Radeon VII

[TaKeN]

That video has been removed it would seem, I watched it last night and you aren't missing much.

They slot it into the motherboard, you see the fans spinning, it boots to the windows screen where they wave and the stream ends.

 

[Illuminist]

That video has been removed it would seem, I watched it last night and you aren't missing much.

They slot it into the motherboard, you see the fans spinning, it boots to the windows screen where they wave and the stream ends.

Thank you for that description for us XD glad i didnt see it now

 

[Gerard]

It's just a very expensive card to build. If rumours are true an issue with Navi was found late in the day forcing AMD to release a card they never even wanted to make.




Vega VII is a stopgap, built at high cost, to keep AMD relevant in the high end until Navi is ready.

Once you see wccftech, that's where you should stop reading. Amd never wanted to release or make this card and only showed it "late in the day" yet somehow had time to make a full heatsink and shroud for it as well as have availability about a month after announcing it with packaging etc? Yeah, seems legit.

 

[HeX]

Once you see wccftech, that's where you should stop reading. Amd never wanted to release or make this card and only showed it "late in the day" yet somehow had time to make a full heatsink and shroud for it as well as have availability about a month after announcing it with packaging etc? Yeah, seems legit.

It's a MI50. It's not like they weren't making them already, and the decision would have been made a few months prior. There's nothing to say all the prototyping wasn't done to show proof of concept but then thrown out because of the cost of manufacture, especially if Navi was due soon, only to be brought back as a fall back plan when Navi got shunted further down the line. Hardly beyond the realm of possibility.

 

[Davedree]

It's a cut down instinct card with a core that failed to hit 64CU's. It's not like they weren't making them already, and the decision would have been made a few months prior. There's nothing to say all the prototyping wasn't done to show proof of concept but then thrown out because of the cost of manufacture, especially if Navi was due soon, only to be brought back as a fall back plan when Navi got shunted further down the line. Hardly beyond the realm of possibility.

No this is the same mistake panos made, and remained quiet and pretended to let it go un-noticed. It's not a rebadged instinct card.
V7 has it's own pcb, different voltage regulation and had to put back on the display I/o.
Aswell as the pcb design cooler and shroud.

Panos was trying claim they are all the same card, and it would be a $50 loss .
https://forums.overclockers.co.uk/posts/32458007/

 

[Gerard]

It's a MI50. It's not like they weren't making them already, and the decision would have been made a few months prior. There's nothing to say all the prototyping wasn't done to show proof of concept but then thrown out because of the cost of manufacture, especially if Navi was due soon, only to be brought back as a fall back plan when Navi got shunted further down the line. Hardly beyond the realm of possibility.

They were making the cores but not with that type of heatsink and shroud assembly. And the story wccftech posted about the guy in question is at odds with other stories that say he simply retired. That site posts more bs than facts, i'm surprised anyone bothers to post their news anymore as its always treated with a raised eyebrow.

 

[Davedree]

They were making the cores but not with that type of heatsink and shroud assembly. And the story wccftech posted about the guy in question is at odds with other stories that say he simply retired. That site posts more bs than facts, i'm surprised anyone bothers to post their news anymore as its always treated with a raised eyebrow.

+1 exactly

 

[nashathedog]

It will be £50 cheaper than a 2080 by the looks of it, which to me is all raytracing and dlss is worth currently

Gibbo said it'll be somewhere around the 650 to 700 mark.

 

[stuartharrison]

I'm sure they weren't intending to release it as it wouldn't have been economically viable.
Then they saw the pricing of RTX cards.
A card with 16gb of hbm2 shouldn't be able to compete with a card with 8gb of ddr6, but that's all down to Nvidia's short-sighted and greedy price strategy.

 

[nashathedog]

No this is the same mistake panos made, and remained quiet and pretended to let it go un-noticed. It's not a rebadged instinct card.
V7 has it's own pcb, different voltage regulation and had to put back on the display I/o.
Aswell as the pcb design cooler and shroud.

Panos was trying claim they are all the same card, and it would be a $50 loss .
https://forums.overclockers.co.uk/posts/32458007/

So how & when do they test a core for quality control? Do they not test the Core & HBM package prior to it going on the PCB then?

I presumed they were failing M chips as well due to it having the cut down core & 16gb's of HBM but if they don't do any testing prior to fitting the core HBM package/interposer/whatever on the board and it is a different board to what's used for the M cards it can't be that & in turn that makes me wonder if they aren't also saving up a supply of full core models to release down the road.

 

[Gerard]

in turn that makes me wonder if they aren't saving up a supply of full core models to release down the road.

Full core models later would be a pointless release imo, between the vega 56 and 64 there's about 10% difference at stock 56 vs 64 cu's. 60 vs 64 cu's would be around 5% or thereabouts.

 

[Davedree]

So how & when do they test a core for quality control? Do they not test the Core & HBM package prior to it going on the PCB then?

I presumed they were failing M chips as well due to it having the cut down core & 16gb's of HBM but if they don't test testing prior to fitting the core HBM package/interposer/whatever on the board and it is a different board to what's used for the M cards it can't be that & in turn that makes me wonder if they aren't saving up a supply of full core models to release down the road.

