Vega/Navi Rumors (Updated)

[Crumpet]

Probably fake but doing the rounds (can't find it in the search, don't know if results can be removed or not)

At least it's a tasty score this time.

From post #1172
http://forums.guru3d.com/showthread.php?t=408618&page=47

 

[Riek]

Well that result would be more believable and more in-line with expectations.

It is also clocked @ 1600MHz which is also more in-line with the expected clock of VEGA.

Doesn't mean that number is more or less valid than the previous one... Just that I prefer to believe these numbers over the previous ones .

 

[CatMerc]

Probably fake but doing the rounds

At least it's a tasty score this time.

No link posted to it. Could easily be inspect element.
It does pass the sniff test, but nothing more than that.

 

[Crumpet]

No link posted to it. Could easily be inspect element.
It does pass the sniff test, but nothing more than that.

Just edited my post with the original forum post, a little bit more info on there but nothing to verify it.

 

[Paratus]

What does a 1080 TI get in that bench?

 

[Det0x]

Probably fake but doing the rounds (can't find it in the search, don't know if results can be removed or not)

At least it's a tasty score this time.

From post #1172
http://forums.guru3d.com/showthread.php?t=408618&page=47

Wow you beat sweeper to the punch, as in iam sure he was just about to post these scores
Thanks.

What does a 1080 TI get in that bench?

http://www.guru3d.com/articles_pages/geforce_gtx_1080_ti_review,30.html

1080TI FE score as per guru3d.
10158/9521 = ~7% increase

 

[Det0x]

double post

 

[Glo.]

According to the guys from Guru3D forum, it is a leak, so there will not be a direct link to the score.

Lets say it is true. Vega is running circles around GTX 1080 Ti, with this score.

 

[Jackie60]

That Timespy score is above 1080ti (9502 Guru 3d). I don't know what settings were used but if true we could have a monster coming. That's with Ryzen at 1700 so no more than 4/4.1Ghz vs 5960X at 4.3Ghz for the 1080ti. If it's not fake then this suggests Vega plus 4ghz Ryzen smashing 1080ti with 5960x at 4.3ghz. That's assuming (a big assumption I know!) all else is equal. Graphics score alone for 1080ti is 9521 vs 10158 for Vega. I dearly hope its true and I can have an excuse to go AMD again and if it is I'm in for 2 assuming price no more than 1080ti.

 

[Glo.]

The funniest part is that none of current set of applications are going to use two most important hardware features, that increase performance of the GPU, until the apps are rewritten to use them: FP16, and Primitive Shaders.

FP16 is reported in SiSoft Sandra as 1:1 ratio, because it has not been rewritten. So bare this in mind, that we still may have not seen the end performance of the GPUs...

 

[raghu78]

We have seen numerous times such faked results before launch turn out completely false. I would advise people to not get carried away by the rumours/fake benches/hype. ATI/AMD has not released a flagship GPU which beats Nvidia's top GPU in more than a decade. So that alone should tell people to rein in their expectations.

 

[Crumpet]

We have seen numerous times such faked results before launch turn out completely false. I would advise people to not get carried away by the rumours/fake benches/hype. ATI/AMD has not released a flagship GPU which beats Nvidia's top GPU in more than a decade. So that alone should tell people to rein in their expectations.

Yeah that's why I bolded that it's probably fake.

But figured people would want to see it anyway, it certainly can't do more harm than circulating one that shows 1070 performance.

 

[Paratus]

We have seen numerous times such faked results before launch turn out completely false. I would advise people to not get carried away by the rumours/fake benches/hype. ATI/AMD has not released a flagship GPU which beats Nvidia's top GPU in more than a decade. So that alone should tell people to rein in their expectations.

My 7970GHZ begs to differ with your flagship assertion.

It's another reason I'll lean AMD if the basic performance is there. I tend to keep my cards for several years. I've been able to do that for over a decade because AMD performance scaling in future games allows me to.

Not a great business strategy compared to NV buy a new one every year or two but it keeps me a reasonably loyal customer.

 

[Jackie60]

It must be true, I'm going all in on AMD stock . If it is true AMD will have done an incredible job competing with both Intel and Nvidia with a fraction of either company's resources. Here's hoping.

