Nvidia 1080ti with new drivers in external enclosure quadruples MacBook Pro native perform...

AppleInsider

Enthusiasts have wasted no time in testing the new Nvidia Pascal video card drivers, and have found external GPU performance nearly four times that of the Radeon Pro 450 in the 15-inch MacBook Pro.

External GPU enthusiast site egpu.io has affixed a GTX 1080 Ti to an AKiTiO Note and Mantiz Venus enclosure -- similar to that used for AppleInsider testing in January. While the cards may be hamstrung slightly by the Thunderbolt 3 interface not being as fast as a 16x PCI-E slot, the results are nonetheless impressive.

In a best-case scenario utilizing benchmarks, the 15-inch MacBook Pro with Radeon Pro 450 scores 5822 on the Luxmark 3.1 benchmark, with the Radeon Pro 460 scoring 6056.

An external GPU feeding video back to the screen of the same MacBook scores 22,673 with a Nvidia GeForce GTX 980 Ti, and 23,172 with the newly enabled 1080 Ti. At this time, the group does not have the new Nvidia Titan Xp video card for testing.

Other benchmarks have similar gaps in performance. the Valley 1.0 test scores 706 on the Radeon Pro 450, and 2353 with the 1080 Ti on the internal display -- climbing to 3031 on an external.

The Heaven 4.0 test scores 360 on the Radeon Pro 450. Demonstrating what can happen when Thunderbolt 3 bandwidth is constrained, the Heaven 4.0 test on the 1080 Ti routed to the internal display measures 1422, but nearly doubles to 2640 when sent to an external display.

Also recently discovered by eGPU.io is the fact that on the 13-inch MacBook Pro with Touch Bar, the Thunderbolt 3 controllers are attached to the PCH-H controller. As a result, available bandwidth to Thunderbolt 3 devices can be constrained by other components connected through the PCH-H controller -- including the flash storage on the computer.

On the 15-inch MacBook Pro with Touch Bar, the Thunderbolt 3 controllers are directly connected to the CPU, eliminating any possible bottleneck through the PCH-H controller.

AppleInsider's own testing on eGPU suitability for users will continue with a Mantiz enclosure, and a casing from Bizon.

mdriftmeyer

And when the next Macbook Pro debuts with a Vega HBM2.0 GPGPU discrete GPU that lead gap will plunge considerably to a cost loss for purchasing those eGPUs.

zimmie

PCIe throughput matters less than most people think for most GPU use. You can block off PCIe lanes with tape. On most GPUs in most games, there is no difference at all dropping it from 16 lanes to eight. When you drop to four lanes, you typically get longer loading times and a small framerate drop. Dropping to two lanes typically gives significantly longer loading times and >30% framerate drops.

This can matter for OpenCL use, but it frequently does not. PCIe throughput is only really used getting your dataset into the video card's RAM and getting the result out. It will slow down some really trivial data manipulation, and it will be slower to work on datasets too large for the card's RAM. Anything which requires more than a few seconds to compute won't be meaningfully slower on an eight-lane or four-lane bus.

tylersdad

mdriftmeyer said:

And when the next Macbook Pro debuts with a Vega HBM2.0 GPGPU discrete GPU that lead gap will plunge considerably to a cost loss for purchasing those eGPUs.

I think it's very unlikely that Apple's first GPU will be able to leapfrog, or even come close to the same performance that nVidia achieves. nVidia's been doing this for decades. 

tmay

tylersdad said:

mdriftmeyer said:

And when the next Macbook Pro debuts with a Vega HBM2.0 GPGPU discrete GPU that lead gap will plunge considerably to a cost loss for purchasing those eGPUs.

I think it's very unlikely that Apple's first GPU will be able to leapfrog, or even come close to the same performance that nVidia achieves. nVidia's been doing this for decades. 

Yet, Nvidia couldn't translate that performance to a low power envelope with Tegra. Perhaps Apple's advantage lies with integrating a Cannon Lake generation CPU w/integrated graphics with a Apple GPGPU optimized for Mac OS and Metal, rather than Windows and Direct X.

Either way, I like the promise of a TB 3 connected GPU, though can imagine a higher level of integration including a power supply in a smaller form factor being more desirable than through an external drive enclosure with PCIe slot.

foggyhill

tylersdad said:

mdriftmeyer said:

And when the next Macbook Pro debuts with a Vega HBM2.0 GPGPU discrete GPU that lead gap will plunge considerably to a cost loss for purchasing those eGPUs.

