Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
AMD Confirms RX 480 At $199 USD, Other APU & Polaris Announcements
- Thread starter lowendlinux
- Start date
- Sort by reaction score
Maybe that is it.
But HBM was co-developed by AMD and that didn't stop NVidia going all out with HBM2 - or being in the process of.
Anyway, HBM2 has more manufacturers going, so it's not exactly the same situation I guess.
But anytime a mem tech comes along, vendors will use it regardless of who started it, AMD has been pioneering there.
It was just bothering me, that's it, not knowing for sure.
Also, for a time even after launch AMD didn't disclose the FP64 ratio. Why?
Those are details that matter but that are kept hidden for some reason.
I'm curious as to the ratio on the FirePros, is it a hardware limitation or was it capped on purpose for the gaming cards?
NVidia does it too, not attacking AMD here.
But HBM was co-developed by AMD and that didn't stop NVidia going all out with HBM2 - or being in the process of.
Anyway, HBM2 has more manufacturers going, so it's not exactly the same situation I guess.
But anytime a mem tech comes along, vendors will use it regardless of who started it, AMD has been pioneering there.
It was just bothering me, that's it, not knowing for sure.
Also, for a time even after launch AMD didn't disclose the FP64 ratio. Why?
Those are details that matter but that are kept hidden for some reason.
I'm curious as to the ratio on the FirePros, is it a hardware limitation or was it capped on purpose for the gaming cards?
NVidia does it too, not attacking AMD here.
ManuelGomes, what's wrong with two different GPUs? ~8TFLOPs from Fiji and DP 1.4 from Polaris (and 5TFLOPs). With GPUOpen and Radeon ProRender and ROCm... mostly open source, it shouldn't be that difficult to utilize them in future. That is if macOS is fully supported. But I believe that is the idea of AMD/Apple companionship, that HW/Software run as optimized as possible.
Last edited:
It depends on the hardware - certainly Nvidia cards run at PCIe 3.0 speeds on modern systems.
The various older cMP systems have restrictions, so you need to mention which cMP you have.
I have a 2009 4,1 upgraded to a 5,1 thats now a 12 core.
I also have another question i forgot to ask before. Do you or anyone else know if the 5,1 mac pro's can run two of these in crossfire on osx or windows 10?
It cannot.
"Crossfire" is AMD's multi-GPU hack. Nvidia calls their similar technology "SLI".
Nvidia cards can be flashed to run at PCIe 2.0, not entirely sure what the real-world differences are in terms of performance. For the 9xx series, the only person really offering this service is MacVidCards. No one has announced this can be done for the 10xx series yet, especially since there are currently no drivers for OS X.
Crossfire/SLI is not supported by OS X either, only Windows. OS X does have the ability to see and use both GPUs, if they're supported in the first place, but not in the traditional Crossfire/SLI sense.
If you're talking about the 480 specifically, technically the 2009/2010/2012 can power two, though the card has been seen to draw more power than is safe. There are drivers out to fix this but these only run under Windows. I'm not sure what the results would be with running this card on a Mac Pro.
The 470 might actually be a better bet, since we know it'll draw less than 150w out of the box, making it the less worrying solution. However, only Sierra looks like it *might* have support for this card, after a hack or two. And even then, it's no guarantee the performance of said driver is any good, or if this will stick around.
I'm personally more interested in seeing if the R9 Nano is supported by Sierra. The 2009/2010/2012 Mac Pro can power one of these cards (you'd need to convert 2x 6-pin to 1x 8-pin to power it), and it looks like it'd consistently outperform the 480, mainly I suspect because of the use of HBM. You still wouldn't get boot screens for this card, and you'd be purely at the mercy of Apple for OS X driver support, but it'd be a good card to get.
No. Nothing is displayed until the GPU is initialized. If you don't have a Mac EFI in your graphics card, but there are drivers for your card, then the card is not initialized until after the OS boots up and loads the GPU drivers.
You don't necessarily need boot screens to choose the OS though. You could just load whichever OS loads up, then switch the OS there, and reboot. This should work for Windows and OS X anyway... I don't know about Linux.
Zarni, there's nothing wrong, specially with all this support for different cards, but honestly, do you really see Apple using 2 different GPUs on the nMP? I don't.
That would ruin the balance of the thing.
And not just that, having different GPUs with different power draws would require some thermal balancing.
And Polaris is already above the GPU power limit for the nMP. They could downclock it a lot to save power for the Nano, or downclock the Nano also a lot to be within the total GPU power budget, but then the higher crunching power is almost gone. Maybe the 480 didn't need to be that fast, just driving the display, but then you'd have people moaning how slow it is.
