Vega/Navi Rumors (Updated)

[zinfamous]

16GB HBM2 with 512GB/s bandwidth. Right.
Not much effort went into this rumor. D-

Also pretty sure that AMD pressers showed Vega 11 to be the big Vega, and Vega 10 to be small Vega (reverse of Polaris naming scheme)

 

[tviceman]

My money is on that the increase in efficiency is basically due to HBM, like on Fury, and maybe some due to a slight lowering of clock speed. That and as you increase die size less of the power budget as a percentage goes to non-scalable uncore and other fixed power pieces.

I'd be pleasantly surprised if there is an increase in power efficiency due to even further architectural improvement but I wouldn't bet on it myself.

This is probably spot on. The perf/w improvements full Fiji had over full Tahiti were almost entirely due to HBM power savings and Fiji came more than a year after Tahiti. People better start tempering expectations.

 

[zinfamous]

There are rumors that Aliens built Egyptian pyramids. Using your logic, the rumors confirm that.

You got it all wrong. The pyramids were built to store grain.

 

[Mondozei]

This is probably spot on. The perf/w improvements full Fiji had over full Tahiti were almost entirely due to HBM power savings and Fiji came more than a year after Tahiti. People better start tempering expectations.

To lazily extrapolate from past trend to the future is... lazy. At this point we know nothing.

 

[Headfoot]

To lazily extrapolate from past trend to the future is... lazy. At this point we know nothing.

I keep seeing this "We know nothing." I had a lengthy post in which I nailed the polaris price within $20 based on what we knew when I was told "we know nothing." So wrong again here.

1) We know GPUs take a long time to design, longer than 6 months, thus we can conclude Vega coming out 6 months after Polaris will not have silicon changes based on what has happened since Polaris
2) We know it has HBM and we know from Fury HBM has significant power/perf implications
3) We know what new features Polaris has, and since we only have 6 months more time to build new features and since we know building new features takes time and some of that time is already used by working in HBM and a bigger die, that it can't possibly be massively different from Polaris unless they intentionally sandbagged Polaris to make Vega look better. Which is highly unlikely.

Sure we dont know "everything" but we do know quite a bit. Objectively more than "nothing"

The likelihood Vega is drastically different from Polaris is vanishingly small.

 

[ConsoleLover]

I think late Q1(March) will be Vega 10 and probably late Q2(June) vega 11

 

[Maverick177]

Fit in perfectly for a complete Zenga upgrade .

 

[Enigmoid]

The biggest takeaway I have is this:
If AMD is planning on replacing the Radeon 470/480 not even a year after I think they may have something special with Vega's architetcture. We know Vega will be on the same 14nm process, so if this is indeed true it sounds like Vega may bring another good step down in power efficiency. Why else replace a fairly new SKU on the same process unless the uArch is improved?

I've seen plenty of threads on reddit where an undervolted (maintains stock frequency) 480 matches a 1060 in power efficiency even in DX 11 titles so I bet AMD is really focusing on optimizing around this metric in both uArch and software/firmware tuning.

Other than that I would take the Vega 10 specs with a grain of salt, something doesn't seem right about those memory numbers, OTOH it is a good sign we are hearing Q1 now instead of H1. I imagine a lot of buyers out there would like to try something new with a Zen/Vega combo build!

AMD will not be replacing Polaris 10 and 11. Vega will be an extension to the product stack, slotting in above Polaris - Vega will replace Polaris as the *70/*80/*90 level cards. Polaris will continue to be sold on the lower-middle end.

We will see the same thing as we did with Pitcarin, with Polaris being sold and rebranded several times.

 

[swilli89]

I keep seeing this "We know nothing." I had a lengthy post in which I nailed the polaris price within $20 based on what we knew when I was told "we know nothing." So wrong again here.

