New Upcoming ATI/AMD GPU's Thread: Leaks, Hopes & Aftermarket GPU's Part 2

Anyone know how far away this is. From my search, it should have started already. My bad, got the days mixed up.

 

Tommorow

 

9AM PT
5->7 PM in EU (depends on location)

 

Yep 6am here on Thursday. Can't wait, I'm really excited to see what AMD has in store for us. I have a feeling it's going to be good.

 

Can´t wait.New Ryzen and new GPUs. Really interesting times ahead!

 

Not expecting much on the high end as far as GPU. From what I hear, Navi will be a very good mid-range.
But I am very excited and hope a lot of details are provide for CPU/GPU.

 

I'm hoping for something to compete with a GTX 1080 Ti or RTX 2080, but at a lower price. One more day to go.

 

That's what I'm hoping for. I want to move from my 980Ti to a RTX 2080 but don't feel like paying the Nvidia prices this time around.

 

Vega 2 or as AMD calls it Radeon 7. 25% more performance and 16gb HBM memory. There is your 1080Ti performance right there.
February 7th for $699. That's more like it.

Nothing about Navi so far.

 

Pretty extreme if they wring out a 25% performance gain over Vega 64 with VII so I wonder what they've done to it.
Overclocking a Vega 64 or underclocking it first to say 1400 Mhz core and 800 Mhz RAM and then going up to 1700 - 1750 Mhz core and 1100 Mhz RAM is a ~8% gain or so I believe it was from the tests back then mostly thanks to memory itself and it requires pretty much the full 1.250mv core voltage though some models can be undervolted and still retain near 1700 Mhz boost clocks too.

I wonder what else it has to gain the rest for that gain because the 1800 Mhz core clock isn't going to be it so that will be interesting to hear more about.
(It also has less core clusters at 60 instead of 64 but these have usually not shown too big of a difference although Vega saw a better improvement than Fury at least.)

EDIT: Then again with tweaks and the die shrink and who knows what else scaling could actually shift and be much better for VII than Vega was so there is that possibility too.
(Not just voltage and thermals either even if these are important factors too.)


EDIT: Though a 20 - 30% gain or somewhere in between (25%) would be rather spot on for a jump from one GPU to a newer one too I guess although it's not a completely new architecture here but a shrunk Vega one well we will see how it performs but I am impressed if that's what it lands at in actual benchmarks and games and not just these marketing slides early on.

Well higher would be nice but I think we're slowly moving past that and getting diminishing returns and it's a bit more of a uphill battle for these gains even from die shrinks and various tweaks. I might be wrong though, that's not the first time that would happen.

 

Maybe they fixed to broken features that didn't work on the original Vega GPU. From what i had heard that could well have been 15% performance right there. Will be interesting to see.

 

AMD Launches The Radeon VII Graphics Card – The World’s First 7nm Gaming GPU for $699 MSRP


As I have been talking about for the last few weeks, AMD has just launched their next-generation 7nm GPU: the Vega for consumer, called the Radeon VII.
I wasn’t sure whether the VII stood for Vega II or 7 and it turns out it stands for 7 (the roman numerical 7). This was one of the last projects of Mike Rayfield as I discussed in my exclusive almost a month back.

AMD’s Radeon VII Is Faster Than NVIDIA’s RTX 2080 – 60 Compute Units and 16 GB of HBM (memory) for 13.8 TFLOPs of compute

I must admit that I did assume the VII would stand for Vega II but it actually stands for 7.
All that said, this is a much-awaited comeback from AMD in the high-performance gaming market and the Radeon VII rocks 3840 stream processors capable of handling up to 1.8GHz in clock rates.
This puts its theoretical compute capabilities at a solid 13.8 TFLOPs. The power requirement will remain the same as its predecessor.

The Radeon VII reference design rocks three fans on the shroud and looks like a beefed up Vega GPU. This is the 7nm Vega revealed at Computex last year, customized for the gaming/prosumer market.
As far as performance goes, it looks like this is going to be an absolute beast because the rumors I heard last night was accurate.

Just heard a rumor from someone on the ground that AMD Vega II will beat 2080 in DX 12. ?