I think it's very unlikely that Apple's first GPU will be able to leapfrog, or even come close to the same performance that nVidia achieves. nVidia's been doing this for decades. 

NVDIA gets creamed in low power env, so yeah, they can "leapfrog" it, not problem, especially if they design the GPU to integrate tightly with CPU.
This will provide them with something nobody else has. the SOC is probably littered with DSP's already so that's just one step more in co-processing for the CPU.

nht

mdriftmeyer said:

And when the next Macbook Pro debuts with a Vega HBM2.0 GPGPU discrete GPU that lead gap will plunge considerably to a cost loss for purchasing those eGPUs.

For $300 the Node is more than affordable for the potential boost it provides.  And no, no GPU built into any MBP with around a 35W TDP will come close to Titan or 1080 Ti level performance.  And you will most likely be able to run any Vega based GPU in any enclosure that supports a Titan with its 250W TDP.

The Radeon Pro 460 (35W TDP) vs the same generation Polaris RX480 (150W TDP) is 1.85 TFLOPSs vs 5.8 TFLOPs so any vega based mobile GPU that Apple will use will have the same rough performance delta between mobile and desktop part.

Even a tech support engineer should understand that...I'm getting really tired of your negative nanny attitude every time something new gets announced.  Easy to use and affordable eGPUs will be a freaking awesome enhancement for any mac user with graphic needs.

lorin schultz

zimmie said:

PCIe throughput matters less than most people think for most GPU use. You can block off PCIe lanes with tape. On most GPUs in most games, there is no difference at all dropping it from 16 lanes to eight. When you drop to four lanes, you typically get longer loading times and a small framerate drop. Dropping to two lanes typically gives significantly longer loading times and >30% framerate drops.

This can matter for OpenCL use, but it frequently does not. PCIe throughput is only really used getting your dataset into the video card's RAM and getting the result out. It will slow down some really trivial data manipulation, and it will be slower to work on datasets too large for the card's RAM. Anything which requires more than a few seconds to compute won't be meaningfully slower on an eight-lane or four-lane bus.

I know less than nothing about computer architecture, so please forgive me if I misunderstood what @"Mike Wuerthele" wrote, but the impression I got is that the bandwidth limitation becomes an issue because the Thunderbolt 3 buss is a two-way street in this scenario. It has to carry both the instructions to the card AND all the pixel data back to the internal display. That's why performance was better when using an external display.

Is my understanding correct or have I missed something?

edited April 2017

wizard69

mdriftmeyer said:

And when the next Macbook Pro debuts with a Vega HBM2.0 GPGPU discrete GPU that lead gap will plunge considerably to a cost loss for purchasing those eGPUs.

The other thing here is why would you even bother to compare a GPU card designed for desktop machines with one designed for use in a laptop.    Makes absolutely no sense to me.

mike wuerthele

lorin schultz said:

zimmie said:

PCIe throughput matters less than most people think for most GPU use. You can block off PCIe lanes with tape. On most GPUs in most games, there is no difference at all dropping it from 16 lanes to eight. When you drop to four lanes, you typically get longer loading times and a small framerate drop. Dropping to two lanes typically gives significantly longer loading times and >30% framerate drops.

This can matter for OpenCL use, but it frequently does not. PCIe throughput is only really used getting your dataset into the video card's RAM and getting the result out. It will slow down some really trivial data manipulation, and it will be slower to work on datasets too large for the card's RAM. Anything which requires more than a few seconds to compute won't be meaningfully slower on an eight-lane or four-lane bus.

I know less than nothing about computer architecture, so please forgive me if I misunderstood what @"Mike Wuerthele" wrote, but the impression I got is that the bandwidth limitation becomes an issue because the Thunderbolt 3 buss is a two-way street in this scenario. It has to carry both the instructions to the card AND all the pixel data back to the internal display. That's why performance was better when using an external display.

Is my understanding correct or have I missed something?

I believe Zimmie is more referring to this: "While the cards may be hamstrung slightly by the Thunderbolt 3 interface not being as fast as a 16x PCI-E slot, the results are nonetheless impressive."

Your understanding is correct.

wizard69

nht said:

mdriftmeyer said:

And when the next Macbook Pro debuts with a Vega HBM2.0 GPGPU discrete GPU that lead gap will plunge considerably to a cost loss for purchasing those eGPUs.

For $300 the Node is more than affordable for the potential boost it provides.  And no, no GPU built into any MBP with around a 35W TDP will come close to Titan or 1080 Ti level performance.  And you will most likely be able to run any Vega based GPU in any enclosure that supports a Titan with its 250W TDP.

