AMD releasing Pro 460, 455, 450 GPUs in Macbook Pro Laptops

[Azix]

I thought it was the physical package that was reduced in height,but it seems they actually reduced the height of the actual wafer itself.

I was wondering about this. On the AMD site they mentioned

To enable the thinnest graphics processor possible, AMD also employs a complex process known as 'die thinning' to reduce the thickness of each wafer of silicon used in the processor from 780 microns to just 380 microns, or slightly less than the thickness of four pieces of paper.

which suggests it wasn't just the power etc that made them go with AMD, but also the physical chip dimensions. should this not be in the desktop polaris 11 (rx 460)? My assumption was these MBP gpus are the same as rx 460

 

[Erenhardt]

Ohhh amd...

Slap 4 Zen cores at 3.5GHz, this 35W thingy and 4GB HBM into single package, and I'm sold.

 

[jpiniero]

which suggests it wasn't just the power etc that made them go with AMD, but also the physical chip dimensions. should this not be in the desktop polaris 11 (rx 460)? My assumption was these MBP gpus are the same as rx 460

Apple's not going to use nVidia anyway because of the patent lawsuit. Unless AMD really pisses Apple off I think they will stick with AMD when they want to use a dGPU in a product.

 

[Gikaseixas]

Looked at any NVidia thread lately? Same thing or worse.

Your buddies are plain wrong by doing this. Its not the right thing to do in either brand related thread so instead of saying "oh but others do it too" you should tell them to cool it off.
Like the saying - two wrongs don't make a right

 

[Gikaseixas]

Ohhh amd...

Slap 4 Zen cores at 3.5GHz, this 35W thingy and 4GB HBM into single package, and I'm sold.

That could be a turning point for APU's but right around 100W or thereabouts

 

[antihelten]

They're not the only ones who've reduced clocks to hit power targets.

So basically for the GP104, Nvidia had to cut performance by 39% (from 8.87 TFLOPS to 5.44 TFLOPS) to get a 58% reduction in TDP (from 180W to 75W), or in other words they achieved a 47% increase in perf/TDP*

Meanwhile AMD had to cut performance by 15% (from 2.15 TFLOPS to 1.86 TFLOPS), to get a 53% reduction in TDP (from 75W to 35W), which would be an 85% increase in perf/TDP*.

All in all this looks like a perfectly decent scaling from AMD, so I don't know why some people are up in arms.

*Obviously TDP isn't necessarily a particularly accurate measurement of actual power usage, and AMD and Nvidia's Boost mechanism also brings a fair bit of uncertainty, so the above number has to be taken with a healthy dose of salt.

 

[Piroko]

I thought it was the physical package that was reduced in height,but it seems they actually reduced the height of the actual wafer itself.

Yes, seems like this was only talking about the die thickness. The wafer thinning is typically used for die stacking applications but it does shave off 0.4mm in total assembly height and probably brings the GPU package closer in height to Intels CPU package.

 

[AtenRa]

Ohhh amd...

Slap 4 Zen cores at 3.5GHz, this 35W thingy and 4GB HBM into single package, and I'm sold.

Unless Apple or someone else (HP, Dell etc) would want a APU like that, im afraid AMD will not even dare to spend resources to create such a die.
But, ZEN APUs with 12 CUs + DDR-4 3000MHz and higher at 35W to 65W TDP could be a nice cheaper alternative for entry SFF gaming systems.
If RavenRidge be able to reach 1080p 30fps (Low/Med settings) in the majority of new games (latest DX-12 titles as well) at an affordable price (up to $200 for 4C 4T + 12 CU iGPU) , it could create new interest for APUs.

 

[dark zero]

Apple of course mostly cares about compute power, and at 1.8 TF is pretty good at 35W.



No, only the 15" comes standard with dGPUs when it didn't before.

Considering that it has better FP64 than nVIDIA GPUs is pretty interesting to watch.

 

[Flapdrol1337]

1.84TFlops at <35w sounds very good, is there a reason why there are posts here thinking otherwise?

It's up to* 1.84 tflops, so the maximum turbo clock is a little over 900 MHz.

A base "guarantied" speed would be nice to know, as well as the tdp, which is also unknown.

 

[frozentundra123456]

Your buddies are plain wrong by doing this. Its not the right thing to do in either brand related thread so instead of saying "oh but others do it too" you should tell them to cool it off.
Like the saying - two wrongs don't make a right

Well, they arent my "buddies", and I have no control over what anyone else posts, since I am not a mod.

 

[Enigmoid]

This is probably going to be like Tonga in the imac. Theoretically better on paper but throttled like crazy.

