Radeon VII

[D.P.]

Or maybe wafers are needed for the CPU chiplets, and this card is just a way of offloading left over 7nn Vega. Why announce Navi when it's going to be announced latter in the year. As part of the new PlayStation and kill any chance shifting old stock. Navi for PC will be announced and launched this time next year my guess.

They dont have any leftover Vega20 cores.

You seem to be afgreeing with me that Navi will be announced later. This means Navi is not close to being ready yet, probably due to the process.

If you think Navi for PC is not released unitl next year then this makes a very long wait with AMD having nothing to show. AMD's market share might become precariously low by that point, 10%?

 

[D.P.]

I'd hardly call it a chase, and this is no halo product contest. AMD are putting out a card with the performance levels we had access to nearly 2 years ago with the 1080Ti! That's all the 2080 is, with some extra tensor cores that aren't being used... which the Radeon VII obviously doesn't have. It's basically a 7nm 1080Ti, 2 years late. At the same price a 1080Ti was 2 years ago. Way to innovate AMD!

And using at least 80w more than the 1080ti despite being on a 7nm process.

 

[TNA]

Its a shame as I was really hoping that AMD would shake up the Industry. Its a decent enough card but basically its a 1080ti or 2080 with more ram. There is nothing compelling about it. I think that the 1080 ti will become a classic and dare I say it I may keep mine for a couple of years maybe more. There is nothing in the GFX card space that compels me to upgrade.

This isn't an Easyrider going on about his 1080ti again post. Its a genuine reflection on the Market today.

People defending the Radeon 7 two years after the release of the 1080ti when its the same speed and the only justification is a few few games need to have a serious chat with themselves.

Its a mediocre release....The only saving grace is its worth buying over the 2080... its just a shame its the same performance just with more ram.

I would have preferred less ram say 12GB and a retail of £499 this would then make the 2080 obsolete.

There we go again, easyrider on about his 1080Ti again...

Yeah I agree. £499 with even 8GB would have made it a lot more attractive. It the the pricing of this card that makes it a dud.

 

[Illuminist]

Hmmm,

I have mixed feelings about this VII
a 1080ti is currently around £800-£900
a 2080 is currently around £750
The vega 2 will be about £650 right ?
On one hand the vega 64 was the best I could get with AMD, now its 1080ti / 2080 performance so it makes sense to sell my vega 64 and add a little to get the new vega 2.

But... if someone said to me here is either a vega 2 or a 2080 ... even though i am a massive AMD fan, i think for the sake of gaming I'd accept the 2080..
THEN AGAIN, if you can undervolt and overclock the vega 2 to 2 ghz core and 1050+ HBM its going to be 10-15% faster ... my maths is that a 2080ti is 25% faster than a 2080/vega 2 meaning i am at best 10% under a 2080ti for £650...
THEN AGAIN you can overclock a 2080ti probably 10-15%
I am obviously confused and sorry for this post.. i really wanted the vega 2 to be either slightly cheaper or be 2080ti performance and £899-£999

 

[Marine-RX179]

There we go again, easyrider on about his 1080Ti again...

Yeah I agree. £499 with even 8GB would have made it a lot more attractive. It the the pricing of this card that makes it a dud.

Agreed also.

The 16GB HBM2 vram pointlessly put up the price to higher than it needed to be. AMD should have released 8GB version at sub £500 or even sub £550.

 

[Kaapstad]

HBM for a gaming card

Meh

 

[D.P.]

They can't just change the VRAM size though since the GPU was designed for 16GB 4 stack HBM2 for HPC tasks (with a 32GB version at some point).

i also expect the cards are binned after all the HBM and GPU die are mounted on the interposer. At that point you can't remove any HBM ram, so the final GPU has 4 working HBM stacks and some defective CUs. I believe Vega architecture coudln't be made to work with 3 stacks if one stack was not working, so you either get the full memory or the GPU does in the trash can.

The price point was taken solely to match the 2080. These GPUs could be sold as the cutdown Instinct MI50 but AMD probably don't see enough demand ( I suspect the performance for the power consumption is just not appealing) . Actual production& retails costs are probably over $700 (or close), but sice these are salvaged chips then getting 700 is better than nothing

 

[Illuminist]

Also if they announced pci ex 4.0 surely they need a new card at the end of 2019 early 2020 to go with that?

 

[DragonQ]

Going by AMD's own numbers, power requirements have probably gone up for Vega7. If they Claim Vega7 is 25% faster at same power and Vega 7 is 29% faster than Vega64, then we might be looking at 320w or soemthign outrageous. However, I suspect the reality is sticking to a 300w ceiling and performance is closer to the 22-25% improvement, so at least 10% behind 2080.

Erm no, some of that performance improvement will come from doubling the number of ROPs. Also the boost clock is already known: 1800 MHz. Whether it can actually hold those clocks for more than a millisecond is another matter...

Some sites suggest the TDP will remain at 295 W.

 

[LePhuronn]

I'd hardly call it a chase, and this is no halo product contest. AMD are putting out a card with the performance levels we had access to nearly 2 years ago with the 1080Ti! That's all the 2080 is, with some extra tensor cores that aren't being used... which the Radeon VII obviously doesn't have. It's basically a 7nm 1080Ti, 2 years late. At the same price a 1080Ti was 2 years ago. Way to innovate AMD!

By virtue of you splicing up my quote, I think you've missed my point.

I would much rather have AMD say "we have a strategy, we're sticking to it" and not release the Radeon 7, rather than release another short-term PR stunt and raise even more questions about their ability to produce gaming cards.

 

[Nasher]

There we go again, easyrider on about his 1080Ti again...

