Radeon RX Vega to Compete with GTX 1080 Ti and Titan Xp

Precisely.
And with these little bumps in performance, like 30% for the 1080 Ti, or whatever Vega can do, it's not worth it.

Gone are the days when new GPUs were twice as fast every year... these days it's just Rebrandeon and GeRepeats (mostly)

If I do get that 4K screen, I can just get another 1080 and be done with it (many games still support multi-GPU AFR, and with DX12/Vulkan we might even see a rebound of multi-GPU), so lookin' good !

 

AMD has already changed the CPU landscape, and the GPU landscape just based on speculation and engineering sample performance. If they can deliver 1080-1080ti performance at a price point of $475-575, they will have to be reckoned with. If they can show benefits of pairing it with Ryzen, all the better. Follow that up with VegaVR for XBox Scorpio and keep on rollin'.

 

I think AMD's response to the question was actually pretty understated. There wasn't much else he could say. "Nice" is ambiguous enough. Also, keep in mind, they have shown demos with 1080 level performance already. Optimized games and API, yes, so what? I would like to see the NAB 8k editing demo though ...

 

Could this mean... "5% less is nice" or "10% less performance is nice"?

Well, for market's benefit (us), I will hope for equal performance.

 

Now, not only do we have the usual nVidia x AMD discussions, but we seem to be engaging in the exegesis of the new gospel that came from an AMD engenieer that used the arcane word "nice". That's wise. :3eyes:

 

Nice could mean anything given that we don't know how the tone of nice was used.

Give up this speculative "nice means it'll perform like this or that" and just wait for the release, it's supposed to be revealed at Computex or at least an announcement should be made regarding Vega at Computex.

 

How do you actually check the HBM2 availability? With Vega being as late as it is, the only HBM2 has been going to the Nvidia top of the line pro cards, which probably aren't selling like hotcakes but rather like Rolls-Royce. After all this time, there should be plenty of HBM2 chips already.

 

You can't really - you just have reports from Nvidia saying availability is low, hence the high prices and tech journalists like the guys from PCPer talk about industry insiders mentioning it.

It's also not so much the HBM2 chip itself but the fabrication. You have to mount each stack on an interposer along with the die then grow ~5K TSVs per stack and then validate each stack. Which is significantly more complicated than mounting GDDR. It takes extra time for validation and if you screw up you lose entire interposer and remaining stacks or you bin it for a lower value if you get lucky with where you screwed up.

And in the end, the Vega chip we saw is only getting 2 stacks, 8GB - which is roughly ~520GB/s which Nvidia is already doing with GDDR5x. Granted the HBM2 is lower latency, slightly lower power and reduces die size - but it significantly increases the cost, lowers the yields and potentially delays the launch/supply. And if you don't need lower latency (it's not really required for gaming) or reduce the die size, then why use it for gaming cards?

AMD has to use it with Vega because they can't afford to spin another chip specifically for compute. Nvidia can afford that (GP100 is essentially an entirely different GPU than Pascal GP102/104) and since they delivered that chip 6 months ago they can essentially set the price at whatever they want and cover the cost of the manufacturing.

 

Screw VR, AI is the new sh1t!
No one wants to be left behind in AI, so Nvidia couldn't meet the demand for quite some time. It's more like Hotcakes than like Rolls-Royce.

Lookie:
IBM, Alibaba, Amazon, Baidu, Google, Facebook, Microsoft, Tencent.

All of them buying DGX1 for usual number crunching, as well as for AI and for cloud computing. Nvidia's Datacenter division alone has grossed $300M, and netted more than the entire AMD.

HBM2 plenty or not, it should be handsomely more expensive than DDR5/5x.
Especially when you are competing with the likes of 8x $10k worth of GPUs in one DGX1.

 

For me Vega has got to come out with something that can compete with Volta not Pascal, Pascal is old tech, it's almost a year old at this point. if they bring out something to challenge pascal or even beat it, then so what? Volta is expect end of this year by the sounds of it, it will quickly be crushed and forgotten. Amd can't be sending cards out from a way different time zone than Nvidia is.

