New Macbook Pro with vega graphics coming soon

[Premal212]

Maybe, can imagine people returning them after 6 months for an "upgrade" either ways good luck.

Q-6

Yeah all the best with that, I hope they do honour that, Apple will have rocketed in my estimated if they allow such a thing.

Trade in value is pretty much peanuts, you rather shift it on eBay etc. If they allow for such a thing, I'll be going in a few months for an upgrade of sorts.

 

[swerve147]

So what is the Nvidia equivalent of this new Vega chip. Are we in GTX 1060 range?

 

[leman]

Yeah all the best with that, I hope they do honour that, Apple will have rocketed in my estimated if they allow such a thing.

I will report

 

[Queen6]

Yeah all the best with that, I hope they do honour that, Apple will have rocketed in my estimated if they allow such a thing.

Trade in value is pretty much peanuts, you rather shift it on eBay etc. If they allow for such a thing, I'll be going in a few months for an upgrade of sorts.

I'm sure that you can trade them in with Apple, at a rock bottom price mind

Q-6

 

[leman]

So what is the Nvidia equivalent of this new Vega chip. Are we in GTX 1060 range?

Of course not. That is a completely different category and its not realistic to fit it into Apple's thermal budget. But again, I'd guess that Vega 20 is on par or maybe even slightly above the 1050 Ti, while using less power.

 

[Queen6]

I will report

As stated another member already has a one time 90 day return concision, so they may in such cases, but I doubt it's a general trend.

Q-6

 

[koyoot]

4 pages and nobody posted the specs and features.

Vega GCN5 architecture.
20 CUs - full chip, with single stack of HBM2 with 192 GB/s bandwidth. 1280 Cores, with 1.3 GHz, totalling with 3.3 TFLOPs of power. Radeon Vega M GH/GTX 1060 Max-Q range performance in 35W TDP power envelope. GTX 1060 Max-Q is 80W TDP.
For comparison Radeon Pro 560X had 80 GB/s memory bandwidth, 1024 GCN cores, and 900 MHz core clock, with 1.8 TFLOPs of compute power.
Vega 16 has the same memory bandwidth, but 16 CUs(1024 GCN cores) and 1185 MHz core clock. It should be GTX 1050 performance level.
4 Shader Engines, 5 CUs per Shader, 8 ROPs per Shader engine, 32 ROPs total.

Of course not. That is a completely different category and its not realistic to fit it into Apple's thermal budget. But again, I'd guess that Vega 20 is on par or maybe even slightly above the 1050 Ti, while using less power.

Vega 20 will be slightly slower, than GTX 1060 Max-Q. But not that much.

 

[Thysanoptera]

From all I've seen from Vega, it appears to be very efficient at lower clocks. Its as you say, AMD has overlocked it way beyond its optimal point to compete with Pascal.

The problem then is that the performance suffers, the "Radeon Pro Vega" was in iMacs for a year, the Pro 56 having 3.5x CUs and higher clocks than 560x is about 3x faster while sipping 210W. So I just don't see how it can be an improvement after cutting it down to 20 CU and lowering the clocks to get down to 35W, to me it looks like another i9 case. The Vega 16 will be on par with RX560x while the Vega 20 theoretically 25% faster will be limited to about 15% faster due to having only 5W of thermal headroom to play with.

I have RX580 and Vega FE at home and honestly FE is such disappointment, it is like I keep wondering where does all this power go. Unless AMD did some magic and vastly improved efficiency while scaling it down - just in such case they would announce a new architecture name and flood the marked with stuff that competes with 1060.

Thanks for the info about the upgrade, I hope it is true and applies also to BTO, that's what Apple should do for those who bought the short lived iteration of 2018 MBP, even if it won't bring any significant improvements but rather for resale value down the line - because of the badge.

 

[koyoot]

The problem then is that the performance suffers, the "Radeon Pro Vega" was in iMacs for a year, the Pro 56 having 3.5x CUs and higher clocks than 560x is about 3x faster while sipping 210W. So I just don't see how it can be an improvement after cutting it down to 20 CU and lowering the clocks to get down to 35W, to me it looks like another i9 case. The Vega 16 will be on par with RX560x while the Vega 20 theoretically 25% faster will be limited to about 15% faster due to having only 5W of thermal headroom to play with.

The GPU has 1280 GCN5 cores, 1.3 GHz core clock, 192 GB/s bandwidth and 35W TDP.

