Vega/Navi Rumors (Updated)

[Krteq]

Just checked a ComputeBench results DB and this is still the same result posted in March :/

for example - https://videocardz.com/67242/amd-vega-with-64-compute-units-spotted

//Freq -1000/1200MHz and a same driver version

 

[Bouowmx]

AMD vs NVIDIA in CompuBench is all over the place: AMD Radeon RX 480 vs NVIDIA GeForce GTX 1060 6 GB, and AMD Radeon R9 Fury vs NVIDIA GeForce GTX 980 Ti (no specific entry for Fury "X")

Hard to conclude gaming performance.

 

[Glo.]

If you want to look at compute performance between competitive GPUs don't look anywhere else than here:
https://wiki.blender.org/index.php/Dev:Source/Render/Cycles/OpenCL

This is comparison between NVidia CUDA and OpenCL(AMD) performance.

This is what happens when you optimize software for AMD. And do not tell me that here you get gimped performance on Nvidia hardware, because optimization for AMD. It is using CUDA, so there is no need for optimization.

But that is off-topic.

 

[EXCellR8]

"OpenCL works fine on NVIDIA cards, but performance is reasonably slower (up to 2x slowdown) compared to CUDA, so it doesn't really worth using OpenCL on NVIDIA cards at this moment."

i think they mean "isn't really worth using" but engrish

 

[Janooo]

If you want to look at compute performance between competitive GPUs don't look anywhere else than here:
https://wiki.blender.org/index.php/Dev:Source/Render/Cycles/OpenCL

This is comparison between NVidia CUDA and OpenCL(AMD) performance.

This is what happens when you optimize software for AMD. And do not tell me that here you get gimped performance on Nvidia hardware, because optimization for AMD. It is using CUDA, so there is no need for optimization.

But that is off-topic.

Where does it say it's NVidia CUDA for NV card? I do not see it on the linked page.

 

[Crumpet]

Fantastic explanation of AMD's stock dropping;

I suggest you watch, but tl:dw is cost of production of Ryzen chips over 1 month of selling time, large console sales with low profit margin, and low profit margin on Polaris. Overall profit margin expected to drop another 1% next quarter as R5 and R3's have lower profit margins but Vega, APU's, Naples and OEM market expected to bring a tidy income Q3 through to next year. AMD actually predicted the amount of profit they would make incredibly accurately.

 

[railven]

Waiting to see what AMD delivers. I'm pretty much gonna have to pass on it regardless if it's better (NV secured me for another year). But hoping AMD brings something to the table to bring them back.

Also getting tired of the same posters with their asinine predictions. During my red army days these same posters I put a lot of stock into. Now I can see just how misleading they are.

With Prey releasing soon, I hope we get a few more morsels of info because some of these predictions are setting up another hype train for derailment.

 

[daxzy]

I actually think R9 Fury is a great architecture - with quirky oddities:

1. The 4GB HBM limitation made a questionable upgrade path from R9 290X.
2. AMD seems to have terribly underestimated Maxwell. If you kept Fury at 900 MHz, its perf/watt is actually really good (matching Polaris refresh).
3. AMD seemed to have real issues scaling core clock speed above 1 GHz. Design problem?

I'll probably add one to my hardware collection when it goes for peanuts.

 

[Paratus]

Waiting to see what AMD delivers. I'm pretty much gonna have to pass on it regardless if it's better (NV secured me for another year). But hoping AMD brings something to the table to bring them back.

Also getting tired of the same posters with their asinine predictions. During my red army days these same posters I put a lot of stock into. Now I can see just how misleading they are.

With Prey releasing soon, I hope we get a few more morsels of info because some of these predictions are setting up another hype train for derailment.

Which predictions are you complaining about?

The one where Vega is barely faster than the 1070, 1080, or the 1080Ti?

Or is it the one where if Vega isn't faster than Volta it is a complete and utter failure?

 

[Paratus]

I actually think R9 Fury is a great architecture - with quirky oddities:

1. The 4GB HBM limitation made a questionable upgrade path from R9 290X.
2. AMD seems to have terribly underestimated Maxwell. If you kept Fury at 900 MHz, its perf/watt is actually really good (matching Polaris refresh).
3. AMD seemed to have real issues scaling core clock speed above 1 GHz. Design problem?

I'll probably add one to my hardware collection when it goes for peanuts.

I'll probably get whatever has the best price/performance metric in the $500 + $200 around Christmas this year.