The Radeon Pro 460 (35W TDP) vs the same generation Polaris RX480 (150W TDP) is 1.85 TFLOPSs vs 5.8 TFLOPs so any vega based mobile GPU that Apple will use will have the same rough performance delta between mobile and desktop part.

Even a tech support engineer should understand that...I'm getting really tired of your negative nanny attitude every time something new gets announced.  Easy to use and affordable eGPUs will be a freaking awesome enhancement for any mac user with graphic needs.

Value is in the eye of the beholder and frankly I wouldn't consider such an investment a good use of my computing dollar.   As far as the rest of the world goes, we don't see these external enclosures taking off, so I don't believe the value equation please many at all.

In any event there was nothing negative at all negative about the posted comments.    The presentation just didn't make sense and got the response required.    It simply makes no sense to compare a high performance desktop video card with a laptop GPU.     

zimmie

lorin schultz said:

zimmie said:

PCIe throughput matters less than most people think for most GPU use. You can block off PCIe lanes with tape. On most GPUs in most games, there is no difference at all dropping it from 16 lanes to eight. When you drop to four lanes, you typically get longer loading times and a small framerate drop. Dropping to two lanes typically gives significantly longer loading times and >30% framerate drops.

This can matter for OpenCL use, but it frequently does not. PCIe throughput is only really used getting your dataset into the video card's RAM and getting the result out. It will slow down some really trivial data manipulation, and it will be slower to work on datasets too large for the card's RAM. Anything which requires more than a few seconds to compute won't be meaningfully slower on an eight-lane or four-lane bus.

I know less than nothing about computer architecture, so please forgive me if I misunderstood what @"Mike Wuerthele" wrote, but the impression I got is that the bandwidth limitation becomes an issue because the Thunderbolt 3 buss is a two-way street in this scenario. It has to carry both the instructions to the card AND all the pixel data back to the internal display. That's why performance was better when using an external display.

Is my understanding correct or have I missed something?

They didn't give the pixel dimensions of the external display. The 15" MBP's internal display is 2880x1800 pixels for a total of a little under 5.2 million pixels. The most common external displays I see are 1920x1080, which is just under 2.1 million pixels. I don't know if these benchmarks correct for the output pixel count. If they don't, a 2x improvement could come solely from running 1/2 the pixels.

bobolicious

Is the Blade Pro worth a performance comparison (OS aside) with its GTX 1080 and phase change cooling system ?
https://www.razerzone.com/gaming-systems/razer-blade-pro
http://www.pcmag.com/news/352646/razer-blade-pro-laptop-gets-thx-certification-kaby-lake

colinng

zimmie said:

They didn't give the pixel dimensions of the external display. The 15" MBP's internal display is 2880x1800 pixels for a total of a little under 5.2 million pixels. The most common external displays I see are 1920x1080, which is just under 2.1 million pixels. I don't know if these benchmarks correct for the output pixel count. If they don't, a 2x improvement could come solely from running 1/2 the pixels.

I am going entirely by the description (YouTube won't stop foisting this insanely irritating 1.5 minute Gramarly ad on me - so to heck with Google and YouTube) but I think the output goes back along the ThunderBolt 3 to the MacBook Pro and is subsequently rendered on the internal display. So they could be using the same resolution. 

xzu

bobolicious said:

Is the Blade Pro worth a performance comparison (OS aside) with its GTX 1080 and phase change cooling system ?
https://www.razerzone.com/gaming-systems/razer-blade-pro
http://www.pcmag.com/news/352646/razer-blade-pro-laptop-gets-thx-certification-kaby-lake

That is first Laptop I would consider a desktop replacement. Kaby Lake, 32gb Ram, dual PCIe drives, and a 1080, oh my goodness. 

The eGPU is exciting as well. At least we are moving in the right direction.

edited April 2017

rezwits

foggyhill said:

tylersdad said:

mdriftmeyer said:

And when the next Macbook Pro debuts with a Vega HBM2.0 GPGPU discrete GPU that lead gap will plunge considerably to a cost loss for purchasing those eGPUs.

I think it's very unlikely that Apple's first GPU will be able to leapfrog, or even come close to the same performance that nVidia achieves. nVidia's been doing this for decades. 

NVDIA gets creamed in low power env, so yeah, they can "leapfrog" it, not problem, especially if they design the GPU to integrate tightly with CPU.
This will provide them with something nobody else has. the SOC is probably littered with DSP's already so that's just one step more in co-processing for the CPU.

