Poll: ** The AMD VEGA Thread **

[Chuk_Chuk]

Good reason to believe that vega will be priced more competitively than we think.. am taking the FE price as base here.. they manufactured and assembled all FE parts in US and it mustve been executed as less than EOQ contract.. so that alone should shave of $200 from RX cost.. further the 8GB excess HBM maybe another $200. total impact on cost $350-400... if FE is making 30% GM we are looking at per piece cost of $700 --> for RX that means anything between $300-350.. again extrapolating the 30% GM (Actually given the larger scale amd should be okay with lower margins). the RRP is approx. $428-500.

Vega is not currently good enough for a price that can allow amd to recoup their R&D. they can only hope to make good GM's...R&D is a sunk-cost amd can now only plan for the future

Depends on how they've done there costing. If they intend on recuperating most of the R&D in the server market then most of the cost in vega will be in material and will help to reduce the price.

If they can sell it cheaper the cheaper manufacturing costs will be IMO a result of that wafer supply agreement AMD has with GF. Lets face it if AMD is contracted to buy X amount of wafers from GF who's process isn't as good as TSMC then the upside for AMD should be cheaper wafers.

Have we seen this affect previous AMD GPU prices?

Plenty of people have already given you a more detailed explanation, certainly enough for you to be able to google die sizes and yields and to work out why what applies to a moduled CPU doesnt apply to a much larger GPU.

If you think my post was against forum rules you are free to report me, but last time i checked it was a pretty open set of rules, nothing that requires me to do what you're saying i need to.

My time is mine to do what I want with, expecting me to hand feed you basic information on a subject you seem to want to know about isnt within my remit. Thats what google is for.

First sorry for calling you arrogant is was unnecessary.

Now it seems to me from what I have read in your post; that you have missed understood my post.

Raja said that Vega will use IF. There are two uses that i know of for this. One of them is to make a modular CPU like Ryzen. We have seen with Ryzen that the smaller die size has lead to greater yields. They have also achieved high die utilisation per wafar. All of which has contributed to low manufacturing costs.

If we apply this to vega; If vega is more modular, would the smaller dies not have a better yield than a larger one?
If they can mix and match dies together to form different SKUs would they not get better die utilisation per wafar?

Which one of these Vega statements is wrong? Why is it wrong? Do smaller GPU dies have worst yields than larger ones?

As for the other statement. Its nothing to do with forum rules. The point of forums is discuss things and trade information?
So you disagree with me, that fair nothing wrong with that.
You can't be bothered to write why you disagree. Okay then
When pressed for more information you response is basically "Get lost i don't have time to tell you why I disagree"

Then why are you here if you don't have time to properly formulate post. You might as well not post since by your own admission you have nothing of any use to contribute.

Sorry about the delayed reply

In the video Adored talks about the RX having a 30MHz increase compared to the FEs 1600MHz so he was clearly not talking about base clocks.

Videocardz does talk about 3DMark11 not detecting overclocked unreleased cards properly but I don't think he's saying the base clock is 1630MHz as the performance would be higher -

Sorry I should have mentioned that it was my own speculation based on Adoredtv video and videocardz article

Either RX vega will have a base clock of 1630 MHz and boost higher or RX vega is capable of sustaining a boost speed of 1630MHz on at least a test bench (could be in a case).

 

[Kaapstad]

Thats the point, you kept blaming HBM on the 1080p performance, which using DX12 and Vulkan showed it was not the problem. And the DX11 drivers caught up over time.

Of course i agree that 4GB held it back, but the performance of the HBM itself was not a major reason for fiji's performance issues at lower resolutions.

There are plenty of AMD cards that managed 1080p fine on DX11 without HBM.

One of the selling points of DX12 is to make up for the shortfalls of older inferior hardware, does HBM fall into this category?

 

[Mauller]

There are plenty of AMD cards that managed 1080p fine on DX11 without HBM.

One of the selling points of DX12 is to make up for the shortfalls of older inferior hardware, does HBM fall into this category?

There you go again, making up circular logic even though older AMD cards show improved performance as well with DX12/vulkan etc. The issue was that the drivers couldnt feed it fast enought, considering its a 4000 core part, when the previous largest core from AMD was ~2500 that is a huge difference.

But even with the above, there are still other shortfalls in the architecture that reduce lower resolution performance.

And the point of DX12 etc is to remove the overhead caused by DX11 in the driver and the shortfall in drawcalls it causes.

 

[andybird123]

If we apply this to vega; If vega is more modular, would the smaller dies not have a better yield than a larger one?
If they can mix and match dies together to form different SKUs would they not get better die utilisation per wafar?

Which one of these Vega statements is wrong? Why is it wrong? Do smaller GPU dies have worst yields than larger ones?

Vega isnt using smaller dies, its a single big die. There's rumours kicking about that Navi might be able to go with smaller dies on an interposer but thats not Vega and its not even confirmed. Loads of people have pointed this out to you but you keep ignoring them. I could write a massive long post explaining the basics of GPU's but I dont feel this is an appropriate medium, you could just google the info and realise why Vega isnt a smaller die just because of IF. Its really basic entry level stuff, but you just keep regurgitating the same misocnception refusing to read or understand why its not correct. Thats not my lookout, sorry.

It is also Wafer, not wafar.

