units: 16 (That would mean it has a shader count of 1024)

 

Higher clock than the R7 265? That would make it faster than R9 M290X. I call BS.

 

Well, the clock speed could be courtesy of Apple's overclock. Apple did do the same with the R9 m295x, clocking it at 850 MHz as opposed to the AMD-stock 723 MHz.
Wouldn't be too far-fetched if thr m290x was overclocked as well.

 

No what I'm saying is R9 M290 would be clocked higher than R7 265, the desktop GPU it's based off of. That almost never happens when making a mobile GPU because it's a lot more efficient to drop clocks and voltage while keeping the core intact than it is to disable parts of the core and keep clocks and voltage. That's why R7 265 (1024 shaders @ 925 MHz) has the same 150W TDP as R9 270 (1280 shaders @ 925 MHz). A 975 MHz R9 M290 would likely have a higher TDP than R9 M290X unless it's on a more efficient process. Considering R7 265 at a lower clock is already a 150W GPU, I don't really see how a MacBook/iMac can cool the R9 M290.

 

Before anyone else says more random things that might not be true I would like to point out a couple of things here.

1) AMD has not confirmed how much memory the flagship Fiji card will have, the whole 4GB is the limit of HBM 1 is nothing but speculation. In fact 8GB is possible for HBM 1, although the method of implementing this is more expensive. All AMD have said is that they put enough VRAM for the GPU, so that doesn't mean they hinted at 4GB, 8GB, or both.

2) Even if it will be only 4GB to keep the costs down it doesn't mean it will be similar to current 4GB cards.
4GB Hawaii < 4GB Tonga < 4GB Fiji
The Delta Color Compression of Tonga effectively gives it on average 40% extra bandwidth and frame buffer, so basically because of compression it can hold ~ 5.6GB in 4GB space. Now what AMD also said recently is that GDDR5 memory controllers are quite inefficient and don't use the frame buffer properly and that thanks to HBM, it's logic layer, and the interposer connecting the memory to the GPU directly Fiji is much more simple and efficient when it comes to the memory controller so effectively there is less space "wasted" in memory. They didn't give any indication of how much this will improve things but I guess we will have to wait for an in depth review to test it.

 

Oh...
But then again, it's Apple. They don't make the wisest of decisions when it comes to GPUs and cooling them in the iMac (not sure about the Macbook Pros, though).

 

MBP wouldn't stand a chance, that's why they've always had low mid-range dGPUs at best, e.g. 6770M, GT 650M/750M, R9 M370X.

 

Nvm it seems Apple does OC all their mobile GPUs, so a 975 MHz R9 M290 is plausible.



But we're still no closer to actually new AMD silicon on the mobile side...

 

http://wccftech.com/upcoming-amd-radeon-flagship-called-fury/ (re-cited source)
http://www.expreview.com/40812.html?utm_source=tuicool (originl source)
Yes, it is unrelated to mobile GPUs. But I do have one question about this. Would AMD bring over the new Fury line (or naming scheme, doesn't have to be based on Fiji) to the mobile market?
Well as it stands, there are three R9 mobile GPUs (all 290s) left to either be discarded or rebranded. So unless there's a m395 and m395x, everything seems to be rebranded once again, unless Fury is brought to the mobile lineup.

 

Don't believe we'll see a Fury mobile GPU... Fury is essentially a Titan competitor. There aren't any Titan mobile chips. The mobility side there is only "refreshes" between 2015 and 2016 until AMD attempts bring out their new architectures for Arctic Isles GPUs and Zen CPUs.

 

They could possibly introduce a cut down version of Fury with HBM that would fit into laptop thermal constraints.
Titan is merely a full core version of Maxwell architecture if I'm not mistaken.
Hence, if AMD already has lower end mobile solutions which differ in core count and thermal aspects, it should be possible to make a cut down Fiji HBM mobile gpu.
HBM alone would already cut power consumption by a good margin, and keeping the core count to something manageable in combination with new GCN architecture could be good... then again, the new GCN architecture may not be on par in terms of efficiency like Maxwell.