1) We know GPUs take a long time to design, longer than 6 months, thus we can conclude Vega coming out 6 months after Polaris will not have silicon changes based on what has happened since Polaris
2) We know it has HBM and we know from Fury HBM has significant power/perf implications
3) We know what new features Polaris has, and since we only have 6 months more time to build new features and since we know building new features takes time and some of that time is already used by working in HBM and a bigger die, that it can't possibly be massively different from Polaris unless they intentionally sandbagged Polaris to make Vega look better. Which is highly unlikely.

Sure we dont know "everything" but we do know quite a bit. Objectively more than "nothing"

The likelihood Vega is drastically different from Polaris is vanishingly small.

The precise uArch differences Vega may or may not have compared to Polaris are completely immaterial. What matters of course, is how it performs compared to the competition.

How is it possible to discern HBM's power savings when we do not have a Fiji without HBM to compare to? Wasn't an engineer quoted as saying it was around 10-15 watts?

 

[tviceman]

To lazily extrapolate from past trend to the future is... lazy. At this point we know nothing.

Dream those big dreams! I hope they come true; it would be awesome to have Nvidia actually lose out for a change.

 

[4K_shmoorK]

See, I don't think AMD is 'losing' at all, its just that they're image and brand could benefit from a well timed win. Secretive specifications, limited marketing, unclear launch windows, etc.

If I were head of AMD Graphics division I would hire as many Nvidia/Ex-Nvidia marketing directors as I could. If you've got a solid lineup, TELL people. Is there even a definitive ad campaign for Polaris? The company needs to go for gold instead of prematurely conceding to silver before the race even starts.

 

[Mondozei]

Dream those big dreams! I hope they come true; it would be awesome to have Nvidia actually lose out for a change.

I'm not wildly speculating, unlike you. I'm merely stating facts that it is lazy and sloppy to assume past patterns will continue in the future. That doesn't translate into a prediction, which if you had a scintilla of reading comprehension, would have noted that my post had none of.

You post low quality content.

 

[AtenRa]

I want to see the new NANO, if at 28nm they managed to have that performance at 175W TDP at 14nm im expecting higher performance than GTX1080 (DX-12/Vulkan)

 

[Glo.]

I keep seeing this "We know nothing." I had a lengthy post in which I nailed the polaris price within $20 based on what we knew when I was told "we know nothing." So wrong again here.

1) We know GPUs take a long time to design, longer than 6 months, thus we can conclude Vega coming out 6 months after Polaris will not have silicon changes based on what has happened since Polaris
2) We know it has HBM and we know from Fury HBM has significant power/perf implications
3) We know what new features Polaris has, and since we only have 6 months more time to build new features and since we know building new features takes time and some of that time is already used by working in HBM and a bigger die, that it can't possibly be massively different from Polaris unless they intentionally sandbagged Polaris to make Vega look better. Which is highly unlikely.

Sure we dont know "everything" but we do know quite a bit. Objectively more than "nothing"

The likelihood Vega is drastically different from Polaris is vanishingly small.

Actually, we know few things. You just don't know them.

Vega and Raven Ridge APU iGPU is the same family as Greenland. All of them are Graphics IPv9. Why does Raven Ridge GPU be different? Because it needs to have 2048 bit memory controller. Reminds you of anything?

What does mean Graphics IPv9? It can mean in the context of previous generations of Graphics IPv8, from which Polaris architecture comes from, that Vega can have for example new rasterizing technology. It can have next generation schedulers, for example. Both of those things can bring next generation Graphics IP number, and differentiate it from previous generation of GPUs.

To the degree that according to Linux AMD Drivers "alchemist", it appears that Vega would require layer of abstraction in drivers to keep that architecture compatible with previous generations of GPUs. That was the reason why it cannot go into PS4.5. It had to be binary compatible with previous generations of GPUs. And Polaris is binary compatible with previous gen of GPUs, because it is the same family: Graphics IPv9.

