AMD Radeon Pro Vega Graphics Coming to MacBook Pro Next Month

[dmartin]

Today Apple has introduced its new iPad Pro and has resurrected both the MacBook Air, now with Retina display, and the Mac mini. In addition to all these new announcements, there's another one the company didn't talk about at the event: an update for the MacBook Pro that will now feature AMD Radeon Pro Vega graphics chips. According to Apple, this will allow a 60% improvement in performance over the Radeon Pro 560X, the most powerful GPUs available to date for the MacBook Pro.

These professional-oriented chips are manufactured on a 14 nm process, this is not the 7 nanometer Vega 20 that's expected to be released later this year for the enterprise space. The GPU is codenamed "Vega 12" and comes with HBM2 memory, which helps reduce space requirements in a notebook, compared to traditional GDDR5 memory. Vega Pro 20 in the MacBook Pro has 20 CUs (1,280 shaders), Vega Pro 16 has 16 CUs (1,024 shaders). Both MacBook Pro configurations will be available on November 14th at a price yet to be determined.



View at TechPowerUp Main Site

 

[Daven]

You do know that the Radeon Vega Pro 20 is not the same as Vega 20 right? The Radeon Vega Pro 20 is 20x64 Stream processors (1280 total). Just like the Radeon Pro Vega 56 and 64 in the iMac Pro are 3564 (56x64) and 4096 (64x64) stream processors. The Vega 20 at 7 nm is the full 4096 workstation class version with 32 GB of HBM2 that AMD has been teasing and is not in any Apple product yet.

I'm thinking the Radeon Vega Pro 20 with 1280 stream processors in this MBP upgrade has a 30-40 W TDP.

Edit: Oh and the Radeon Vega Pro 20 is probably a 12 nm FinFet process.

 

[Divide Overflow]

So much for the story the other day claiming "Vega 20" would be the first 7 nm GPU, with a launch at the end of Q1.

 

[T1beriu]

These professional-oriented chips are manufactured on a 7 nm process

This is not true. You're confusing it with Vega 20 die codename for machine learning, the first 7nm GPU that will come to maker later this year as Lisa Su specified a few days ago.

This is Vega 20 SKU for consumers which is built on 14nm as it's specified on AMD's website. This was announced/teased as CES 2018.

 

[Daven]

This is not true. You're confusing it with Vega 20 die codename for machine learning, the first 7nm GPU that will come to maker later this year as Lisa Su specified a few days ago.

This is Vega 20 SKU for consumers which is built on 14nm as it's specified on AMD's website. This was announced/teased as CES 2018.

You are correct. dmartin has no idea what he is talking about in his post.

 

[jabbadap]

Surely this is Vega 12 gpu, that dB entry is from different gpu altogether. Product names are Radeon™ Pro Vega 20(20*64=1280 stream processors) and Radeon™ Pro Vega 16(16*64 = 1024 stream processors). By looking at that die size, I kind of doubt fully enabled chip is only 20CUs though.

https://www.amd.com/en/graphics/radeon-pro-vega-20-pro-vega-16

 

[WikiFM]

Wait what!? Vega 20 on Apple laptops!?
Then I read the comments... Now it makes sense

 

[W1zzard]

Fixed

 

[T4C Fantasy]

You do know that the Radeon Vega Pro 20 is not the same as Vega 20 right? The Radeon Vega Pro 20 is 20x64 Stream processors (1280 total). Just like the Radeon Pro Vega 56 and 64 in the iMac Pro are 3564 (56x64) and 4096 (64x64) stream processors. The Vega 20 at 7 nm is the full 4096 workstation class version with 32 GB of HBM2 that AMD has been teasing and is not in any Apple product yet.

I'm thinking the Radeon Vega Pro 20 with 1280 stream processors in this MBP upgrade has a 30-40 W TDP.

Edit: Oh and the Radeon Vega Pro 20 is probably a 12 nm FinFet process.

its 14nm

 

[ArbitraryAffection]

So this is that baby Vega chip with a single stack of HBM2 memory, that was originally advertised as 'Vega Mobile'? Huh.