And if they don't beat pascal that is another hype train crashed and burned :/ if it really is HBM2 delaying them so much they should have given up on that idea and gone for GDDR5X instead. When the 480 came out they should have followed it up with small vega then maybe end of the year end it with big Vega... instead we are almost 5 months into 2017 and still nothing

Sorry for the rant, just frustrates me with a lack of competition right now on the GPU front, be nice if someone gave AMD some sense and direction on when to release products

 

It probably will compete with GTX 1080 Ti but only pricewise, because when it comes to performance I'd say it will be ~GTX 1070 and maybe ~1080 in low CPU demanding titles, but that's it!

 

So expecting barely better card then Fury X. You don't have any expectations at all

 

So you think it's going to get slower than the cards they already demoed, slower than their previous generation Fury X? Lol

If AMD just took Polaris and made it 4096 cores and scaled the rest of the architecture to feed the cores, it would be as fast as a 1080Ti. With zero architecture changes. But we know it has architecture changes - granted most of them are for energy efficiency - it still has them. So unless AMD's architecture changes reduce performance, or they screw up the balance and don't feed the pipeline correctly, it's definitely going to be really damn close to the 1080Ti.

I expect it will be $600-650 for the highest configuration and will likely trade blows with the 1080Ti in DX12 titles and be slightly behind in DX11 ones. AMD owners will look at DX12 results and say "its faster for future workloads and thus better card" Nvidia owners will look at overall results and say "it's slower than a 1080Ti in majority of titles thus worse card". AMD owners will then bring up lower cost Freesync monitors being a benefit while Nvidia owners will bring up that Volta is coming soon, even though it's probably not coming for an entire year.

That's my prediction.

 

For me there´s no doubt that Vega is going to be very competitive against Nvidia performance wise and that is gonna be better priced.

The problem is that is going to be so late for the party!!! When it finally arrives, almost everyone who wanted to upgrade have already bought a Nvidia GPU...

 

Slower than Fury X? No, definietly no. But right now Fury X is faster than 1070 only on paper, in real life it's slower in 95% scenarios.

I assume that new high end AMD GPU will suffer from similar issue like old one - will require much more powerful CPU than competition to feed all cores and achieve good results.

 

You forget that the world is rotating... and the amount of people who want to upgrade doesn't just vanish ... you will always and at ANY time have a good amount of people who are willing to upgrade.

In two months, there will be people who didn't want to upgrade until then and will then just decide to upgrade.

Your view is very static. But the market is extremely dynamic. If Vega is what you expect it to be ("very competitive and fair-priced") then a lot of people will buy it. Even if a lot of people buy 1070s and 1080s until Vega comes out.

There is always demand. And being late always depends on what you are delivering. Because basically there is no such thing as being late.
If AMD delivers 1080 - 1080Ti performance for a better price than 1080 ... if they really can do that... then from release on everyone who wants that performance and doesn't have a card yet ... will buy VEGA...

 

How I would predict it to happen too.

 

Can you give your technical explanation as to why games with low CPU demand would run better on VEGA than ones with high CPU demand?

By what logic? Are you just making this up?
If at all...wouldn't it be the other way round?

 

Well, I am waiting how the RX Vega does on benches,temps,etc before I make any move on buying a 1080ti.

I am really looking into 2 1080ti's at this point, but really holding off on how the Vega performs, I am waiting, this shall be interesting.

I mostly game 4k or higher with all bells and whistles intact, if its lower: like 1600p, I add more graphic filters to make it up, depending on game.

 

There have been a lot of rumors and speculation batted about. Just because someone presented some words on the internet for people to read just makes it the author's opinion, nothing more. Reading into subjective statements is a waste of time. Nice is not ambiguous it is subjective. There is no accounting for taste.

The majority have missed the big picture. AMD has two battle fronts. The cpu front is a mess at the moment. They are trying to patch up Ryzen. Vega will be plagued with the same general type of issues. They will be scurrying about to patch up it's performance as well. Just because the Radeon Group functions autonomously with the cpu segment still puts them at want with the same resource pool. All these decisions are based on their cash flow.

Based on early projections they figured they could release Vega in the first half of 2017. The article pushing October 2016 was a laugh. All the funds were tied to Ryzen. Triage is cheaper and requires less funds. This is what AMD needed to be able to push Vega. So at this point they are still on track. Vega will be released in the same shape as Ryzen...needing a triage team. Expecting anything else would require a miracle from AMD.

Even with that constraint going against them and nice is the word issued, just means the rabbit did not die.