Vega 16 will be much faster than Radeon Pro 560X, because it has more bandwidth and much higher core clocks(560X 900 MHz, 80 GB/s vs 1185 MHz, 192 GB/s).

Praise the circuit design and HBM2 efficiency.

 

[leman]

So I just don't see how it can be an improvement after cutting it down to 20 CU and lowering the clocks to get down to 35W, to me it looks like another i9 case. The Vega 16 will be on par with RX560x while the Vega 20 theoretically 25% faster will be limited to about 15% faster due to having only 5W of thermal headroom to play with.

I think you are being a bit pessimistic here. The Vega 11 in the Ryzen APU is already very close to the Pro 560... and that is an integrated part without HBM2, and with only 11 CUs.

Vega 20 has 20 CPUs, and it has ridiculously fast RAM. In addition, HBM2 power consumption is very low compared to the GDDR5. I think power consumption is the main reason why Apple was clocking the video RAM rather conservatively on the MBP, and that was a big limiter of Polaris chip performance. With HBM2, this limit is simply gone, the GPU can really flex its muscles and has more thermal headroom for itself as well.

The problem then is that the performance suffers, the "Radeon Pro Vega" was in iMacs for a year, the Pro 56 having 3.5x CUs and higher clocks than 560x is about 3x faster while sipping 210W.

Where did you get this data? I haven't seen much reliable benchmarking done on the iMac Pro graphics... according to the 3dmark tests from notebookcheck, the Vega 56 is between 4-4.5 times faster than the Pro 560. I couldn't find any information on power consumption of these parts. The 210 Watts is the TDP of the RX-branded Vega 56, but the Vega in the iMac is at least 30% underclocked...

 

[koyoot]

Vega 20 has 20 CPUs, and it has ridiculously fast RAM. In addition, HBM2 power consumption is very low compared to the GDDR5. I think power consumption is the main reason why Apple was clocking the video RAM rather conservatively on the MBP, and that was a big limiter of Polaris chip performance. With HBM2, this limit is simply gone, the GPU can really flex its muscles and has more thermal headroom for itself as well.

The reason why Vega has more thermal headroom(higher clocks) is very simple. It is not due to HBM2 memory, but AMD simply burning a lot xTors for circuit design, and increasing the pipeline. The GPU is bigger than 1536 Polaris/Vega GPU from Kaby Lake-G, but has smaller amount of cores. The cores are bigger, there are more features, and circuit is different. That is main reason why Vega M GH in that canyon NUC from Intel has 55W TDP, while having 1190 MHz core clock, and Vega 20 has 1.3 GHz core clock, with 1280 GCN cores, in 35W TDP envelope.

Like I have said in previous post: Praise the Circuit design.

 

[fokmik]

so still 14nm..so i guess this thing will throttle like hell. Another useless option like i9 ...
Change the cooling and after that offer powerful upgrades Apple

 

[Thysanoptera]

I think you are being a bit pessimistic here.

After being burned multiple times by the hype surrounding new AMD products and their breakthrough technologies, yeah, I'm pessimistic.

Where did you get this data?

From notebookcheck also, LOL. From their iMac Pro review. They didn't list the 3dmark for 560x, but compared Luxmark to 555, so 560x is about 30% faster than 555 making the Pro 56 3x faster than 560x. And they measured the power, 50 to 90 W at idle, 260W in Valley. Not very scientific, it is probably less than 210, there would be no way they could scale it down to 35W.

Like I have said in previous post: Praise the Circuit design.

Is this circuit design any different than desktop Vegas or the Vega Pro in iMac Pro?

 

[koyoot]

Is this circuit design any different than desktop Vegas or the Vega Pro in iMac Pro?

Every GPU has different Circuit design... Vega Mobile was, unlike desktop Vega's, designed with Mobile First philosophy in mind.

 

[leman]

From notebookcheck also, LOL. From their iMac Pro review. They didn't list the 3dmark for 560x, but compared Luxmark to 555, so 560x is about 30% faster than 555 making the Pro 56 3x faster than 560x.

They have scores for Pro 560 and since the differences between 460, 560 and 560X is negligible anyway, I just took 560's scores.

And they measured the power, 50 to 90 W at idle, 260W in Valley. Not very scientific, it is probably less than 210, there would be no way they could scale it down to 35W.