 

[xkm1948]

Didn't the macbook have overheating issues? Now it is gonna be even hotter thanks to Vega.

 

[phanbuey]

Didn't the macbook have overheating issues? Now it is gonna be even hotter thanks to Vega.

They've had massive throttling issues since Haswell - the last d gpu enabled mbp that didn't constantly throttle both the cpu and gpu was the Sandy Bridge/Ivy Bridge with the Nvidia 650/750m cards.

ever since then they've just been cramming hotter components in them (6 core gpu, higher wattage AMD gpus) and making the chassis slimmer with less airflow. I imagine these don't perform anywhere near spec.

I really feel sorry for the people that get to learn that after spending a few grand on one of these.

 

[Valantar]

So much for the story the other day claiming "Vega 20" would be the first 7 nm GPU, with a launch at the end of Q1.

You're (understandably, but still) confusing two different chips. Vega 20 is the rumored code name for an upcoming 7nm workstation/server/HPC focused full-fat desktop/server Vega die design, rumored to have at least 64CUs/4096SPs and four stacks of HBM2. This, the Vega Pro 20, is the marketing name (not code name) for a 20CU/1280SP product based on the Vega Mobile chip shown off at CES 2018. Similar names, but that's due to different naming conventions being in use. Apple seems to like the "count the CUs" naming scheme, while AMD gives chips code names from the arch name+XY, where X is the "generation" within the architecture and Y enumerates the order the chips were designed in (supposedly, though this seems a bit random - this seems to be codenamed Vega 12, so where is 11?).

 

[ghazi]

Good design win for AMD. I wonder if this chip really only has 20 CUs. Seems kind of skimped, but I wouldn't be surprised if it performs exactly the same as the 24CU GPU in KBL-G given that this is actually legit Vega.

They've had massive throttling issues since Haswell - the last d gpu enabled mbp that didn't constantly throttle both the cpu and gpu was the Sandy Bridge/Ivy Bridge with the Nvidia 650/750m cards.

ever since then they've just been cramming hotter components in them (6 core gpu, higher wattage AMD gpus) and making the chassis slimmer with less airflow. I imagine these don't perform anywhere near spec.

I really feel sorry for the people that get to learn that after spending a few grand on one of these.

Yeah, apparently the new 6-core model is slower than the previous quad-core one, runs at 2.2GHz under full load. LOL

You're (understandably, but still) confusing two different chips. Vega 20 is the rumored code name for an upcoming 7nm workstation/server/HPC focused full-fat desktop/server Vega die design, rumored to have at least 64CUs/4096SPs and four stacks of HBM2. This, the Vega Pro 20, is the marketing name (not code name) for a 20CU/1280SP product based on the Vega Mobile chip shown off at CES 2018. Similar names, but that's due to different naming conventions being in use. Apple seems to like the "count the CUs" naming scheme, while AMD gives chips code names from the arch name+XY, where X is the "generation" within the architecture and Y enumerates the order the chips were designed in (supposedly, though this seems a bit random - this seems to be codenamed Vega 12, so where is 11?).

Vega 11 is Raven Ridge.

 

[Valantar]

Vega 11 is Raven Ridge.

Isn't that based on CU count, i.e. a marketing name?

Yeah, apparently the new 6-core model is slower than the previous quad-core one, runs at 2.2GHz under full load. LOL

Apparently they largely fixed the throttling issues of the latest generation through a firmware update. Still quite tragic that it launched in a barely-working state, but supposedly the throttling is not really an issue any longer.

 

[ArbitraryAffection]

Didn't the macbook have overheating issues? Now it is gonna be even hotter thanks to Vega.

Vega is actually very efficient when not pushed to its absolute limit, close to Pascal efficiency. Even if it were an NVIDIA GPU it'd still likely be an overheating mess.