 

[Azix]

Unless Apple or someone else (HP, Dell etc) would want a APU like that, im afraid AMD will not even dare to spend resources to create such a die.
But, ZEN APUs with 12 CUs + DDR-4 3000MHz and higher at 35W to 65W TDP could be a nice cheaper alternative for entry SFF gaming systems.
If RavenRidge be able to reach 1080p 30fps (Low/Med settings) in the majority of new games (latest DX-12 titles as well) at an affordable price (up to $200 for 4C 4T + 12 CU iGPU) , it could create new interest for APUs.

http://www.bitsandchips.it/52-english-news/7622-rumor-two-versions-of-raven-ridge-under-development

Do you remember that news? So ... it was not a BR APU, but a RR APU. Our sources were fooled again by AMD. BTW, we can't always win (Here it is some of our news come true). 🙂

However, according to our sources, at the present moment there are two version of Raven Ridge under development, one with a 12CUs GPU and one with a 16CUs GPU. You can see the main differences in the table below.

AMD will attack the Mobile and Desktop markets with two different kind of products, one aimed to equip the Mobile devices, one aimed to equip Multimedia Desktops . Or even Apple products: as usual (See Polaris 11 and Tonga in the new Macs), AMD could sell to Apple a fully unlocked Raven Ridge APU – 16 CUs GPU – and sell in the other markets a castrated Raven Ridge APU – 12CUs GPU.



We have to say that AMD is studying these two versions of Raven Ridge, but we don't know if both will be commercialized (AMD in the recent past has canceled a lot of products: Nolan and Amur (SkyBridge), Komodo (5 modules CPU), etc).

 

[Glo.]

http://video.cnbc.com/gallery/?video=3000563435

Lisa Su claims that AMD co-developed the Polaris GPUs with Apple. Similar situation as LG Displays.
http://semiaccurate.com/2014/03/25/details-apples-gpu-emerge/
http://semiaccurate.com/2014/04/17/semiaccurate-right-apples-foundry-plans/

They did the same for the first part of the story too, the main customer for this 14nm synced process.

SemiAccurate appears to be completely right on this.

 

[crisium]

Now that Apple has their supply maybe we can see 16CU 465 to more properly battle the 1050.

 

[LTC8K6]

Now that Apple has their supply maybe we can see 16CU 465 to more properly battle the 1050.

Apple's supply is probably all there is.

By the time we got a full Tonga card, it was way too late.

 

[Phynaz]

Yeah the quoted article states they got it down to 380 microns in height. That's a seriously thin die for a GPU. This is a good partnership for AMD, they need to appear in premium products and not throttle terribly at low TDP to show that they can make a good premium product.

This would be TSMC's InFO packaging.

 

[raghu78]

This would be TSMC's InFO packaging.

the press release says 14nm FINFET Polaris GPUs . So it has nothing to do with TSMC InFO

http://www.amd.com/en-us/press-releases/Pages/radeon-pro-400-2016oct27.aspx

 

[ACSFernalld]

Does the 35W TDP across all three GPUs mean that all three should provide roughly the same battery life? Specifically I'm wondering about projected heat/battery life differences between the 455 and the 460. I know it's hard to know exactly since they haven't been released yet, but any insight would be really helpful.

 

[Headfoot]

Looked at any NVidia thread lately? Same thing or worse.

The thread crapping in all threads has gotten particularly blatant and consistent

 

[mohit9206]

Pro 450 has same tflops as hd 7750 and Pro 455 has same tflops as hd 7770. Say 20% faster due to architecture improvements. So maybe these could come to desktop as R7 450 and R7 450X for under $80 next year ?
Call it irrelevant as much as you want but its still going to be faster than kaby lake i3 integrated gpu and better driver support.

 

[nurturedhate]

Looked at any NVidia thread lately? Same thing or worse.

Then stop threadcrapping in threads.

 

[VirtualLarry]

Pro 450 has same tflops as hd 7750 and Pro 455 has same tflops as hd 7770. Say 20% faster due to architecture improvements. So maybe these could come to desktop as R7 450 and R7 450X for under $80 next year ?
Call it irrelevant as much as you want but its still going to be faster than kaby lake i3 integrated gpu and better driver support.

Make them single-slot, low-profile-capable, and optionally passive (if they can swing it, engineering-wise), and I think that they would have that niche market for HTPC cards / SFF business workstations to themselves.

I would much rather have Polaris (even cut-down), than Fermi (GT610/620) or Kepler (GT710/720/730).

 

[krumme]

Make them single-slot, low-profile-capable, and optionally passive (if they can swing it, engineering-wise), and I think that they would have that niche market for HTPC cards / SFF business workstations to themselves.

I would much rather have Polaris (even cut-down), than Fermi (GT610/620) or Kepler (GT710/720/730).

There is tons of profit in mobile and especially a mac for a 35w part. Can a htpc market compete for that? I dont think so. Its not worth going for.

 

[Headfoot]

Make them single-slot, low-profile-capable, and optionally passive (if they can swing it, engineering-wise), and I think that they would have that niche market for HTPC cards / SFF business workstations to themselves.

I would much rather have Polaris (even cut-down), than Fermi (GT610/620) or Kepler (GT710/720/730).

An inexpensive, modern half height single slot HTPC GPU would be a game changer