 

Value for money. The Razer has better specs and beats the MBP on everything except battery life and operating system.

I love OSX but I am getting impatient with unfulfilled promises. Most notably how bad icloud works in my use case. One drive is much better and I get a TB worth every year with O365.

I ordered a QHD Razer from MS Canada with 970m for $2160 CDN plus tax. That's what, $1600 USD?

The comparable mbp is over $3,000 CDN.

The only ties to apple I have are music, movies and photos. Music is DRM free and I've uploaded my itunes collection to google. Movies can be DRM stripped, then I can play them in non apple devices. And photos are something I am finally using sensible software such as lightroom to manage, making uploading to multiple cloud services easy.

 

There are no MXM specifications available to build a mobile GPU with HBM.

Pascal and AMD cards with HBM will either come soldered on the motherboard or MXM Sig (the company that makes the specifications for the MXM card) will need to make a new revision or brand new MXM type.
2012 was their last update. Don`t think its 100% certain we will get Pascal/R9 400 series in MXM.

We may not have the momentum from the industry to make it happen anymore. OEMs probably want soldered components anyway... :/

 

Wrong thread?

 

So long as the mounts and connector positioning are all kept to and the heatsink matches the core then it makes little difference really. The spare area can be used for power circuitry.

 

Seems like AMD might have a fight on their hands. In a totally unprecedented move, the 980Ti's MSRP is $650.
http://www.pcgamer.com/nvidia-geforce-gtx-980-ti-review/

If the R9 390 is at that price or more expensive and has less vRAM (HBM or not) and is not decidedly stronger, guess what the next best selling card might become. Even *I* have to admit that $650 ain't bad for what it brings you, even if the Titan X is overpriced.

 

Yeah, I was taken aback at that price. It's actually not bad. Hopefully AMD can compete for less.

 

It's priced so well I wonder if there's something broken in it like in the 970... I'm so skeptical about nVidia right now.

 

Definitely not unprecedented, more like a repeat of history. Original Titan launched at $1000, 780 with 2 SMXs disabled and half the VRAM launched a few months later at $650. Titan X launched at $1000, 980 Ti with 2 SMMs disabled and half the VRM launched a few months later at $650.

If anything, 980 Ti's "competitive" pricing shows that Nvidia knows AMD has a great GPU on their hands with Fiji, so Nvidia is striking first.

 

I suppose... maybe when the 985Ti (6GB Titan X) comes out it'll be $700 again like the 780Ti was haha.

But still... AMD is reeling from that one and they haven't even put out their GPU yet.

 

It is pretty amazing how close the performance is... I feel sorry for the people who got suckered into the Titan X

 

@D2 Ultima Actually the $650 price just makes me even more confident Fiji will rip them a new one, so they're rushing the 980 Ti to market to cash in on day 1 impulse buyers.

OR they know Fiji is a flop, and the 980 Ti also has some unexpected "special features" XD

 

Exactly. Or 1080, which seems more likely.

Nah. If rumors of Fiji's performance are true, I think AMD's gotta be feeling pretty good about its position right now. It could potentially upset both 980 Ti and Titan X with 390 and 390X respectively. Nvidia made the first move, which makes it all the easier for AMD to one-up them.

 

R9 390 MSRP $550
R9 390X MSRP $700-850
R9 390>/=GTX 980 Ti
R9 390X>Titan X

Come on AMD. Give Nvidia the sucker punch they deserve.

 

I actually wonder if AMD is deliberately taking their sweet time in order to force nVidia to show their hand first. Because you know, one-upsmanship only works if you're late to the party!

 

I totally gave my hopes after hearing AMD's HBM (if there will be any) mobile flagship GPU won't be able to compatible with MXM3.0b due to its high bandwitdth. So either rebrand or brand new GPU, it will be no good for me. It is sad, because I was really ready to give a chance to AMD. Stuck with NVIDIA then...

 

Cloudfire is guessing... Its not like the manufacturers currently all follow MXM spec anyway. Why do you think Clevo makes the only MXM cards that actually work in all MXM machines? Because they are the only ones who follow the spec!