mattspace
macrumors 68040
right, but goMac is having problems with a 680, which can use either the Apple driver, or the web driver, right?
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.
Radeon Vega Frontier Edition Plug and Play in 10.13.1+
- Thread starter theitsage
- Start date
- Sort by reaction score
right, but goMac is having problems with a 680, which can use either the Apple driver, or the web driver, right?
From my understanding, fixes from Nvidia are making their way into the Apple driver. But Nvidia's driver also had issues.
10.13.2 is the first time I'm starting to see serious fixes (and new problems), but I can't compare with the Nvidia driver on 10.13.2.
(I don't want to get into exactly what I'm working on and risk working relationships.)
MisterAndrew
macrumors 68030
Thanks for testing this card! I suppose we can assume the WX 9100 would also behave similarly. It's interesting that it scores lower than my Vega 56 with 64 bios in the LuxMark benchmark.
Now I’m reading that the Vega 56 doesn’t evenly draw power (as noted in the other thread) and doubting my choice.
It’s really strange that on BIOS b Furmark is causing shutdowns. I wonder if even the aux connectors are pulling unevenly.
Kind of sucks. Wouldn’t be a problem if AMD just pulled a bit more from the slot. Wonder what Vega Nano will end up looking like.
It’s really strange that on BIOS b Furmark is causing shutdowns. I wonder if even the aux connectors are pulling unevenly.
Kind of sucks. Wouldn’t be a problem if AMD just pulled a bit more from the slot. Wonder what Vega Nano will end up looking like.
MisterAndrew
macrumors 68030
Yeah, to safely power any Vega card (so far) you need Pixlas mod or secondary PSU.
Yeah. Something is funny here. If Vega 56 is drawing mostly from the aux connectors it should still be ok-ish. It’s going to pull the Mac Pro out of reconmended spec but not out of design tolerance. On BIOS B it shouldn’t be pulling more than 90-ish watts from each aux connector. Again, out of Apple supported spec but not out of tolerance. (It seems Apple programmed the tolerance to be about 120.)
I’d have to plug in and take a look, but the only other explanation would be that either Vega 56 has a max much higher than we’re thinking, or the draw is uneven and pulling more than 120 on an aux connector.
Not sure yet how you’d rebalance that on dual 8 pins. You’d almost need a fully custom cable.
Or I could just draw from SATA bay power, or step down to a 580.
AMD is making good products but from a power efficiency standpoint they aren't there yet. There are AMD cards that are dual 8 pin that are only single 8 pin or even dual 6 pin on the nvidia equivalent side.
MisterAndrew
macrumors 68030
Even balanced by using 2 y-cables it could still pull too much and SATA bay power could be dangerous too. If you already have the card then just do the pixlas mod. It's not that difficult. You could get a Sapphire PULSE Radeon RX 580 and not worry about it, but it's about half as fast.
The strange thing about Vega 56 is it's efficient enough. It's actually more efficient than some of the Apple cares. It's just drawing power un-evenly.
I may plop mine in and figure out if one aux is being pulled more than another. But if they're being pulled evenly Vega 56, even without drawing much from the PCIe slot itself, should fit nicely inside 120w per aux connector.
[doublepost=1512064471][/doublepost]
The trick here is I'm not sure you can balance with y cables. At least not with off the shelf cables. Normally you'd push both aux pins into an 8 pin connector but... If I'm actually pulling more than 75w that won't necessarily work. I'm basically trying to shove 1.5x the max load of a single 8 pin into a single 8 pin and that's not probably going to work.
If you could balance I'm pretty sure you wouldn't need the SATA power to begin with.
MisterAndrew
macrumors 68030
The issue is that it can draw spikes of power that are too much. The current draw that you can see it taking from like iStat menus is an average. The Vega 56 is a 64 that didn't quite make the cut in the manufacturing process. They were able to sell the slightly defective chips by disabling some of the compute units and downclocking it without changing anything else. To each their own, but I would not use the mini pci-e connectors on the logic board to power it.
Sure. The issue doesn't look like the max TDP though, especially on BIOS B. It looks like an issue with unbalanced draw on the aux power.
Definitely wouldn't be an issue, spikes and all, if Vega 56 would just draw more from slot power. What I'm reading online is that it seems like Vega only uses slot power for the blower.