 

[Deleted member 158293]

Indeed, Vega is extremely power efficient at lower clocks. It's actually a perfect fit in this "case"

 

[dmartin]

You're (understandably, but still) confusing two different chips. Vega 20 is the rumored code name for an upcoming 7nm workstation/server/HPC focused full-fat desktop/server Vega die design, rumored to have at least 64CUs/4096SPs and four stacks of HBM2. This, the Vega Pro 20, is the marketing name (not code name) for a 20CU/1280SP product based on the Vega Mobile chip shown off at CES 2018. Similar names, but that's due to different naming conventions being in use. Apple seems to like the "count the CUs" naming scheme, while AMD gives chips code names from the arch name+XY, where X is the "generation" within the architecture and Y enumerates the order the chips were designed in (supposedly, though this seems a bit random - this seems to be codenamed Vega 12, so where is 11?).

Thanks for the clarification, fixed now

You do know that the Radeon Vega Pro 20 is not the same as Vega 20 right? The Radeon Vega Pro 20 is 20x64 Stream processors (1280 total). Just like the Radeon Pro Vega 56 and 64 in the iMac Pro are 3564 (56x64) and 4096 (64x64) stream processors. The Vega 20 at 7 nm is the full 4096 workstation class version with 32 GB of HBM2 that AMD has been teasing and is not in any Apple product yet.

I'm thinking the Radeon Vega Pro 20 with 1280 stream processors in this MBP upgrade has a 30-40 W TDP.

Edit: Oh and the Radeon Vega Pro 20 is probably a 12 nm FinFet process.

You're absolutely right, thanks for the info @Mark Little

You are correct. dmartin has no idea what he is talking about in his post.

Yup, I got that wrong, my bad. Lesson learned!

 

[phanbuey]

Isn't that based on CU count, i.e. a marketing name?


Apparently they largely fixed the throttling issues of the latest generation through a firmware update. Still quite tragic that it launched in a barely-working state, but supposedly the throttling is not really an issue any longer.

It isn't throttling anymore under "normal load" as they capped the wattage to the processor / enabled the regulator chip but as soon as you actually push the machine (or run Cinebench) it's still well below the average score for that processor.

Adding the additional heat of the GPU to that design is irresponsible. I'm curious to see the reviews of the top end model.

That's the score drop off on consecutive tests of cinebench - within 1-2 mins you become heat saturated and lose 10-15% performance - not even pushing that 560x.

 

[Frick]

They've had massive throttling issues since Haswell - the last d gpu enabled mbp that didn't constantly throttle both the cpu and gpu was the Sandy Bridge/Ivy Bridge with the Nvidia 650/750m cards.

ever since then they've just been cramming hotter components in them (6 core gpu, higher wattage AMD gpus) and making the chassis slimmer with less airflow. I imagine these don't perform anywhere near spec.

I really feel sorry for the people that get to learn that after spending a few grand on one of these.

You mean throttling as in dropping below base clocks or throttling as in not running max turbo at all times? I've looked around quickly and I can't find much about this.

Nvm saw this.

It isn't throttling anymore under "normal load" as they capped the wattage to the processor / enabled the regulator chip but as soon as you actually push the machine (or run Cinebench) it's still well below the average score for that processor.

Adding the additional heat of the GPU to that design is irresponsible. I'm curious to see the reviews of the top end model.

View attachment 109661

View attachment 109662

That's the score drop off on consecutive tests of cinebench - within 1-2 mins you become heat saturated and lose 10-15% performance - not even pushing that 560x.

All laptops drops in performance these days.

 

[Valantar]

All laptops drops in performance these days.

Yep, that's the nature of modern turbo schemes. Seeing how the net result overall is increased performance in a smaller package with better battery life at idle, I'm all for it. The new 4-core "15W" Intel chips easily pull >50W when turboing on all cores. Can't beat physics, after all. Still, if those few seconds far beyond TDP gets the work done, that's all we really need. Nothing wrong with thin-and-lights having worse sustained performance than bigger laptops, as they used to simply have worse performance all the time instead. At least now they're sufficiently fast for pretty much everything.