I really don't think they'll do it, neither use 28nm cards anymore. Just due to thermal/power restrictions.
That would ruin the balance of the thing.
And not just that, having different GPUs with different power draws would require some thermal balancing.
And Polaris is already above the GPU power limit for the nMP. They could downclock it a lot to save power for the Nano, or downclock the Nano also a lot to be within the total GPU power budget, but then the higher crunching power is almost gone. Maybe the 480 didn't need to be that fast, just driving the display, but then you'd have people moaning how slow it is.
I really don't think they'll do it, neither use 28nm cards anymore. Just due to thermal/power restrictions.
You can down volt and power gate the GPU. RX 480 is not that far from 130W thermal target, so it might be possible maintaining the performance of the GPU with lower voltages on the GPU.
I think there are two possibilities for lineup of GPUs:
Two base versions based on Polaris 10 with GDDR5: 2048 GCN cores, 4 GB of RAM, 2304 GCN cores, 8 GB of RAM, as first two tiers, and highest tier based on Vega 10 with HBM2 memory(16 GB?) and 4096(3072?) GCN cores.
Second option is that all of the Mac Pro GPUs will be based on full and cut down versions of Vega 10.
I think there are two possibilities for lineup of GPUs:
Two base versions based on Polaris 10 with GDDR5: 2048 GCN cores, 4 GB of RAM, 2304 GCN cores, 8 GB of RAM, as first two tiers, and highest tier based on Vega 10 with HBM2 memory(16 GB?) and 4096(3072?) GCN cores.
Second option is that all of the Mac Pro GPUs will be based on full and cut down versions of Vega 10.
If it comes out now, I'd say the GPUs are based on existing 460/470/480 lineup.
460 is in fact the successor to the D300 really, so it would make sense. P11 with 2GB, which is kinda low for today's standards, specially in a Pro card. Maybe that would go for the full 4GB instead.
470 is a nice mid range card, my choice probably. they can even bump it to the full 8Gbps mem. Either with 4GB or even 8GB, which is what I hope for.
And finally the 480 full fat with either 8GB or even 16GB, which would be nice. And that should be GDDR5X to make a difference but I won't go into that yet again
So, it's either 4607470/480 with 2/4/8Gb or 4/8/16GB, the latter being the preferred solution of course.
Vega is still too far off, and I don't believe in "later" coming cards.
If they wait for Vega, they might just as well wait for Skylake-W and that would be a dream machine, but still almost 2 years in the future - bummer!!
460 is in fact the successor to the D300 really, so it would make sense. P11 with 2GB, which is kinda low for today's standards, specially in a Pro card. Maybe that would go for the full 4GB instead.
470 is a nice mid range card, my choice probably. they can even bump it to the full 8Gbps mem. Either with 4GB or even 8GB, which is what I hope for.
And finally the 480 full fat with either 8GB or even 16GB, which would be nice. And that should be GDDR5X to make a difference but I won't go into that yet again
So, it's either 4607470/480 with 2/4/8Gb or 4/8/16GB, the latter being the preferred solution of course.
Vega is still too far off, and I don't believe in "later" coming cards.
If they wait for Vega, they might just as well wait for Skylake-W and that would be a dream machine, but still almost 2 years in the future - bummer!!
It will be a shame if they wait until Vega. That means we won't see a new mac pro until next year when the current model has been shipping for 3+ years. They need to just get a model out now with Polaris 10/11 configurations, broadwell-ep and thunderbolt 3 and then they can do the Skylake/Vega update next year (or the year after that...).
Its not even clear if Vega will fit in the current mac pro form factor. Right now Polaris 10 is a ~155 W card and assuming Vega is bigger and consumes more power it might be overkill for the 125 W thermal envelope the mac pro is currently limited to.
Its not even clear if Vega will fit in the current mac pro form factor. Right now Polaris 10 is a ~155 W card and assuming Vega is bigger and consumes more power it might be overkill for the 125 W thermal envelope the mac pro is currently limited to.
It's hard to tell. Who can predict that Apple rebrand a 7970 as D700 and put that into the 6,1? The standard 7970 6G card is way way outside the 6,1's GPU thermal envelope.
I googled a bit and found quite a few report about down volt the RX480 to 1.05V (stock 1.1V), which makes the card save >10W and run cooler and perform better (no more thermal throttling, and stay a max boost clock). Obviously this is just a one voltage fit all strategy. By running more test, and fine tune the voltage, some one suggested that can save up to 30W.
Of course, you can always down clock a card to fit it in a smaller thermal envelope. The problem is there is diminishing returns on this sort of thing. The 7970 was a ~180 W card that was downclocked to fit in a 125-150 W envelope. But if you wanted to put a bigger and hotter GPU in like a Fury derivative it doesn't make sense to take a 250 W card and squeeze it in the same envelope.
