AMD announces Radeon M300 series notebook video cards

I didn't say they don't throttle. I was asking the question.

 

3.6GHz.
4980HQ boosts like
3.6GHz 3-4 core
3.8GHz 2 core
4GHz 1 core
Or... it's supposed to do that. That's what 4980HQ owners have reported, even though the wiki you linked to me before says it's a 3.8GHz boost for 3-4

Also, no it cannot handle that.

Not even close. Windows and OS X are GUIs that run instruction sets. If instruction sets X are run in order X (say... a benchmark?) on windows and then run on a mac, they're gonna use the CPU in the same way.

 

So, as for the flagship mobile, there has been no details (that I've found) leaked as to performance (including the speculation on HBM). Here are what I see as the most likely possibilities:
1) they use a 28nm design based on an abandoned 20nm design
2) they use a 28nm version of the 14nm design to tease next year's processors (not likely)
3) they use one of the abandoned designs from skus that never showed up in production (likely cut to focus money on R&D on the die shrink for next year)
4) they use a cut down Fiji design (cooler than Hawaii, and most likely choice if HBM appears on the card)

As for HBM, I have not heard of them using it outside of the flagship, yet. This is because it is still quite expensive. It will be a shame of them to squander this opportunity, as they likely have a one year exclusivity on HBM due to helping develop it.

Meanwhile, AMD has pulled back on this years releases due to 20nm wafers becoming scarce and to focus the money on major releases for next year. We have to remember that Lisa Su only took the reigns at the end of 2014. Since then, she has fired top people at the company and hired talent from across the industry. She pulled projects and tech off of the shelves to be included in new products (which take time to design, incorporate, and produce). She then announced the shift from focusing on low end APUs to maintaining (and trying to gain more) custom APUs, while trying to reclaim a space in the high end (and mid-range) computing. This is why a lot of previous products got scuttled. She has poured a lot of R&D directly into next years releases, which is why this year is lackluster.

As for serendipity, IBM sold their patents to GloFo. This gave many useful techs up for AMD to use as a strategic partner and lured Samsung into giving a cross license on its more developed 14nm node (which will be used in Zen). Some of this, I wouldn't doubt, has found its way into redesigning the GPUs which could explain the claims on heat reduction better than just a redesign (same goes for the guidance on a 95W TDP Zen processor since we all know how hot the 9000 series 8 cores were). To, in mid-design, switch from the 14nm GloFo transistor to Samsung transistor would have taken a chunk of cash. Considering they are reporting less R&D expenditures than Intel, many investors are scared that they are blowing smoke, but many of the changes were either their old tech or already developed by IBM or Samsung. This should give great hope on the integration of these products, although it leaves to question whether the 14nm cross-license included Samsung's 10nm. The reason this is important is GloFo was developing the 14nm to scale directly to 10nm (just like TSMC and Samsung). If they do not have the rights to 10nm, there would have to be a larger redesign for Zen+ to switch back to GloFo's 10nm.

Because of the above, in conjunction with SMT regularly (across the industry) providing a 20% gain and each die shrink regularly giving 10% (except for Intel's jump to 14nm), AMD's claim of 40% seems plausible on the surface. This also suggests that this year's mobile offerings (like AMD's majority of offerings this year) will be ****e. For that reason, I'm not expecting anything amazing out of them except for the desktop flagship (which will become king!).

Edit: Sorry to go off topic, in part, but I did include some info pertinent to CPU development due to some comments in this thread.

 

if you read this already, I apologize, but this has some info on GPU as well, like 4 GB max RAM, maybe offset with compression etc.
http://www.anandtech.com/show/9266/amd-hbm-deep-dive

 

It suggests max ram. But Macri directly stated it is not limited to 4GB. Further, look at the bandwidth per chip stack (128GBps), Total bandwidth of 512GBps... The Titan comes in at a total bandwidth of 336GBps. Also, other articles suggest the base Fiji will come with 4GB, while they may release an 8GB version as well.

Macri also discussed the lower heat due to less ram, with the above stated benefit.

 

Well with the APUs I'm not suprised AMD is just re-branding there GCN 1.0 for mobile. I believe the APUs used in the 7800K series actually supass the lower end mobile GPUs

 

So, the first true new 300 series AMD GPU is finally here:



Repi, rendering guru from Dice / Frostbite engine technical director, just got it from AMD:

https://twitter.com/repi/status/601739457960763392

Looks like those earlier leaks:

http://www.pcper.com/news/Graphics-...ow-Small-Form-Factor-Water-Cooler-Integration

 

R9 M370X = 640 shaders
Its a rebrand.

https://compubench.com/device.jsp?benchmark=compu20&did=26119942&os=OS X&api=cl&hwtype=dGPU&hwname=AMD Radeon R9 M370X

max_compute_units 10
1 compute unit = 64 shaders

R9 M265X up to R9 M275X all have 640 shaders.