Yeah I agree. £499 with even 8GB would have made it a lot more attractive. It the the pricing of this card that makes it a dud.

The pricing on Nvidia's current cards make them a dud, yet people keep buying them and even defending them :/

 

[KillBoY_UK]

Hmmm,

I have mixed feelings about this VII
a 1080ti is currently around £800-£900
a 2080 is currently around £750
The vega 2 will be about £650 right ?

the GTX 1080ti is EOL, so ofc remaining cards will be more costly,

Hopefully AMD wont pull a vega and say its $699 ...but the first 1000 units at a retailer rebate(or how ever many it was with Vega at lunch) and its really $799+

 

[DragonQ]

By virtue of you splicing up my quote, I think you've missed my point.

I would much rather have AMD say "we have a strategy, we're sticking to it" and not release the Radeon 7, rather than release another short-term PR stunt and raise even more questions about their ability to produce gaming cards.

While I kind of agree, unfortunately shareholders win out always. Additionally, this card will be significantly faster than their current flagship, even if it is just a die shrink with clock boosts. It's a legitimate product for consumers even if it doesn't excite us here.

 

[KillBoY_UK]

Erm no, some of that performance improvement will come from doubling the number of ROPs. Also the boost clock is already known: 1800 MHz. Whether it can actually hold those clocks for more than a millisecond is another matter...

Some sites suggest the TDP will remain at 295 W.

The ROP increase will be Interesting as I've felt Fuji and Vega where unbalanced and held back by the ROP counts

 

[nashathedog]

People are always expecting too much when we're waiting for the next jump, It's an affliction that strikes every time we look towards what comes next, When compared to this generation I think Navi will sit as Polaris did to the Fiji cards when we dropped from 28 to 14, Someone posted yesterday how it must be a Polaris replacement, well yeah, that's all it was ever meant to be, The problem is people spent the last year or more raising expectations in others by fantasizing about what they want it to be & what they think it should be. The shrink to 7nm will knock Nvidia's socks off according to some, offering a significant performance increase,
Well guess what?
Here's 7nm in the flesh,
Is it everything you hoped it to be?

 

[Shaz12]

I think once nvidia goes 7nm, there will be a much bigger performance improvement

 

[nashathedog]

Hmmm,

I have mixed feelings about this VII
a 1080ti is currently around £800-£900
a 2080 is currently around £750
The vega 2 will be about £650 right ?
On one hand the vega 64 was the best I could get with AMD, now its 1080ti / 2080 performance so it makes sense to sell my vega 64 and add a little to get the new vega 2.

But... if someone said to me here is either a vega 2 or a 2080 ... even though i am a massive AMD fan, i think for the sake of gaming I'd accept the 2080..
THEN AGAIN, if you can undervolt and overclock the vega 2 to 2 ghz core and 1050+ HBM its going to be 10-15% faster ... my maths is that a 2080ti is 25% faster than a 2080/vega 2 meaning i am at best 10% under a 2080ti for £650...
THEN AGAIN you can overclock a 2080ti probably 10-15%
I am obviously confused and sorry for this post.. i really wanted the vega 2 to be either slightly cheaper or be 2080ti performance and £899-£999

Hoping for 2ghz will lead to disappointment. Today's Vega can only reach their claimed clocks after you undervolt and overclock it so if VII is stated to run in the 1800's I'd bet on that meaning that's around what we'll manage to get in a best case scenario after manually tweaking it.

 

[D.P.]

Found the link

https://wccftech.com/exclusive-mike-rayfield-amd-retires/

AMD’s SVP and GM RTG, Mike Rayfield leaves the company amidst chatter surrounding ‘disengaged behavior’, David Wang to be interim lead
He presented suggestions that werent really feasible such as ‘Radeon VII ‘ which was to be a Vega 20 based consumer facing part that cost $750 to build and would barely tie in with an [NVIDIA] GTX 1080 Ti.

This is one of the reasons why I said Vega20 wont come to consumers. Something must have gone badly wrong at AMD to do such a U-turn.

 

[Marine-RX179]

I think once nvidia goes 7nm, there will be a much bigger performance improvement

But not bigger performance per £ though.

The Nvidia cards has already pushed 192-bit card at sub £200 up to £200+, and now all the way up to £350+ with the 2060.

It's ridiculous that the cards not even at least 256-bit anymore with a price tag of £350.

I mean sure the lower power consumption with the small bus size is nice and all, but I think most people would rather have a 2060 with 256-bit bus (which would also mean 8GB vram instead of 6GB) and higher memory bandwidth leading to better performance, rather than lower power consumption. Overclocking in most cases represent that people would rather have higher performance even if it means tossing efficiency out the window.

 

[ANDREW GREEN]

I suppose the question to gamers is, when does having more than 8gb of VRAM actually become useful? Do you want the extra VRAM or Ray Tracing?

They can't just change the VRAM size though since the GPU was designed for 1I6GB 4 stack HBM2 for HPC tasks (with a 32GB version at some point).

i also expect the cards are binned after all the HBM and GPU die are mounted on the interposer. At that point you can't remove any HBM ram, so the final GPU has 4 working HBM stacks and some defective CUs. I believe Vega architecture coudln't be made to work with 3 stacks if one stack was not working, so you either get the full memory or the GPU does in the trash can.

The price point was taken solely to match the 2080. These GPUs could be sold as the cutdown Instinct MI50 but AMD probably don't see enough demand ( I suspect the performance for the power consumption is just not appealing) . Actual production& retails costs are probably over $700 (or close), but sice these are salvaged chips then getting 700 is better than nothing

Given they are salvaged chips does this mean this is a stopgap card? Not a custom design, using old card parts, using HBM probably because they have excess HBM stock to shift.