From what I'm reading about GPUs on the Mac Pro in general, total power isn't usually a problem and the limit is definitely not 75w (it's not 150w either.) It's when a card is imbalanced and not evenly pulling power.
(The Vega 64 is still just no good though because even if you're willing to go beyond 75w the power draw is still just way too much. Especially if it draws, like 56, nothing from slot power.)
Last edited:
h9826790
macrumors P6
This is what I am interested to know as well. The Vega 56 should not pull more than the 1080Ti. And this is confirmed in almost all review. And if I can run Frumark on a 1080Ti with just the mini 6pin. The Vega 56 should able to do that as well. Unless the Vega 56 is programmed to draw almost no power from the slot, and nearly totally rely on the aux power. There is one report here about Furmark will shutdown the Mac straight away even with proper power management (balanced the load).
Since there is just one report, I really want to know more if any other user able to share the load between the mini 6pin and successfully run Furmark without any shutdown. Also, what's the power draw reading when doing it.
Well the difference again would be that Vega doesn't seem to really pull slot power, but puts almost it's full load on the aux connectors.
But even if the slot is only pulling 30w, aux should still be ok. It's more than 75w, but the Apple six pins aren't 75w six pins.
h9826790
macrumors P6
Same "issue" observed on the 7950 Mac Edition card. The card only draw 25W from the slot, and all remaining power from the 6pins. If it draw up to around 50W from the slot. It will make the power management so much easier to run 2x 7950 on the cMP.
Just check the Tom's hardware review. You are completely right. The card's actual power draw is just about 230W (can go up to ~260W depends on BIOS selected). But it can only draw up to 20W from the slot. Which means the remaining 210-250W will be pulled from the mini 6pins.
If select BIOS 2 (max draw ~230W), the real world power draw can just stay within the mini 6pin's practical limit. However, if select BIOS 1, the power draw on each mini 6pin may still exceed that ~120W hard limit even the draw is perfectly balanced.
Also, we really don't know how to card react / perform in MacOS. The card's power draw is straightly controller by the drivers in Windows. If TDP set at 230W, the card won't draw more than that (average, not spike). However, on the Mac side, it seems all current Apple computers' model are thermal limiting. The driver don't need to straightly control the power draw, the GPU will throttle itself anyway. I am thinking if the Apple driver is actually programmed to fully benefit from the +20% TDP. Which let their GPU can boost the performance for a very short period of time before thermal throttling occur. And of course they will take that little boost as the "GPU's performance reference". What Apple does to the driver should be based on the iMac Pro's design, not cMP. Then won't care about the power draw distribution, or possible hard shutdown. It just won't happen in the iMac Pro, everything will be thermal limiting.
Oooooook. Just spent some time with Vega 56 on the internal connectors. Took some measurements. All measurements are from BIOS B, not A.
I ran Furmark for 10 mins. I know some people might want a longer test but I have to eat so I might run more later. Machine never shut down.
Max power draws according to iStat (approx):
PCI Booster A - 96w
PCI Booster B - 84w
PCI Slot - 25w
Never really moved.
I'm a Furmark newbie so I ran it at the default of 1280x720 because it's not frame capped. If there are better settings let me know.
Both those readings are higher than the 75w of a minimum spec 6 pin, but within known spec for Apple's cables/supply with the extra lead supplied.
I ran Furmark for 10 mins. I know some people might want a longer test but I have to eat so I might run more later. Machine never shut down.
Max power draws according to iStat (approx):
PCI Booster A - 96w
PCI Booster B - 84w
PCI Slot - 25w
Never really moved.
I'm a Furmark newbie so I ran it at the default of 1280x720 because it's not frame capped. If there are better settings let me know.
Both those readings are higher than the 75w of a minimum spec 6 pin, but within known spec for Apple's cables/supply with the extra lead supplied.
h9826790
macrumors P6
So you mind provide more info about how you power the card?
My thought as follow
1) 96+84+25 = 205. A bit too low for Furmark
2) No need to worry about FPS, run it at full screen, native resolution, AA OFF is the best way to stress the GPU
3) 96 is the maximum display limit of iStat. Base on point 1, I suspect it draw at least another 20W from Booster A
4) Base on point 3, I suspect you didn't use anything to share the loading, but just use two independent mini 6pin -> 8 pin cable.