 

Even so, that's not the mobile flagship.

Edit: many of the m300 series already announced seam to be rebrands (maybe with minor tweaks). I can't remember if the m370x was among those already listed in the OEM sku document that has been circulated. If not, I apologize for the tone and thank you for the additional info.

Edit 2: Here is an article I know of talking about the rebadges/rebranding. I cannot find the one that had a nice table simplifying getting the info.

http://online.bs/2015/05/18/update-amd-announces-radeon-m300-series-notebook-video-cards/
http://www.fudzilla.com/news/graphics/37721-amd-also-announces-new-radeon-m300-series-for-notebooks

It seems according to fudzilla that the supposed flagship is the m375 (no x). If so, it seems it may be a rebrand. This is why I'm waiting until E3 when it is all on the table. Then the model numbers are known and we'll see what model is the flagship.

 

Yeah, back when AMD announced the OEM 300 and m300 cards, only the R9 m375 was included for the R9 m300 series. The m370x was not included.
Currently, only Lenovo and Apple are using the R9 m370 series GPU. I haven't seen Toshiba update their high-end Satellite laptops with the m365x yet.

 

R9 M370X have not been announced by AMD. Just bragged by Apple as "80% faster bla bla bla".

More unannounced cards revealed through drivers this month:

AMD6646.1 = 'AMD Radeon (TM) R9 M385
AMD6646.1 = "AMD Radeon R9 M280X

AMD6647.1 = 'AMD Radeon (TM) R9 M380
AMD6647.1 = "AMD Radeon R9 M270X

 

Just FYI, Windows and OSX are both operating systems, which may or may not have a GUI (OSX can be run with only a CLI, iirc). An instruction set is the set of commands that a processor is capable of understanding (e.g. ARM, x86), not a set of instructions that make up a program ( so it doesn't really make sense to say an OS runs instruction sets). Also, in modern out-of-order CPUs there's no guarantee that, even if you run programs with the exact same instructions in the exact same order (which is unlikely to begin with since Windows and OSX use different APIs for most things), the processor will execute them in the same way.

But yes, boost frequencies should be OS agnostic. From what I've seen, 3/4 core boost frequencies are typically 200mhz under the 1 core boost frequency, so 3.8ghz for 3-4 cores sounds right (and this is also what notebookcheck reports). I don't see any reason why a 4980HQ in a Macbook can't hit 3.8ghz on 4 cores for a short period (though obviously it would be unable to maintain 3.8ghz for long periods due to lack of thermal headroom).

 

Fair enough, however the instruction sets are in fact what cause the load to be what it is. Even if you run it out of order, the instructions being run are the same. It shouldn't make any matter of efficiency.

As for the boost frequencies, I always thought that the 4980HQ was 3.8GHz on 4 cores, but as far as what I had come to find out from actual owners, it was 3.6GHz for 4 cores. Unless I found something exceedingly wrong somewhere in my research. If someone has it and can show me a screenshot from XTU with the default settings I'd believe it, but as far as I know it goes as I listed it above.

Finally, the TDP limits, the software the mac uses to throttle the CPU before it gets to loud-fan-range or overheat-range and finally the lack of heat mean that anything even halfway-heavy in load wouldn't work. Maybe for a few seconds?

 

Not rocket science. Just look at the 2nd and 3rd number in every R9 mobile GPU name.

6x or 7x: Cape Verde (640 shaders)
8x: Bonaire (768-896 shaders) example
90: Pitcairn (1280 shaders)
95: Tonga (2048 shaders)

http://www.amd.com/en-us/products/graphics/notebook/r9-m200

 

Didn't know AMD's stock m280x was so crippled, it ran at 110 MHz as opposed to 1000 MHz.
But seriously, the m280x in the ASUS laptop doesn't match the official specs. In fact, they match the specs of this m270x. Just something I now noticed.

 

AMD's official specs list R9 M270X as a 640-shader Cape Verde. M280X core clock typo aside, I'd sooner trust AMD than TPU, who consistently prints a lot of mistakes. It doesn't make sense from a logical POV either. Why would M270X have more shaders than M275? Desktop 260X certainly doesn't have more shaders than 265.

 

Well, according to Cloudfire and AMD (through the drivers), there's a m380 that is a rebrand of the m270x, and that m270x is different from the usual one.
AMD6646.1 = "AMD Radeon R9 M280X"
AMD6647.1 = "AMD Radeon R9 M270X"
...
AMD6820.1 = "AMD Radeon R9 M275X"
AMD6821.1 = "AMD Radeon R9 M270X"
AMD6822.1 = "AMD Radeon E8860"
AMD6823.1 = "AMD Radeon R9 M265X"
So to answer your question, it's because AMD labelled it so.
@Cloudfire, question. Where does one find these identifiers?

