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.
2016 nMP
- Thread starter ManuelGomes
- Start date
-
- Tags
- nmp
- Sort by reaction score
- Status
Only one 8 pin connector. 6 pin provides 75 Watt, 8 pin provides 150 Watt. So no problem with a 2*6 pin to 8 pin adapter:
Last edited:
The GTX 1080 is interesting, but there are a lot of places where it vs Polaris is kind of up in the air. In pure Gigaflops the 1080 seems like it might be about 30% faster than Polaris. But 180 watt TDP is starting to get close to comfortable limits for dual cards in a Mac Pro. (Has there been a TDP released on the 1070?)
It's a good card, especially if you're looking for a single card. But I think if you're looking at a 2016 Mac Pro things are still murky. Both Nvidia and AMD seem to be targeting performance per watt over raw performance right now, but the Nvidia hype is a little overboard.
Polaris 10's TDP honestly seems a little bit too low for the nMP, which also makes me think that if Apple goes Polaris it will be overclocked. 150-160 seems like a good target, and Nvidia is a little over, and AMD is a little under. Polaris and Pascal seem pretty neck and neck in performance per watt.
Last edited:
Wonder why they went with single 8-pin instead of double 6-pin
it usualy happens when they need to save space on pcb: example R9 nano
in this case, they said they were using vapor chamber to cool the card down. probably the chamber take so much space so they needed to go with single 8-pin.
now why they are using such a big vapor chamber. is the chip that hot ?
it usualy happens when they need to save space on pcb: example R9 nano
in this case, they said they were using vapor chamber to cool the card down. probably the chamber take so much space so they needed to go with single 8-pin.
now why they are using such a big vapor chamber. is the chip that hot ?
goMac, if P10 will have 4096 GCN cores and 1000 MHz core clock, than yes, in pure compute GTX 1080 will be 10% more powerful. That core config looks a bit unrealistic for 232mm2 GPU.
GTX 1070 has 6.5 TFLOPs at 150W. Similar performance per watt level to Fury, and Fury Nano. The difference is that because of that really high core clocks GTX 1070 will retain performance much better, and... Mobile GTX 1080M will be based on GP104-200 chip, clocked at 1300 MHz. So we are looking at 5.3 TFLOPs GPU with TDP between 100 and 125W.
This is direct competitor for Polaris GPUs.
GTX 1070 has 6.5 TFLOPs at 150W. Similar performance per watt level to Fury, and Fury Nano. The difference is that because of that really high core clocks GTX 1070 will retain performance much better, and... Mobile GTX 1080M will be based on GP104-200 chip, clocked at 1300 MHz. So we are looking at 5.3 TFLOPs GPU with TDP between 100 and 125W.
This is direct competitor for Polaris GPUs.
Max temps for Maxwell: 94 C. What you expect from 1.7 GHz chips?
I would think it would work better with existing 2x8 + 2x6 power supplies for SLI.
I take with an big load of salt that leaked "Polaris" benchmarks, also for me an AMD-Mac optimistic, those benchmarks are too good to be true.
I'm considering that Fury Nano benchmarks 8.5 TFlop on 175W, vs 9 TFlop at 180W for gtx1080, assuming true AMDs premise about Polaris delivering same performance as previous generation on half power, Polaris should easy reach 9 TFlop on 100W (tcMP max gpu TDP is 110W).
Whatever everything on paper it's perfect aaaaawwwwesome etc, lets see independent test's benchmarks to have an criteria on this on factual data and not PR Labs "promised benchmarks".
9 TFLOPs from P10 at 100W? When Nvidia is not able to achieve this on 16 FF+ at that TDP with GTX 1080M that is based on GTX 1070? The hype is REAL 
.
2816 GCN cores, clocked at 1150 MHz give 6.5 TFLOPs of compute power. AMD may want to bring also HBM version of Ellesmere GPU that would have 3072 GCN cores and due to saved power higher core clocks, even 1250 MHz. That would put it at 7.6 TFLOPs, and 125W of power.