Ah, just found it. Well, you also need to take into account that that machine has a CPU with the TDP of 140 Watt. And besides, notebookcheck said that they measured 150W total power consumption when running 3rdmark06. Its really difficult to put an estimate here, but the GPU will be probably somewhere around 120-150Watts... Scaling down to 35W, especially when one considers non-linear relationship of clock and power + binning + minor efficiency improvements is very possible I think.

But of course, you are right that one should get too exited prematurely. We will see
[doublepost=1540998390][/doublepost]

Every GPU has different Circuit design... Vega Mobile was, unlike desktop Vega's, designed with Mobile First philosophy in mind.

Wait, maybe I am misunderstanding something, but aren't these all same chips anyway just binned/clocked differently? I know that AMD does post processing (thinning) for Apple, but its ultimately still the same Polaris chip they use in any other desktop or laptop.

 

[koyoot]

Wait, maybe I am misunderstanding something, but aren't these all same chips anyway just binned/clocked differently? I know that AMD does post processing (thinning) for Apple, but its ultimately still the same Polaris chip they use in any other desktop or laptop.

No they are not the same chips. Vega 10 has 64 CUs, two memory controllers, for two stacks of HBM2, and was designed for Desktop use. Vega 12 has 20 CUs, single memory controller, for single stack of HBM2, and designed with Mobile First philosophy.

Vega is also different architecture than Polaris, because it has Tile Based Rasterization, Rapid Packed Math(FP16x2 of FP32), High Bandwidth cache Controller, and ROPs are clients of L2 cache instead of Memory Controllers.

Vega burns a lot of transistors to increase the core clocks, compared to Polaris, and that results in higher clock speed in lower thermal envelope. There is a lot of design changes in Vega, compared to Polaris, and they are not the same GPU family.

Remember: Vega is GFX9 family, and Polaris is GFX8.

Vega 64 is limited on the front end. Vega 12 appears to not be limited on front end, and I was never so excited to actually see how it translates into gaming performance. Untill AMD will implement any form of Operand Reuse Cache, high core count AMD GPUs will always be front end limited. 1280 will not be front end limited, because it has 5 CUs in each of 4 Shader Engines. Compare this to Polaris 11: 2 Shader Engines, 8 CUs in each of them. Vega is much more balanced design.

 

[_Kiki_]

4 pages and nobody posted the specs and features.

Vega GCN5 architecture.
20 CUs - full chip, with single stack of HBM2 with 192 GB/s bandwidth. 1280 Cores, with 1.3 GHz, totalling with 3.3 TFLOPs of power. Radeon Vega M GH/GTX 1060 Max-Q range performance in 35W TDP power envelope. GTX 1060 Max-Q is 80W TDP.
For comparison Radeon Pro 560X had 80 GB/s memory bandwidth, 1024 GCN cores, and 900 MHz core clock, with 1.8 TFLOPs of compute power.
Vega 16 has the same memory bandwidth, but 16 CUs(1024 GCN cores) and 1185 MHz core clock. It should be GTX 1050 performance level.
4 Shader Engines, 5 CUs per Shader, 8 ROPs per Shader engine, 32 ROPs total.

Vega 20 will be slightly slower, than GTX 1060 Max-Q. But not that much.

Vega M GH 100W TDP (for whole IGP) is equal GTX 1060 Max-Q so no way Apple will fit into MacBook Pro, but AMD Radeon RX Vega M GL TDP 65W so it's possible to fit into 15", but is equal GTX 1050 Ti Max-Q, so probably VEGA M GL = VEGA 20, VEGA 16 "cutted" version

 

[leman]

No they are not the same chips. Vega 10 has 64 CUs, two memory controllers, for two stacks of HBM2, and was designed for Desktop use. Vega 12 has 20 CUs, single memory controller, for single stack of HBM2, and designed with Mobile First philosophy.

I see what you mean. Well, Vega is certainly a very flexible architecture (being designed as such from the start). Still, I'd think that Vega 12 is more or less Vega 10 cut in half, with some minor improvements here and there. I wouldn't expect any major redesign.

Vega burns a lot of transistors to increase the core clocks, compared to Polaris, and that results in higher clock speed in lower thermal envelope. There is a lot of design changes in Vega, compared to Polaris, and they are not the same GPU family.

Well, that is clear, I didn't mean to suggest that Polaris and Vega are the same design. Sorry if my wording was a bit unclear.