Apple is known to do this sort of thing. They typically ask for the best binned cards from AMD. This allows them to shave a few watts off the standard retail cards.
It would seem to make more sense for the WX 7100 (etc.) "workstation" cards to be used over the RX 480 (etc.) in Apple's "pro" machines, wouldn't it?
The 7100 would have 5 T-flops and 8 GB each vs. 3.5 T-flops and 6 GB each for the D700. Not a super update but not so bad either.
(Plus, how would they justify $1000 for an upgrade if they were using 480's?)
The 7100 would have 5 T-flops and 8 GB each vs. 3.5 T-flops and 6 GB each for the D700. Not a super update but not so bad either.
(Plus, how would they justify $1000 for an upgrade if they were using 480's?)
The WX7100 has very little difference in hardware compared to the RX 480. VRAM is even the same. Workstation cards are more expensive some of the time because of better driver support for things like 3d drawing. The GPU itself is often the same. Sometimes these kind of cards will offer ECC VRAM or lots of VRAM but neither of these are the case for the WX 7100.
When it comes to the DX00 series the "professional" label is more about branding than it is the feature set. The GPUs themselves were more similar in features to AMD's consumer line than its professional line. The D700 though did have lots of VRAM for its time.
May be you are right, but that doesn't looks like what Apple did in the pass. AFAIK, all GT120, 4870, 5770, 5870 are not workstation card.
The D700 is also just a downclocked 7970 6G card. Apple can call the RX480 as DX700 (or whatever they want), and put 12G VRAM on it, and marketing it as a workstation card. TBH, at this moment, I still can't see how D700 is a true workstation card. It's failure rate is high (compare to the true workstation card), it doesn't has ECC VRAM, it doesn't has specific driver, it doesn't has specific support from Apple (the same Apple care as all other Apple hardware), it doesn't even has a unique device ID.
May be the card work really well in 6,1 and it's super fast in real world OSX professional apps, however, I just can't see how it's a true workstation card.
If Apple can make the 7970 a workstation GPU in 6,1. They can do exactly the same thing and make RX480 a workstation on 7,1. And sell them as a $1000 as an optional upgrade. If a rebranded 7970 can works so well in the 6,1 as some owners report, and worth the money (because if perform exactly like a true workstation card). Then the rebranded RX480 can also work very well in the 7,1 and worth that $1000 (if Apple do it right).
Right. Much of that is why the WX-7100 would make more sense at this point. The only down-side from Apple's point of view might be the infamous "Not Invented Here" syndrome. But they will call them something else and service them like everything else just exactly as if it was a pure Apple product.
That means that they'd replace them, and you'd be back the next week because it didn't fix the problem ?
To Apple, there is no difference between the RX 480 and the WX-7100. The cards are identical except for some minor clock speed adjustments. Both feature GDDR5 without ECC. AMD charges extra for the WX-7100 for access to its professional drivers, which Apple obviously doesn't care about since they are for windows. Apple can call the card whatever they want, but it will have the same features and performance as these cards.
Apple used a HD 7970 in Mac Pro. According to tech power up numbers this GPU have had 925 MHz core clock, and consumed under load 163W of power. Apple declocked it to 850 MHz and fitted in 130W thermal envelope.
If Apple would decide to put Polaris 10 on standard core clocks of 1120 MHz, and standard voltage at:
0.96V it will draw much, much less power. Let it run even on 1175 MHz, and you still have way below 130W power target at 1V.
Secondly, every of the clues so far points to small Vega - Vega 10 - being released this year.
Tile-Based Rendering is the secret of efficiency and performance of Maxwell and Pascal GPUs. Watch this:
And Vega 10 according to "whispers" floating around the industry will have 2 new features architecture wise: new scheduler, and new rasterizer.
If Apple would decide to put Polaris 10 on standard core clocks of 1120 MHz, and standard voltage at:
0.96V it will draw much, much less power. Let it run even on 1175 MHz, and you still have way below 130W power target at 1V.
Secondly, every of the clues so far points to small Vega - Vega 10 - being released this year.
Tile-Based Rendering is the secret of efficiency and performance of Maxwell and Pascal GPUs. Watch this:
Last edited:
Yeah, but you can turn these voltages too low or you won't get enough chips to qualify to run at these voltages. You need a bit of headroom to maintain stability.
AMD themselves have said its coming in 2017. This is the type of thing public companies don't lie about since they are accountable to their stockholders. There were "whispers" that Polaris was coming early this year, had 2560+ cores, was a 125 W card and could match Fury. None of these were correct.