Yeah, Apple is going in the oppossite direction of NVIDIA, as in GPUs less than 16nm in size that are 1/5 the speed, instead of the size of a shoebox... I mean I guess you kinda have to look at what's going on here? So if logic follows, (we'll use moore's law for FUN), in 2-3 years not really gonna be the same speed but that would be pretty good, if it was close. While they're heading towards that end where they still have a GPU the size of a shoe box but then can't make it smaller, cooler, quieter, and more energy efficient. Then NVIDIA would have to what? start from scratch maybe, then..? I don't know if this article is still relevant but http://appleinsider.com/articles/15/01/23/how-amd-and-nvidia-lost-the-mobile-gpu-chip-business-to-apple-with-help-from-samsung-and-google-

edited April 2017

nht

wizard69 said:

Value is in the eye of the beholder and frankly I wouldn't consider such an investment a good use of my computing dollar.   As far as the rest of the world goes, we don't see these external enclosures taking off, so I don't believe the value equation please many at all.

Given that in recent months we have eGPU enclosure offerings actually ship from AKiTiO, PowerColor, Mantiz, Razer, Asus, Bizon, Sonnet and if you count proprietary Alienware there appears to be a market out there.  There are more that have been announced but haven't shipped and unlike before with TB and TB2 when new annoucnements were likely to be vaporware there's a reasonable expectation that these will result in actually shipping product.

It's been enabled by TB3 becoming commonplace on laptops and the much lower price on enclosures compared with a few years ago when Sonnet was pretty much the only game in town.  The USB-C/TB3 form factor is likely going to be the standard for a while.

Sure, you can build a gaming PC for $500 (Core i3 + RX 470) but using the same $500 for an enclosure and a GTX 1060 and use the Core i5 laptop you already own isn't a bad way to go.  This holds for certain Pro users as well.  A eGPU + Titan gives you similar performance to a Mac Pro when using Resolve which is heavily GPU dependent.

bobolicious

xzu said:

bobolicious said:

Is the Blade Pro worth a performance comparison (OS aside) with its GTX 1080 and phase change cooling system ?
https://www.razerzone.com/gaming-systems/razer-blade-pro
http://www.pcmag.com/news/352646/razer-blade-pro-laptop-gets-thx-certification-kaby-lake

That is first Laptop I would consider a desktop replacement. Kaby Lake, 32gb Ram, dual PCIe drives, and a 1080, oh my goodness. 
The eGPU is exciting as well. At least we are moving in the right direction.

I'd also be curious if simply a networked (TB2/3) folder from rMBP to iMac/Pro may be a better option for some - the scant reviews I could find on the Razer suggest, as might be expected, it may be loud, or even throttle when pushing pixels... Phase change cooling is an intriguing idea, reminding me of the liquid cooled G5 pro of years ago.

zimmie

colinng said:

zimmie said:

They didn't give the pixel dimensions of the external display. The 15" MBP's internal display is 2880x1800 pixels for a total of a little under 5.2 million pixels. The most common external displays I see are 1920x1080, which is just under 2.1 million pixels. I don't know if these benchmarks correct for the output pixel count. If they don't, a 2x improvement could come solely from running 1/2 the pixels.

I am going entirely by the description (YouTube won't stop foisting this insanely irritating 1.5 minute Gramarly ad on me - so to heck with Google and YouTube) but I think the output goes back along the ThunderBolt 3 to the MacBook Pro and is subsequently rendered on the internal display. So they could be using the same resolution. 

For the "eGPU" test, yes. The "eGPU external display" test where the scores are so much higher was presumably run targeting an external display.

This is the same reason the smaller iPhones have more apparent graphics power than the larger ones. They're pushing fewer pixels. Last I checked, the SE (A9) was roughly on-par with the 7+ (A10) (in frames-per-second terms during games) in spite of using an older GPU simply because it has fewer pixels.

krisarkade

Can we see tests on a 2016 Core i7 13 inch MacBook? I mean it's amazing that this works but it's kind of redundant to be testing this on a laptop that already has a pretty decent dedicated GPU.

seeing as the 13 inch Core i7's are running the Iris 550 chipsets I can see where an eGPU would be far more useful in this scenario.

lorin schultz

wizard69 said:

[...] It simply makes no sense to compare a high performance desktop video card with a laptop GPU.     

Of course not, but the context makes the article useful and valuable to me. It's only because of an article on AI that I even know I can use an external GPU in an enclosure.

Now, knowing that, and knowing what it costs, the next question is: "Is it worth the hassle/expense?" To know that, I need to know how the performance compares to the laptop by itself.

In that context, the comparison makes perfect sense.