 

[maddie]

For those interested.

Should dispel some of the more outrageous posts on the extremely high costs of HBM implementation. Vega should be lower cost than Fiji to manufacture as it has a smaller interposer and 2 stacks of 4-Hi HBM2 modules. With AMD's ability to work with lower margins, this bodes well for a competitive selling price.

Article is well worth reading.
http://electroiq.com/insights-from-leading-edge/category/uncategorized/page/6/

System Plus Consulting showed an interesting cost comparison between AMD graphics modules with DDR5 vs HBM memory as shown below.

 

[raghu78]

My 7970GHZ begs to differ with your flagship assertion.

It's another reason I'll lean AMD if the basic performance is there. I tend to keep my cards for several years. I've been able to do that for over a decade because AMD performance scaling in future games allows me to.

Not a great business strategy compared to NV buy a new one every year or two but it keeps me a reasonably loyal customer.

Between March 2012 when GTX 680 and Feb 2013 when GTX Titan (Kepler) launched the HD 7970 and GTX 680 were close competitors and equals. There was no definite winner by a significant margin (>=10%). Kepler started falling badly when Maxwell launched.

I agree with you that for roughly similar performance at launch( even if say 5% slower) I would also recommend a AMD GPU because of the console legacy and how well they are supported.

 

[AtenRa]

I think it would be expected for Vega to beat Pascal after the 13 months Nvidia's high end has been on the market.
If they don't beat a gtx1080ti by 20 or 30% the Vega is a failure IMHO, especially with Volta 6 months away from the June Vega release.

R7 1800X didnt beat the more than 12 months old Intel HEDT by 20-30% and still is one of the best product launches of the last 4-5 years.
AMD Vega doesnt need to be 20-30% faster than GTX1080Ti to be a good product.

 

[Paratus]

Between March 2012 when GTX 680 and Feb 2013 when GTX Titan (Kepler) launched the HD 7970 and GTX 680 were close competitors and equals. There was no definite winner by a significant margin (>=10%). Kepler started falling badly when Maxwell launched.

I agree with you that for roughly similar performance at launch( even if say 5% slower) I would also recommend a AMD GPU because of the console legacy and how well they are supported.

Splitting hairs, but fair enough. That being said within your decade time frame we had HD5870 vs GTX 285 for 6 months. The HD 5870 was about 30% faster.

Anyway this is OT. So I'll drop it.

 

[raghu78]

R7 1800X didnt beat the more than 12 months old Intel HEDT by 20-30% and still is one of the best product launches of the last 4-5 years. AMD Vega doesnt need to be 20-30% faster than GTX1080Ti to be a good product.

Though I agree that Vega need not be faster than 1080 Ti but should come at a compelling price/perf the problem for AMD is going to be the upcoming Volta launch. We do not want a repeat of the Maxwell generation when AMD went below 20% market share for the first time in ATI/AMD's history. Vega will have to contend with GV104 and GV102 shortly. Another thing is AMD have forgotten to do a complete stack launch in a short span of time after HD 7000 series. Nvidia meanwhile have perfected the launch of a complete stack and the Pascal series launch was amazing. Four GPUs launched in 5 months (GP104, GP106, GP102, GP107) covering USD 120 - USD 1200 price points was simply a lesson in outstanding execution that AMD should learn from. The other major problem is AMD's complete surrender of the high end notebook GPU market to Nvidia. It started with Kepler and was pretty much complete with Maxwell. Pascal reinforced that domination. AMD cannot be a successful GPU company with zero presence in high margin / high end notebook GPUs.

One of the main goals for AMD going forward with future GPU generation launches is to show up for the fight like they did with HD 4870 and HD 5870. These GPU generations are just roughly 2 years. So turning up a year behind the competition is just not going to work. AMD also need to get back to a decent presence in high end notebooks and that will only happen if AMD can compete with Nvidia in perf/watt. I hope Zen and Vega bring in enough cash to help AMD better invest in their product roadmap and allow them to come to market with competitive products on time against competition.