— Usman Pirzada (@usmanpirzada) January 8, 2019

Three games were tested including Battlefield V and Farcry 5 at 4K with maxed out settings and the Radeon VII beats out (according to the company’s first-party benchmarks) the NVIDIA RTX 2080 in all APIs with the highest margin in Vulkan.
AMD’s Radeon VII graphics card is the first consumer AMD GPU to net a solid 60 fps in maxed out 4k settings in Battlefield V and Farcry 5 and if these titles are any indication of the larger set, the company finally has a competitive 4K graphics card.

AMD has priced the graphics card very competitively at $699 and confirms that availability will be from February 7th (ie almost a month later). Its official folks, the company has (soft) launched the world’s first 7nm gaming GPU.

We need to wait at last 6-8months to price come down a little
IMO it will be then ~550-590€ and 3 free games.

Spoiler: screens

 

-> https://twitter.com/Radeon/status/1083060489456742401

 

Woooooooooooooooooooo looks like I am going team Red! Feb is the month I have been planning my upgrade around (Preparing for games like Anthem, Division 2, and so on). Thank you for sharing this!

 

Well, the show was interesting. New Ryzen definitely worth looking into and no need to get a new mobo unless something really ground breaking comes along. The Radeon 7 looks like a keeper, even some seem to moan about the price. Maybe time to sell my PowerColor Radeon RX Vega 64 RedDevil

 

As I understand it they doubled the number of ROPs (to 128) and more than doubled memory bandwidth (to 1TB/s).

Given how much performance we're able to squeeze out of Vega with memory overclocking that may well have been enough to get past the limitations of the current generation.

 

Ah that would help, the ROP's were considered part of why the cards were somewhat front-loaded as I recall where the cores had problems being fed fast enough so utilization and efficiency and thus overall performance could be problematic particularly with draw stream binning rasterization, rapid packed math, primitive discard and I'm probably missing something ended up not making as much of a difference as expected or outright not functioning at all and AMD didn't do much with these features.

Bandwidth is another key area although Vega10 (56 and 64) while they saw gains from increasing memory speed it wasn't that high comparably. (Though it still increased performance more than overclocking the core itself did.)
Combine the faster speed with higher bandwidth and perhaps a revised HBM2 chip maybe using less voltage or similar voltage but at higher speeds and that would definitively help.

Both of those and a higher clock speed and other tweaks and yeah it could push to 20-ish percent I could see that uncorking some of these bottle-neck problems the Vega design had on launch.

EDIT: Yeah 64 and 4 from the initial GCN design and that's been in place until now then hasn't it where the ROP's are doubled.
https://en.wikipedia.org/wiki/Graphics_Core_Next

I have some reading to do but there's been some tweaks but some of the key designs have also remained.
(There's a number of adjustments between GCN generations and then the GPU's also have gotten some tweaks like the improved tessellation performance found in Polaris and newer for one example.)

EDIT: I have some more reading to do and refreshing on the Vega limitations because I feel I'm fumbling it a bit. Been a year or thereabout so yeah I'm forgetting a lot of the finer details.

 

Clock is important too. Take card with 64 ROPs + 256 TMUs. And 2nd card running 32 ROPs + 128TMUs.
But 2nd card has double clock. Now, ROPs and TMUs perform same. But 2GHz card is faster because all those unlisted parts if infrastructure are performing twice as fast.

Polaris RX-580 with 32ROPs and 144TMUs catches to Fury X at around 1500MHz if it has sufficient memory bandwidth. (in games)
Yet Fury X has 1050MHz. That clock ratio would be giving effectively 46ROPs + 205TMUs to RX-580, which is way below 64ROPs + 256 TMUs of Fury X.

 

I'm happy with the performance boost; that would fit my 4K needs quite nicely right now. I really wonder if they will also offer a version with "just" 8GB HBM2 for $599....? 16 GB is mighty nice, but I think they would sell very well if they offered a version with half the VRAM for less money.

 

A link that shows the new AMD card running neck n neck with Ngreedias
https://www.pcgamer.com/amd-unveils-next-gen-radeon-vii-launches-february-7-for-dollar699/