Look here: http://videocardz.com/62250/amd-vega10-and-vega11-gpus-spotted-in-opencl-driver

SI: TAHITI
CI / GFX7: MILOS, KRYPTOS, HAWAII, NEVIS, PENNAR, BONAIRE, Kabini
VI / GFX8: ICELAND, TONGA, CARRIZO, BERMUDA, racerx, FIJI
GFX81: AMUR, STONEY, ELLESMERE, DERECHO
GFX9: GREENLAND, RAVEN1X, VEGA10, VEGA11

Greenland, Raven Ridge iGPU, and Vega GPUs are the same family. Ellesmere is evolved Tonga/Fiji architecture. Vega is quite different. We know also that clock for clock, Polaris is around 20% faster than Tahiti, for example, and 5-10% than Tonga. How will fare Vega compared to Fiji, for example, on this front?

Thirdly, we don't know on which process Vega architecture will be made on. My bet is on 16 nm FF+, because AMD has good production chain, and worked with TSMC and Amkor on packaging, and producing the interposer GPUs. Nvidia was able to get 70% increase in efficiency on this process. Tesla P4 with 2560 CC's has 75W TDP and gives 5.5 TFLOPs of compute power. GTX 980 Ti with 2816 CC's consumed around 225W of power under load. So P4 is a 33% of power consumption compared to 28 nm process GPU with similar compute power. And all of this is done just by process shrink itself. How will fare on this front big Vega, with similar core count to Fiji?

Fourth: If you have 1.5 GHz GPU, and 1 TB/s memory bandwidth, you do not design something like this without any reason. The GPU must have enormous throughput, otherwise resources will be completely and utterly wasted.

Lastly. If Polaris was ported by AMD, from the development process of PS4.5 GPU, and Vega was from the ground up next generation GPU architecture - you get the picture. Vega is not similar to Polaris in architectures.

 

[Glo.]

I want to see the new NANO, if at 28nm they managed to have that performance at 175W TDP at 14nm im expecting higher performance than GTX1080 (DX-12/Vulkan)

If you want Nano form factor - most likely all of HBM GPUs will have that form factor. Apart from those that are water cooled .

 

[AtenRa]

If you want Nano form factor - most likely all of HBM GPUs will have that form factor. Apart from those that are water cooled .

I dont only want the form factor, I want the 175W TDP at 14nm + at least HBM2 8GB .

 

[tviceman]

I dont only want the form factor, I want the 175W TDP at 14nm + at least HBM2 8GB .

Nano wasn't a special achievement in performance. AMD took full Fury X, reduced clocks and voltage by 5-10%, and saved 20% in power. Both AMD and Nvidia do the very same thing in all their mobile parts.

Also, HBM2 is only worth paying for if the overall GPU performance follows. Fury X had amazing next gen ram, but it was slower than GM200 and didn't overclock anywhere close to GM200 either. Paper specs are good for nothing.

 

[tviceman]

I'm not wildly speculating, unlike you. I'm merely stating facts that it is lazy and sloppy to assume past patterns will continue in the future. That doesn't translate into a prediction, which if you had a scintilla of reading comprehension, would have noted that my post had none of.

You post low quality content.

Fool me once AMD, shame on you. Fool me twice AMD, shame on me. Fool me 3, 4, 5.... Oh wait you're right. I won't let AMD's past performance dictate my expectations.

I'll keep it real and grounded, including all relevant data such as R&D, past patterns, release cycles, previous claims vs. real world measures, and current comparisons. You keep it.... however you want.

 

[Glo.]

Nano wasn't a special achievement in performance. AMD took full Fury X, reduced clocks and voltage by 5-10%, and saved 20% in power. Both AMD and Nvidia do the very same thing in all their mobile parts.

Also, HBM2 is only worth paying for if the overall GPU performance follows. Fury X had amazing next gen ram, but it was slower than GM200 and didn't overclock anywhere close to GM200 either. Paper specs are good for nothing.

So why currently Fiji is faster than GM200? Especially in next gen APIs.

What shows true performance of particular GPUs?

 

[crisium]

We will see. 16GB at 512GB/s doesn't seem right, as other posters have said.

 

[Arachnotronic]

If this VideoCardz information is correct then this means that AMD is keeping up a roadmap on its website that it knows to be false (it says Navi in 2018). If that roadmap isn't updated by the next time AMD talks to investors (mid-Oct), then we know that either VideoCardz is full of it or that AMD is illegally lying to its stockholders.