 

[Arachnotronic]

Vega will have to contend with GV104 and GV102 shortly. Another thing is AMD have forgotten to do a complete stack launch in a short span of time after HD 7000 series. Nvidia meanwhile have perfected the launch of a complete stack and the Pascal series launch was amazing. Four GPUs launched in 5 months (GP104, GP106, GP102, GP107) covering USD 120 - USD 1200 price points was simply a lesson in outstanding execution that AMD should learn from.

It's all about financial resources. AMD is trying to attack a wider range of markets/products with a smaller R&D budget than NVIDIA, so of course it won't be able to flawlessly roll out an entire stack like NVIDIA can and does.

The other major problem is AMD's complete surrender of the high end notebook GPU market to Nvidia. It started with Kepler and was pretty much complete with Maxwell. Pascal reinforced that domination. AMD cannot be a successful GPU company with zero presence in high margin / high end notebook GPUs.

It's all about that power efficiency that people keep pooh-pooh'ing Low efficiency is OK in desktop when you can just crank up your TBP (185W, anybody?), but in notebooks it's a death knell.

 

[AtenRa]

Though I agree that Vega need not be faster than 1080 Ti but should come at a compelling price/perf the problem for AMD is going to be the upcoming Volta launch. We do not want a repeat of the Maxwell generation when AMD went below 20% market share for the first time in ATI/AMD's history. Vega will have to contend with GV104 and GV102 shortly. Another thing is AMD have forgotten to do a complete stack launch in a short span of time after HD 7000 series. Nvidia meanwhile have perfected the launch of a complete stack and the Pascal series launch was amazing. Four GPUs launched in 5 months (GP104, GP106, GP102, GP107) covering USD 120 - USD 1200 price points was simply a lesson in outstanding execution that AMD should learn from. The other major problem is AMD's complete surrender of the high end notebook GPU market to Nvidia. It started with Kepler and was pretty much complete with Maxwell. Pascal reinforced that domination. AMD cannot be a successful GPU company with zero presence in high margin / high end notebook GPUs.

One of the main goals for AMD going forward with future GPU generation launches is to show up for the fight like they did with HD 4870 and HD 5870. These GPU generations are just roughly 2 years. So turning up a year behind the competition is just not going to work. AMD also need to get back to a decent presence in high end notebooks and that will only happen if AMD can compete with Nvidia in perf/watt. I hope Zen and Vega bring in enough cash to help AMD better invest in their product roadmap and allow them to come to market with competitive products on time against competition.

My prediction,

End of Q2 early Q3 (August) ,

8 Core Skylake @ $1000
GTX 1080Ti @ $700

= $1700

AMD will be able to give you ~90% of the performance of those two above at $999 (8 Core Ryzen + Vega)

 

[nathanddrews]

Power efficiency, bah! BRING THE HEAT!

If that Time Spy score is real - fully aware that it could be 100% fake - that would be amazeballs, but would that mean $699 for Big Vega? Undercut for $599/649? Topple the entire AIB industry and go back to 2010 with a $499 high end card? The mind, it wanders.

 

[raghu78]

It's all about financial resources. AMD is trying to attack a wider range of markets/products with a smaller R&D budget than NVIDIA, so of course it won't be able to flawlessly roll out an entire stack like NVIDIA can and does.

I think the biggest problem over the last 5 years was the CPU division was hurting the GPU division. Plain and simple. With Zen, AMD can get their CPU division to perform much better. The key here is to make sure that AMD keep up their CPU and GPU R&D pipeline well fed. I think that Naples should help AMD to grow their CPU division revenue as right now they are at close to zero market share. Free cash flow and profits is the lifeblood of any company and the expectation going forward is the CPU division starts to generate meaningful profits. AMD was able to launch successful full GPU product stacks within 3 - 6 months eg: HD 4000 /HD 5000/HD 6000/ HD 7000 series. So they can do it again provided their CPU divison starts to generate meaningful revenue and profits.

It's all about that power efficiency that people keep pooh-pooh'ing Low efficiency is OK in desktop when you can just crank up your TBP (185W, anybody?), but in notebooks it's a death knell.