 

Looks like AMD's mobile GPU numbering is even more of a mess than I thought. Also judging from those numbers it seems that 66XX GPUs are Bonaire based and 68XX variants are Cape Verde. So since 6646 is full Bonaire with 14 CUs, the 6647 must be 12 CUs and 768 SPs.

 

Wow amd hasn't released a new mobile GPU since the 7970m... how pathetic

 

The Bonaire-based ones are newer. Tonga (M295X) is even more recent.

 

Forgive my ignorance, but what would stop AMD from taking the Fiji XT die, downclocking it, undervolting it, and sticking it in a notebook? Would you just lose so much performance that there would be no point in doing so?

 

Same reasons we don't have Titan X in a notebook.

BTW did you miss this? Reference Fiji XT is water cooled.

 

No, I didn't miss it. I was away for a couple weeks (hence my inactivity), but I was still keeping a close eye on AMD.
But I suspect that the R9 390 will be a downclocked, air-cooled version of the 390X, so it still leaves the door a crack open for possibilities.

 

Yeah, let's have a repeat of GTX 480M, but worse.

 

Funds! But yeah it's different with GM200 because of the 384 bit bus, Fiji on the other hand requires HBM so it's not limited to 256bit, but then again you need a new way to implement it into notebooks, current MXM won't work, maybe placing it on the motherboard itself would but at this point we don't know. The die sie wouldn't really be an issue though and neither would heat, as HBM is much more efficient than GDDR5 and lower clocked Fiji would be faster and cooler than Tonga.

 

I doubt Fiji XT will run as hot as Fermi, at least not when downclocked. I feel like Nvidia got cocky with the 480M.

 

Well Hawaii certainly did, and they didn't even need a water cooler that time, at least not for the single GPU Hawaii cards

 

I'm sure it would be possible to at least get Tonga-like temps with a custom performance-limited Fiji XT. Sure, Tonga runs hot, but it wasn't nearly the disaster that the 480M was.
Now whether AMD would actually do that or not remains to be seen.

 

I think this hasn't been stated yet in these forums so I will put it here.
I don't know what the R9 M300 series flagship will be, but according to WhyCry from Videocardz, an M300 GPU based on Tonga XT has 90% of the performance of 980M. He claims that he saw benchmarks (in person I believe). Again not sure if that's the top card or not but that's way better than the M295X, also who knows maybe they will have some special branding for the mobile flagship just like the rumors point to for desktop Fiji.

 

Isn't the R9 m295x a Tonga XT mobile GPU? And it's performance is between the GTX 880m and GTX 970m.
That m300 Tonga GPU would have to be much cooler than the m295x for it to be that good. Even then, the GPU would have to have a greater TPD of 125w.
I have doubts those benchmarks was for a mobile GPU. Or I doubt the validity of this person's claim.

 

There is one possibility, they put HBM1 on it. Why not use it on the two flagships, desktop and mobile. If they did, it explains both the jump in performance and lower heat. As the bandwidth refers to between the memory and the gpu, it is possible. With that said, the bandwidth of the mxm may act as a different bottleneck in the system then. That would make the benchmark 90% of the 980m quite plausible.

 

They don't have to increase the TDP, with a new stepping if they managed to improve clock speeds at lower voltages it may even run cooler. Tonga is almost a year old now (it's probably older than GM204 because it's launch was pushed back), just look at what they managed to do from Trinity to Richland, especially going from A10 4600M to A10 5750M, that was probably a bigger improvement than anything Intel introduced for mobile at the same TDP since sandy bridge, and that was without any architectural improvements, just higher clocks and the ability to hold turbo much longer.

 

M295X and 480M are equally massive fails.

M295X is already an underclocked and undervolted Tonga yet still gets past 90C and throttles. Performance-wise, 970M laughs at it and 980M poops on it, yet both Maxwells have lower TDPs and run much cooler. M295X barely outperforms AMD's own smaller and more efficient M290X. So Tonga-like temps? We might as well go home and wish for 880Ms instead.

GTX 480M sat in the mid to high 80s under load and didn't throttle, but that's because of how crippled it was to fit within 100W TDP. 425 MHz core and 600 MHz memory is just sad for a flagship GPU, not to mention 30% of the GF100 core and memory interface being disabled. It's no wonder Nvidia's own 560M (238mm2 75W TDP) and AMD's 5870M (166mm2 75W TDP) matched 480M (529mm2 100W TDP) in performance. Ofc Nvidia wasn't gonna keep putting its huge expensive GF100 chips in notebooks when smaller cheaper GF106/116 and GF104/114 chips could do the same job or better. That's why 480M lived like 3 months before getting killed.