I don't think AMD would so blatantly break the law, so my guess is that this report is partially -- or totally -- false.

 

[AtenRa]

Nano wasn't a special achievement in performance. AMD took full Fury X, reduced clocks and voltage by 5-10%, and saved 20% in power. Both AMD and Nvidia do the very same thing in all their mobile parts.

Also, HBM2 is only worth paying for if the overall GPU performance follows. Fury X had amazing next gen ram, but it was slower than GM200 and didn't overclock anywhere close to GM200 either. Paper specs are good for nothing.

Fury Nano is as fast or faster than GTX 980Ti in DX-12 games (1440p and higher) at 175W TDP vs 250W TDP for vanila GTX 980Ti 1000MHz base/1075MHz turbo (according to NVIDIA site). At a card size smaller than even 14/16nm GTX 1060/RX 480.

If VEGA 10 does have 4096 shaders (64CUs), that will be a small die at around ~330mm2. This one will only compete against GTX 1080. A 430-450mm2 die with HBM2 at 175W TDP could be the next Nano and will be able to compete against TITAN X Pascal.

 

[Det0x]

If this VideoCardz information is correct then this means that AMD is keeping up a roadmap on its website that it knows to be false (it says Navi in 2018). If that roadmap isn't updated by the next time AMD talks to investors (mid-Oct), then we know that either VideoCardz is full of it or that AMD is illegally lying to its stockholders.

I don't think AMD would so blatantly break the law, so my guess is that this report is partially -- or totally -- false.

AMD has plans to ship Navi in 2019 according to the leaked server roadmap. Now you might believe that Navi has been delayed but i don’t find that to be the case. This roadmap is specific to the server platform and these products have the tendency to arrive later as compared to consumer parts. Hence, Navi could still be a planned 2018 launch on the consumer side but could miss that time frame on the server side of things.

 

[Sweepr]

Fury Nano is as fast or faster than GTX 980Ti in DX-12 games (1440p and higher)

Slower overall and doesn't overclock nearly as much as any GTX 980 Ti.

at 175W TDP vs 250W TDP for vanila GTX 980Ti 1000MHz base/1075MHz turbo (according to NVIDIA site).

Interesting. Now let's check real world results:

https://tpucdn.com/reviews/MSI/GTX_1070_Gaming_Z/images/power_maximum.png

Minor 7-11W difference.

If VEGA 10 does have 4096 shaders (64CUs), that will be a small die at around ~330mm2.

42% larger die for nearly 80% more CUs than Polaris 10? Sure.

This one will only compete against GTX 1080.

Remains to be seen if it will be able to match GP104 outside of Gaming Evolved titles - and rumor says 225W TDP, higher than a GTX 1080 from May 2016.

A 430-450mm2 die with HBM2 at 175W TDP could be the next Nano and will be able to compete against TITAN X Pascal.

Unless they pull a Maxwell-like improvement at the same process, I think you're a (tiny) bit optimistic with 2.4x RX 480 performance (163W @ TPU) at 175W TDP. And then again, the competition is a moving target, it has a lot of time to come up with an improved 3840 SPs GP102 card. That's if Volta doesn't arrive sooner than we expect.

 

[maddie]

We will see. 16GB at 512GB/s doesn't seem right, as other posters have said.

I've seen other posters here claiming this. Why is this a problem?

If that bandwidth [512GB/s] is sufficient then we would expect the memory to clock at the lowest speed needed for power savings. Seeing that 8GB stacks are not available, then 4 x 4GB appears to be the only way to get 16GB onboard. Just run the memory at [1.0 Gbps speed/pin] to satisfy the bandwidth needs and minimize power consumption.

We want 16GB memory onboard.
4 x 4GB stacks is the only way at present to achieve this.
We do not need more than 512GB/s to extract maximum performance.

Answer is 4 x 4GB HBM2 stacks operating at 1.0 Gbps/pin