I believe perf/watt and perf/sq mm are the key indicators of a successful high tech PC CPU or GPU or APU. AMD ruled on these metrics with HD 4800 and HD 5800. With the advent of GCN they fell behind on these metrics because of different design decisions by AMD and Nvidia. In fact GCN became more Fermi like with full hardware scheduling and Kepler became less Fermi like with scheduling partially in hardware and partially in software.

http://www.anandtech.com/show/5699/nvidia-geforce-gtx-680-review/3

AMD is yet to recover from the efficiency loss. Lets see what AMD are going to do with Vega and Navi to address that efficiency gap.

 

[Paratus]

Though I agree that Vega need not be faster than 1080 Ti but should come at a compelling price/perf the problem for AMD is going to be the upcoming Volta launch. We do not want a repeat of the Maxwell generation when AMD went below 20% market share for the first time in ATI/AMD's history. Vega will have to contend with GV104 and GV102 shortly. Another thing is AMD have forgotten to do a complete stack launch in a short span of time after HD 7000 series. Nvidia meanwhile have perfected the launch of a complete stack and the Pascal series launch was amazing. Four GPUs launched in 5 months (GP104, GP106, GP102, GP107) covering USD 120 - USD 1200 price points was simply a lesson in outstanding execution that AMD should learn from. The other major problem is AMD's complete surrender of the high end notebook GPU market to Nvidia. It started with Kepler and was pretty much complete with Maxwell. Pascal reinforced that domination. AMD cannot be a successful GPU company with zero presence in high margin / high end notebook GPUs.

One of the main goals for AMD going forward with future GPU generation launches is to show up for the fight like they did with HD 4870 and HD 5870. These GPU generations are just roughly 2 years. So turning up a year behind the competition is just not going to work. AMD also need to get back to a decent presence in high end notebooks and that will only happen if AMD can compete with Nvidia in perf/watt. I hope Zen and Vega bring in enough cash to help AMD better invest in their product roadmap and allow them to come to market with competitive products on time against competition.

Completely agree with the comments on the notebook segment. I'm not a huge fan of the perf/w argument for desktops, but it's absolutely critical for add in cards for notebooks and AMD has dropped the ball there.

I do wonder if AMD will continue to secede the high performance/gaming laptop market and jnstead concentrate on notebook targeted Zen based APUs. A laptop Zen based APU with HBM and Vega CUs able to drive 4K for productivity apps and maybe 1080P medium gaming might be a cost effective and desirable approach for AMD. It would also mean competing with Intel as opposed to Intel + NV or even weirder AMD CPU + NV GPU.

I know our work laptops have almost completely gone ultrabook with integrated graphics instead of discrete cards. Seems like that's the laptop market that plays towards AMDs current strengths.

 

[raghu78]

Completely agree with the comments on the notebook segment. I'm not a huge fan of the perf/w argument for desktops, but it's absolutely critical for add in cards for notebooks and AMD has dropped the ball there.

I do wonder if AMD will continue to secede the high performance/gaming laptop market and jnstead concentrate on notebook targeted Zen based APUs. A laptop Zen based APU with HBM and Vega CUs able to drive 4K for productivity apps and maybe 1080P medium gaming might be a cost effective and desirable approach for AMD. It would also mean competing with Intel as opposed to Intel + NV or even weirder AMD CPU + NV GPU.

I know our work laptops have almost completely gone ultrabook with integrated graphics instead of discrete cards. Seems like that's the laptop market that plays towards AMDs current strengths.

I do think a Zen+Vega APU with HBM2 based High bandwidth cache is the best method to cannibalize sales from the Intel CPU + entry level Nvidia GPUs like GP107. A Zen+Vega APU with 2GB HBM2 cache would easily provide unbeatable perf/watt which translates to better battery life and competitive perf against such a discrete CPU + GPU combination. But still the performance would be nowhere close to what high end notebook GPUs like GP106 or GP104 provide. Moreover the high end notebook GP104 draw upto 100w of power. Notebook APUs will max out at 35-45w which includes a CPU portion so the GPU portion would only have a power budget of 15-20w.

So AMD needs a double pronged strategy with APUs for low end and full fledged high end notebook GPUs for the high end. For that to happen AMD have to be competitive with Nvidia in terms of perf/watt.