5) 10min is good enough, if shutdown occur, it usually happen in seconds. Of course, once the card warm up, it will draw a bit more, but 10min is good enough to let it stabilise. In the end, it's Furmark, not any normal daily loading.
My own suggestion
A）get a bridge between the mini 6pins and the 8pins. You can use something like the EVGA PowerLink. Or as simple as dual mini 6pin -> single 8pin -> dual 6+2. No need to worry about overload the single 8pin. There are 3x 12V wires inside, and each of them usually able to handle up to 10A. Which means up to 360W in total. You are far far away from that.
B) Run Furmark again at native resolution full screen. Try Benchmark first, then observe the reading in iStat (history) or Hardware monitor. If you are happy with the number, then further try stress mode (10-15 min is enough).
In any case, do NOT go BIOS A, stay at BIOS B.
Last edited:
Still considering if I want to keep the card. If I do I'll get the aux sources balanced.
But without sleep and fan control, I'll still be waiting for good drivers. And by that time, Vega Nano might improve things again. And the resale prices on this are pretty good.
And the card is wicked fast, which is great, but I feel like it's wasted a bit if you don't have 4k (I don't on my Mac Pro).
At least a bunch of Metal bugs dropped off.
h9826790
macrumors P6
Agree that’s a nice OOTB card for cMP (if they fix the issue). Sleep is not a problem for me. I know this function is nice, but with the help of the SSD (and the falling quality of MacOS), i really prefer full shutdown and cold start nowadays. It only take few extra seconds if compare to sleep / wake, but much safer.
However, I really can’t stand with the full fan noise.
Anyway, so poor that AMD just can't do it right, either pulling too much form the PCIe slot (RX480), or way under utilise it (Vega 56).
Last edited:
devon807
macrumors 6502
This is interesting... Vega is already beating a 1080Ti in a game fully utilizing Metal 2:
http://barefeats.com/dirt_rally.html
Interesting to see once drivers are ready from the new iMac.
http://barefeats.com/dirt_rally.html
Interesting to see once drivers are ready from the new iMac.
h9826790
macrumors P6
The info at that page suggest that the game may be CPU limiting.
If you look at the average FPS graph.
iMac 5K is the fastest regardless which GPU using. Then follow by the nMP -> cMP -> MB.
And even the D700 can beat the 1080Ti.
This patten suggest that the game is not GPU limiting (except the Intel 550), but more like CPU limiting (or something else).
I told you guys the Nvidia Metal drivers are not great.
[doublepost=1512155209][/doublepost]
It might be, and I feel like I'm seeing some of that on Vega 56 on the Mac Pro.
But on the same Mac Pro, Vega 64 also destroys the 1080 Ti. But again, I think it's the drivers.
Feeling like the CPU is holding me back on my 2010 Mac Pro is sobering. I think for people who are serious about performance we might be getting to the end of the road for the cMP. On AMD I still have a few Metal performance issues that now I'm getting suspicious might be CPU problems.
If you're running a pure compute workflow, the 2010 Mac Pro might give you some room to work with. But if you're tied to CPU too... I think this time next year you're either going to be buying whatever Apple is releasing, or you're going to be moving to Windows.
I'm using the 3.46 ghz 6 core CPUs and I still feel CPU limited now.
(Would have also liked to see a 580 on a Mac Pro in these benchmarks. It's implied a little bit that the Vega Metal drivers aren't as mature as the Polaris drivers by those numbers.)
Last edited:
itdk92
macrumors 6502a
So far, of all the dual 1080Ti Mac Por 5.1 we built, none has given us problems. All our customers are super happy.
I am using VEGA cards myself for FCPX and the app is crashing like crazy. Today 10 times at least, doing very different things.
Waiting for 10.13.2 and 10.4!
I am using VEGA cards myself for FCPX and the app is crashing like crazy. Today 10 times at least, doing very different things.
Waiting for 10.13.2 and 10.4!
devon807
macrumors 6502
I wonder how soon to the release of the iMac Pro we will begin to see proper drivers come out.