.2816 GCN cores, clocked at 1150 MHz give 6.5 TFLOPs of compute power. AMD may want to bring also HBM version of Ellesmere GPU that would have 3072 GCN cores and due to saved power higher core clocks, even 1250 MHz. That would put it at 7.6 TFLOPs, and 125W of power.
Last edited:
9 TFlops seems high to me at 100W. I'd guess closer to 7 or 8. But again, my suspicion is that if Apple adopted Polaris, they'd overclock it and try and run it up to 150w and call it a custom card. Especially if Polaris 10's lower power usage is due to clock and not core count. I don't think, at least at the top end, Apple would be looking at stock Polaris. Apple has a massive proprietary cooler they can stick Polaris on.
And in case anyone thinks I'm just saying that because of the 1080, you can go back in my post history...
Guys, its even worse than I thought about GTX 1070. It is 2048 CUDA core GPU with... 1575 MHz core clock. That is the core clock that brings 6.5 TFLOPs of compute power. With 150W TDP. On new node... Nope, GTX 1080M with 1300 MHz will not have 100W, it will have 125W.
Lol, of course not. I expected... more. This is new node. This GPU is very comparable to GTX 980 in terms of core config. It has only 30% more performance than GTX 980. But I mean, I expected more...
Enough with the AMD FUD, OK? I realize you have a job, why not just wait until benchmarks come out to toss the mud?
Because someone talks about Nvidia, it also means he talks about AMD? I know that your business is not selling anymore, but stop at least accusing me for being AMD Shill. You have more interest in promoting Nvidia on this forum, than I would have with AMD.
Wasn't it mentioned somewhere that Polaris 10 targets 390 and 390X performance, with TDP 110-150W? That means 5-6TFLOPs GPGPU power.
So it seems that
So it seems that
- GTX1080 is partly competing with Fury (no totally lapping, but there's the $100 difference too)
- GTX1070 is a direct enemy of Polaris 10
- 1080M and Polaris mobile too
- Nvidia wont release anything to compete with Polaris 11?
Forni from Bits&Chips say it might come in Q4 of 2016.
http://www.bitsandchips.it/english/52-english-news/6620-rumor-nvidia-pascal-timeline
Remember the gating mechanism AMD is using, unused cores will be powered down and that will help with total power draw and temperature, at least when both GPUs are not in use.
It's a confirmation that Polaris is going to compete at the low to mid ends, at least initially. AMD's big market right now is consoles and low profile boxes, so it makes sense for them. No one is going to be putting a GTX 1080 into a PS4.
That said, for the Mac Pro, if we're looking at stock components, the GTX 1080 is a better fit. It's slightly above the ideal thermal profile, but that can be fixed with a small under clock, or more load balancing like the current model. Polaris is a tough fit for the Ultimate Mac Pro SKU at this point because TDP/performance is too low, and Fury's TDP is too high. Even Fury Nano is a problem now because it's now a really inefficient architecture compared to Polaris and Pascal.
Hopefully we'll either see a custom-ish Polaris card on the nMP, or a Pascal option on the high end. At the Mac Pro's price point, it's harder to justify a $1000 premium for a Polaris 10 upgrade. I'm still leaning towards AMD having something special up their sleeves for the nMP. Maybe a higher binning. If a Mac Pro ships in August or September (and is announced at WWDC) it still gives AMD time to put something together and announce in a flashy way at WWDC. The lack of a higher end on Polaris is due to timing, not architecture.
The missing Polaris based Fury hole in AMD's lineup is a little conspicuous if you ask me.
Get focus on nVidia marketing moves:
They push their GTX 1080 to deliver a performance expected superior to AMD Polaris ,
They do a sneaky move to include dp1.4 (actually with dp1.3 only capabilities), just to build hype and a differentiator on AMD.
They push their GTX 1080 to deliver a performance expected superior to AMD Polaris ,
They do a sneaky move to include dp1.4 (actually with dp1.3 only capabilities), just to build hype and a differentiator on AMD.
- Status