I'm hopeful it will be Vega as I prefer AMD. Ive gotten years of decent performance out of each card. If I have to it will be NV though.

 

[Hitman928]

Fantastic explanation of AMD's stock dropping;

I suggest you watch, but tl:dw is cost of production of Ryzen chips over 1 month of selling time, large console sales with low profit margin, and low profit margin on Polaris. Overall profit margin expected to drop another 1% next quarter as R5 and R3's have lower profit margins but Vega, APU's, Naples and OEM market expected to bring a tidy income Q3 through to next year. AMD actually predicted the amount of profit they would make incredibly accurately.

I don't have time to watch that video but:

Margins aren't expected to drop due to R5 and R3 sales, it's due to AMD expecting console ramp ups in the second quarter which are very low margin. R5 and R3 are lower margin than R7 for sure, but still should be higher than console margins and not the reason for dropping their margins from this past quarter. AMD also said that they will be selling prior generation (not Ryzen) chips moving forward as there is still a lot of inventory that they need to get through, though they specifically mentioned motherboards so it sounds like AMD is trying to keep partners happy. Not sure on the low profit margin on Polaris, I didn't see that in the transcript but maybe I just missed it.

I know people are upset about the huge negative reaction to AMD matching expectations in Q1, but it wasn't that which was the main cause of the sell-offs, it was their projection for Q2 as well as not wanting to go into some specifics as to what to expect from Naples launching. Investors really expected margins to see a significant bump with Ryzen fully in the market as well as Vega coming into play. They didn't expect lower guidance for margins or an ambiguous timeline for Naples revenue. Does that mean the crazy drop was deserved? Probably not, but it happens, especially with how much movement has been going on with AMD's stock in a pretty short time frame. If you're long on AMD, this shouldn't change much and if Naples can get some big time customers, cash could start rolling AMD's way with high margins, time will tell but they have positioned themselves to be a healthy company again should they execute their plan smoothly for the rest of the year.

 

[Gikaseixas]

Waiting to see what AMD delivers. I'm pretty much gonna have to pass on it regardless if it's better (NV secured me for another year). But hoping AMD brings something to the table to bring them back.

Also getting tired of the same posters with their asinine predictions. During my red army days these same posters I put a lot of stock into. Now I can see just how misleading they are.

With Prey releasing soon, I hope we get a few more morsels of info because some of these predictions are setting up another hype train for derailment.

Didn't you just upgraded from 1080 to 1080TI? I ask that because if Vega gives you the same performance increase over the TI why wouldn't you buy it? You proved that jumps like that are enough to make you upgrade.

Also, what predictions are you complaining about? Don't see you complaining about the equal amount of doom and gloom being posted, i wonder why....

 

[Crumpet]

I don't have time to watch that video but:

Margins aren't expected to drop due to R5 and R3 sales, it's due to AMD expecting console ramp ups in the second quarter which are very low margin. R5 and R3 are lower margin than R7 for sure, but still should be higher than console margins and not the reason for dropping their margins from this past quarter. AMD also said that they will be selling prior generation (not Ryzen) chips moving forward as there is still a lot of inventory that they need to get through, though they specifically mentioned motherboards so it sounds like AMD is trying to keep partners happy. Not sure on the low profit margin on Polaris, I didn't see that in the transcript but maybe I just missed it.

I know people are upset about the huge negative reaction to AMD matching expectations in Q1, but it wasn't that which was the main cause of the sell-offs, it was their projection for Q2 as well as not wanting to go into some specifics as to what to expect from Naples launching. Investors really expected margins to see a significant bump with Ryzen fully in the market as well as Vega coming into play. They didn't expect lower guidance for margins or an ambiguous timeline for Naples revenue. Does that mean the crazy drop was deserved? Probably not, but it happens, especially with how much movement has been going on with AMD's stock in a pretty short time frame. If you're long on AMD, this shouldn't change much and if Naples can get some big time customers, cash could start rolling AMD's way with high margins, time will tell but they have positioned themselves to be a healthy company again should they execute their plan smoothly for the rest of the year.

Well, considering the video explains in detail everything that Lisa Su said in her transcripts, I think it is to do with that. But yes seems like investors didn't like expected guudance profit margin to drop from 34 to 33% in q2

 

[railven]

Which predictions are you complaining about?

The one where Vega is barely faster than the 1070, 1080, or the 1080Ti?