 

[Thysanoptera]

And besides, notebookcheck said that they measured 150W total power consumption when running 3rdmark06. Its really difficult to put an estimate here, but the GPU will be probably somewhere around 120-150Watts...

The 150W was in 3dMark06 in Directx9, I still believe it is around 200W while running something modern.

No they are not the same chips. Vega 10 has 64 CUs, two memory controllers, for two stacks of HBM2, and was designed for Desktop use. Vega 12 has 20 CUs, single memory controller, for single stack of HBM2, and designed with Mobile First philosophy.

This is the very definition of binning the same sillicon. What about that what you wrote before, about transistors, pipelines, size of cores, circuit: are there any differences between desktop and mobile Vega. Any differences between the Vega Pro 56 in iMac pro and the Vega 20?

 

[leman]

Vega M GH 100W TDP (for whole IGP) is equal GTX 1060 Max-Q so no way Apple will fit into MacBook Pro, but AMD Radeon RX Vega M GL TDP 65W so it's possible to fit into 15", but is equal GTX 1050 Ti Max-Q, so probably VEGA M GL = VEGA 20, VEGA 16 "cutted" version

a) Vega M is not really Vega architecture (its Polaris with memory controller cut out from Vega)
b) it also includes the CPU

Vega Pro 16 and 20 are based on a new chip from Vega family.
[doublepost=1540999436][/doublepost]

This is the very definition of binning the same sillicon.

I would disagree with your definition of binning

What about that what you wrote before, about transistors, pipelines, size of cores, circuit: are there any differences between desktop and mobile Vega. Any differences between the Vega Pro 56 in iMac pro and the Vega 20?

Yes, there are. AMD claims that Vega Pro 16/20 comes with "optimised geometry engine" compared to the desktop Vega. The chip itself is the Vega Codename 12 that was leaked some time ago, while desktop 56/64 are Vega 10. All these numbers are rather confusing.

 

[koyoot]

Vega M GH 100W TDP (for whole IGP) is equal GTX 1060 Max-Q so no way Apple will fit into MacBook Pro, but AMD Radeon RX Vega M GL TDP 65W so it's possible to fit into 15", but is equal GTX 1050 Ti Max-Q, so probably VEGA M GL = VEGA 20, VEGA 16 "cutted" version

No... Its NOT the same chip...

Vega M GL is cut down version of Polaris architercture, with Vega features(Draw Stream Binning Rasterizer, and High Bandwidth Cache Controller). It belongs to GFX 8 family. Vega 12, Vega 20 is full GFX 9, GCN5 architecture, with all of its features, including Rapid Packed Math.

This is the very definition of binning the same sillicon. What about that what you wrote before, about transistors, pipelines, size of cores, circuit: are there any differences between desktop and mobile Vega. Any differences between the Vega Pro 56 in iMac pro and the Vega 20?

No there arent any...

Vega 56 is 500 mm2 die sized monster. Vega 20 is under 180 mm2. HOW THEY CAN BE THE SAME CHIP?!

It is completely different silicon. Completely different Circuit.

 

[_Kiki_]

a) Vega M is not really Vega architecture (its Polaris with memory controller cut out from Vega)
b) it also includes the CPU

Vega Pro 16 and 20 are based on a new chip from Vega family.
[doublepost=1540999436][/doublepost]

I would disagree with your definition of binning



Yes, there are. AMD claims that Vega Pro 16/20 comes with "optimised geometry engine" compared to the desktop Vega. The chip itself is the Vega 12 that was leaked some time ago.

I said IGP

 

[koyoot]

I said IGP

No. Whole package of Kaby Lake-G has 100W TDP, and CPU has 45W. The rest is for the GPU. 55W is for vega M GH.

 

[_Kiki_]

No. Whole package of Kaby Lake-G has 100W TDP, and CPU has 45W. The rest is for the GPU. 55W is for vega M GH.

I know this, integrated graphics is always in CPU, whole chip will draw 100W, no way to fit to MacBook Pro 15 with few grams copper heatsink, also it's even to USB-C power limit

 

[leman]

I know this, integrated graphics is always in CPU, whole chip will draw 100W, no way to fit to MacBook Pro 15 with few grams copper heatsink, also it's even to USB-C power limit

I don't really see how this is relevant at all given that the MBP is not going to use Kaby Lake G or Vega M...