480M was based on desktop 465, a 40nm GPU that performed like the 55nm 285 while using more power. Thankfully 465 was replaced by 460, one of the best GPUs Nvidia ever made.

 

Well, if AMD has no mobile solutions based on a new architecture... then perhaps they might have slapped HBM onto M295x and branded it their top of the line for mobile?

That alone could drop the power requirements down by about 30W (from 125W, that would be about 95W, tops), increase the bandwidth quite a lot, as well as reduce the form factor (though, I'm not sure AMD necessarily thought about doing this - unless the top of the line mobile gpu won't be Tonga, but will use a new architecture and use GDDR5).

We know that most of the 300 series will be rebrands as it is... but we have no word yet on top of the line mobile gpu's.

 

I like to go from Intel to Amd and back again, but Amd away seems to disappoint me with the cpu performance for the power used, the gpu is usually good, In recent years, I have had notebooks with A4, A6, A8 and A10, and a number of the models before that going back years, back to desktops and 386 cpu`s

I would love Nvidia to get into the x86 x64 chip making, they already make decent gpu`s and their Android cpu/gpu have been quite good and it would give the other two some competition.

But with Intel owing x86 and Amd the same with x64, what do they do for royalty`s for using each others technology.

John.

 

As for that, it is usually a cross-license as they have blocking patents (can't make the improved without the base tech). With that said, some of those 90s patents are expired now. Also, the law requires FRAND licensing, fair, reasonable, and negotiated deals. This means you cannot unreasonably deny a license for a tech that is a bar to entry for an industry. Think of Samsung and the 3G patents. Everyone has to pay Samsung to be in the game, but the amount paid must be reasonable in proportion to the patent being licensed and the product being sold. Another example is MS making $5 off of every Android phone.

 

Noooooo, I was hoping to upgrade my GPU alone

 

You unpack the exe file from the AMD driver you want to look at, browse through the folders that was created til you find the INF file with the list of cards with the device ID next to them (AMDxxxx). Sometimes you have several cards listed under the same device ID in the drivers, like 6821.1/6821.2 etc, they are all the same card just separated with decimals because they have different names.
Then you compare the device ID`s and see if you can find them on other cards. If its a match, its a rebrand. Just google the "AMDxxxx.x" and you should find them pretty quick.

This is also possible with Nvidia drivers.
laptopvideo2go.com find entries in the INF drivers too of unannounced cards, like this upcoming Tesla mobile Maxwell card thats about to come: http://forums.laptopvideo2go.com/topic/31487-inf-v6019/
The difference between AMD and Nvidia drivers is that AMD only have one INF file you need to look at while Nvidia have like 10-15 different INF files which belongs to different OEMs (Gigabyte, Dell, MSI etc).

 

Eh, I'm only interested n the GPUs supported in general. But nonetheless, thanks for the help.

 

It seems the R9 M370X GPU in the new MBPs is Cape Verde...

http://www.anandtech.com/show/9276/2015-15inch-retina-macbook-pros-dgpu-r9-m370x-is-cape-verde

 

So the M390/M390X will be the only interesting product this year then.

They better slap NVIDIA into action. I want a nice juicy fully unlocked Maxwell GM204 core (or better) this year. Something tasty enough to replace my not-very-good-at-overclocking 980M.

 

AMD has nothing coming out this year which will even remotely threaten Maxwell 2 on mobile. Fiji won't work, so AMD has no silicon to make new mobile GPUs from. I bet M390X and M395X will be Pitcairn and Tonga rebadges.

 

How sad is it that the market has fallen to a level where this will probably be true.

 

Sad but not unexpected at all

 

AMD Radeon R9 M370X benchmarks:

3dmark firestrike (graphics): 2,337 points (71% of GTX 950M, 55% of GTX 960M)
3dmark 11 (graphics): 3,682 points (82% of GTX 950M, 70% of GTX 960M)

http://www.3dmark.com/3dm/7059283
http://www.3dmark.com/3dm11/9842897

 

Its COST. They can't afford to sell crippled chips at all.

 

390X is rumored to be $850. So how much would a hypothetical Fiji mobile GPU be, like $1500?

 

Speaking realistically, I would expect less than that. Perhaps $1100.

 

None of this is realistic in the first place

 

Even if 390X is stronger than Titan X, $850 is a bit too much. Ideally that card needs to not cross $700. It needs to be so good that people stop realizing the Titan X exists.