I'd say they test the package once the Core and Hbm has been constructed on a test bed. And then it's a binning process from there on for Gpu functionality and hbm functionality.
With different production lines for each variation they require.

 

[HeX]

No this is the same mistake panos made, and remained quiet and pretended to let it go un-noticed. It's not a rebadged instinct card.
V7 has it's own pcb, different voltage regulation and had to put back on the display I/o.
Aswell as the pcb design cooler and shroud.

Panos was trying claim they are all the same card, and it would be a $50 loss .
https://forums.overclockers.co.uk/posts/32458007/

Indeed, when I said MI50 I meant the physical GPU core. They were making them already, it's not like they had to develop a new GPU from scratch with all the R&D involved. I also said:

"There's nothing to say all the prototyping wasn't done to show proof of concept but then thrown out because of the cost of manufacture"

Which accounts for the board/cooler design. It could very well have been a parallel project much like Big-Polaris vs Vega.

 

[Davedree]

Indeed, when I said MI50 I meant the physical GPU core. They were making them already, it's not like they had to develop a new GPU from scratch with all the R&D involved. I also said:
Whch accounts for the board/cooler design. It could very well have been a parallel project much like Big-Polaris vs Vega.

Ok, I get you meant the core despite saying card. I agree and totaly get the parrallel project you mention. So say the r&d was already finalised on a demo V7 but cancelled, then brought back. However it still adds additional expense to the v7 production and this is more my point.
Big polaris lol well, was another failure which i'm sure Lephurom would also defend

Thinking about it the mi60 uses 32gb hbm2, so as Nasha mentioned it'd be interesting what they do if the cores don't make the standard on that one.

 

[nashathedog]

Full core models later would be a pointless release imo, between the vega 56 and 64 there's about 10% difference at stock 56 vs 64 cu's. 60 vs 64 cu's would be around 5% or thereabouts.

I agree there'd be very little in it but if the chips aren't failures being taken from the pro line they will have cores that don't need to be cut so why not release them when there's enough? especially if there is an architecture problem setting Navi back (which I doubt).

I'd say they test the package once the Core and Hbm has been constructed on a test bed. And then it's a binning process from there on for Gpu functionality and hbm functionality.
With different production lines for each variation they require.

So if that's right & the module is tested on a test bench before going on to the pcb they may well be failed Pro chips after all, That makes more sense to me than AMD not testing them before putting them on the pcb.

EDIT: The MI50 and VII share the same amount of memory as well as the same number of cu's & stream processors so it looks like they're just diverting a percentage of those cores over for the VII. There definitely won't be a full chip coming if that's the case (certainly not with 32gb's of HBM on it.

https://www.tomshardware.co.uk/amd-radeon-instinct-mi60-mi50-7nm-gpus,news-59390.html

 

[Illuminist]

I wish someone with production knowledge or an insider with amd would find our Threads to lay some smack down with knowledge and truth. Maybe that would ruin the fun of us debating wafer to gpu production techniques and time frames

Thursday can't come soon enough this thread is making me crazy waiting

 

[Tom B]

I won't be installing it until I can get a block but I'm tempted to grab one on launch in case the price goes up... Whether to gamble or not... Hmm.

 

[Gerard]

I agree there'd be very little in it but if the chips aren't failures being taken from the pro line they will have cores that don't need to be cut so why not release them when there's enough? especially if there is an architecture problem setting Navi back (which I doubt).

In all likelihood they will under the radeon 7 name with some of the cores disabled in bios. A hands on with this card one of the youtubers did show that it didn't have a bios selector switch so that seems to be something put in place to stop potential core unlocking. Doubt many people are going to want to take the chance of bricking a $700 card, though some will be sure to try.

In fact i'd be surprised if many of these "failed" cores are failed at all.

 

[Illuminist]

In all likelihood they will under the radeon 7 name with some of the cores disabled in bios. A hands on with this card one of the youtubers did show that it didn't have a bios selector switch so that seems to be something put in place to stop potential core unlocking. Doubt many people are going to want to take the chance of bricking a $700 card, though some will be sure to try.

In fact i'd be surprised if many of these "failed" cores are failed at all.

Ah like the good old amd days where you could sometimes unlock cores like on the fury card to unlock it to a fury x noice if its possible

 

[nashathedog]

In all likelihood they will under the radeon 7 name with some of the cores disabled in bios. A hands on with this card one of the youtubers did show that it didn't have a bios selector switch so that seems to be something put in place to stop potential core unlocking. Doubt many people are going to want to take the chance of bricking a $700 card, though some will be sure to try.

In fact i'd be surprised if many of these "failed" cores are failed at all.

That's what I was thinking now that I've seen that they have exactly the same basic stats as the MI50 which I edited into the earlier post. I don't know enough about the build process to know if they test the actual cores before adding the HBM to it & if so whether the MI50 is actually using failed MI60 cores but it sounds like the VII is using the MI50 chip as is.

EDIT: The MI50 and VII share the same amount of memory as well as the same number of cu's & stream processors so it looks like they're just diverting a percentage of those cores over for the VII. There definitely won't be a full chip coming if that's the case (certainly not with 32gb's of HBM on it.

https://www.tomshardware.co.uk/amd-radeon-instinct-mi60-mi50-7nm-gpus,news-59390.html