Or is it the one where if Vega isn't faster than Volta it is a complete and utter failure?

Not allowed to call out members. Tons of terrible predictions. But being a former red ant, it's always the ones that balloon expectations only to burst at launch that leave me feeling sour.

Didn't you just upgraded from 1080 to 1080TI? I ask that because if Vega gives you the same performance increase over the TI why wouldn't you buy it? You proved that jumps like that are enough to make you upgrade.

Also, what predictions are you complaining about? Don't see you complaining about the equal amount of doom and gloom being posted, i wonder why....

If Vega launched before 1080 Ti and offered any improvements over GTX 1080, I'd have bought it. I normally upgrade every 9-12 months. Considering I'd have to switch to a FreeSync monitor (and I'd most likely want one equal to my current one in resolution/refresh rate/size), swapping sides only 2-3 months after buying a GTX 1080 Ti would be out of the question.

But next year is right around the corner, and if AMD can bring the product, I'd gladly bring the cash

 

[Glo.]

Where does it say it's NVidia CUDA for NV card? I do not see it on the linked page.

https://twitter.com/tonroosendaal/status/852103617742073857

It says here.

 

[Glo.]

Sorry for post under post, but just for laughs

 

[KompuKare]

3. AMD seemed to have real issues scaling core clock speed above 1 GHz. Design problem?

I wonder if the density (transistors/area) has something to do with that? That is, does a design able to hit higher speeds have to have at least some parts designed to be less dense so that a design able to clock faster would have to 'waste' some die space to reach those clocks.
And might power be equally affected by such decisions.
Designing a less dense design might cost a bit more in terms of dies per wafer, but if it means better clocks or perf/watts and hence is able to command a higher price it would make sense, especially if GF's wafers are cheaper than TSMC (which they should be if their process is worse).

 

[beginner99]

especially if GF's wafers are cheaper than TSMC

GF's wafers are by default cheaper for AMD at least due to WSA. More wafers bought means lower penalty. That's why I find it weird that console chips are made at TSMC...

 

[Glo.]

GF's wafers are by default cheaper for AMD at least due to WSA. More wafers bought means lower penalty. That's why I find it weird that console chips are made at TSMC...

For very simple reason 90% of industry is making in TSMC fabs.

They are cheaper, than GloFo.

 

[guskline]

Glo, thanks for the post. Regardless of the content, that video always gets me laughing. THANKS.

 

[Gikaseixas]

If Vega launched before 1080 Ti and offered any improvements over GTX 1080, I'd have bought it. I normally upgrade every 9-12 months. Considering I'd have to switch to a FreeSync monitor (and I'd most likely want one equal to my current one in resolution/refresh rate/size), swapping sides only 2-3 months after buying a GTX 1080 Ti would be out of the question.

But next year is right around the corner, and if AMD can bring the product, I'd gladly bring the cash

Fair enough I'm waiting to see what Vega brings to the table but if it is slower i will definitely go for a 1080TI. My new build (sig) needs a fast card that i can count on for the next couple of years.

 

[CatMerc]

At this point it all feels like this:

I made this a few days ago

Why are they comparing results from different OSes - Windows vs. Mac OS X?

Probably to create unnecessary Vega hype. Windows performance is far better, and 1080 Ti smashes the C1.

 

[nad-]

Seeing all these low clock Vega 687F:C3/687F:C1 benchmarks with varying performance brackets, I got the sneaking suspicion there's a 14nm Fiji dieshrink in the lower end of the Vega stack.

I'm probably completely wrong but if Pascal could do so well being mostly a Maxwell dieshrink at high clocks, Fiji at high clocks could do pretty well too, it'll probably alleviate some of the frontend bottleneck running at those clocks as well.
Unless the frontend runs its own clock independent of the core clock.

http://cdn.wccftech.com/wp-content/...R9-Fury_HBM_OC_1000-MHz_Unlocked-1140x883.png

Here's a non-X Fury on steroids, wonder how much the 1TB/s bandwidth OC helps the graphics score.

 

[Paratus]

Not allowed to call out members. Tons of terrible predictions. But being a former red ant, it's always the ones that balloon expectations only to burst at launch that leave me feeling sour.

Fair enough. We'll call that any prediction faster than a 1080 and I will pencil myself in as one of the asinine.

 

[Janooo]

https://twitter.com/tonroosendaal/status/852103617742073857

It says here.

Thanks.