 

[FredFlint]

The FuryX got/gets a lot of stick, I have used mine daily for a long time now and it was a big improvement over my 290X and it got a lot better over time. I think the 290 cards got the same stick but now it's considered a very strong card for its age. Vega will most likely follow the same path.

 

[Panos]

The FuryX got/gets a lot of stick, I have used mine daily for a long time now and it was a big improvement over my 290X and it got a lot better over time. I think the 290 cards got the same stick but now it's considered a very strong card for its age. Vega will most likely follow the same path.

I agree. I had a GTX1080 @ 2190 for few months, but went back to my Nano and then got a second hand FuryX.
Anyone who bothered to upgrade the stock BIOS on the FuryX/Nano to the AMD UEFI one from April 2016, will find a different a card under there.
It feels faster and overclocks much better than the initial release, while overall the system boots and runs faster because of the UEFI BIOS.
In addition it has received a big boost with the drivers updates over the last 2 years.

But none of the reviewers have upgraded their FuryX/Nano bios, and barely any is benching the card with the latest drivers.

 

[Chuk_Chuk]

Vega isnt using smaller dies, its a single big die. There's rumours kicking about that Navi might be able to go with smaller dies on an interposer but thats not Vega and its not even confirmed. Loads of people have pointed this out to you but you keep ignoring them.

Wrong. Two people mentioned that; here (https://forums.overclockers.co.uk/threads/the-amd-vega-thread.18781388/page-249#post-30956013) and here(https://forums.overclockers.co.uk/threads/the-amd-vega-thread.18781388/page-250#post-30956206) 1 of which i responded to here(https://forums.overclockers.co.uk/threads/the-amd-vega-thread.18781388/page-250#post-30956274), the other I missed.

Can you please show me where AMD has confirmed that vega is not modular and that it is a monolithic die? Because i am not aware of this; as stated in my previous post as far as i know we only know the die size and a die shot (which i admit doesn't look modular but I am no expert). If AMD has confirmed this then there is no point to my speculation. So please give me link or tell me which event that AMD confirmed this?

I could write a massive long post explaining the basics of GPU's but I dont feel this is an appropriate medium, you could just google the info and realise why Vega isnt a smaller die just because of IF.

No you couldn't.
I was referring to the individual modules that form the full die (Like the CCX in Ryzen). I can how you could get confused in my previous post so lets try again. If vega had a modular design (where modules are fused together to form one die) would smaller modules have worst yields than a monolithic die (assuming that contamination per square m remains constant)? Yes or No?

Its really basic entry level stuff, but you just keep regurgitating the same misocnception refusing to read or understand why its not correct. Thats not my lookout, sorry.

It is also Wafer, not wafar.

No you just don't understand my posts. Is my grammar that hard to understand. I know that I have a habit of not including enough information in some of my posts but I don't see how someone can misunderstand me so badly.

Thank you for the correction. However I would recommend that in future you proof read your own post before correcting the spelling of other.

 

[silversurfer]

Infinity fabric allows for the joining of multiple engines on a single die, and offers high bandwidth and low latency.There has been no mention of using Infinity fabric with multiple GPUs.

https://www.reddit.com/r/Amd/comments/6bzu05/raja_kaduri_ama_recap/

Sounds more like it'd be a thing for Navi more then Vega. The ideal scenario: IF and HBCC being fully utilised and engaging with every game to their full theoretical advantage, the reality is probably just some slight use to start with or in the case of Frontier apparently no use at all of major features? (at least in games)

 

[Rroff]

Can you please show me where AMD has confirmed that vega is not modular and that it is a monolithic die?

http://www.guru3d.com/news-story/amd-vega-die-shot.html

https://www.reddit.com/r/Amd/comments/6bzu05/raja_kaduri_ama_recap/

Sounds more like it'd be a thing for Navi more then Vega. The ideal scenario: IF and HBCC being fully utilised and engaging with every game to their full theoretical advantage, the reality is probably just some slight use to start with or in the case of Frontier apparently no use at all of major features? (at least in games)

IF (as per current implementation) is potentially capable of doing similar to current multi GPU - maybe with some advantages but not to any ground breaking extent - to truly combine more capabilities than can fit in a single die required a different architecture to current monolithic cores.

 

[Mauller]

i would think that AMD is going throught the pain with HBM and interposer for another reason as well, since it is readying them for GPUlet based GPU packages with multiple gpu dies and their accompanying HBM.

 

[AthlonXP1800]

https://www.reddit.com/r/Amd/comments/6mfcpp/aibs_get_the_final_source_bios_for_radeon_rx_vega/

AIB partners to get RX Vega final BIOS source on 2 August but partners still not begin mass production RX Vega cards yet until sometime after 2 August. So that mean RX Vega launch at Siggraph will be a paper launch. RX Vega reference cards could hit retailers in September depend on GPU shortage and custom RX Vega cards in October or later just a few months before Volta launch.

Absolutely shamble launch schedule.

 

[nashathedog]

I suppose AMD sticking with HBM is for them to have CPU GPU MEM on one soc ...great for future consoles

That's where it will shine, Once we get APU/HBM packages for PC and as you said the custom APU's that are expected in the next gen consoles I imagine performance will be moving past 60+ 1080p without a problem.

 

[Mauller]

https://www.reddit.com/r/Amd/comments/6mfcpp/aibs_get_the_final_source_bios_for_radeon_rx_vega/

AIB partners to get RX Vega final BIOS source on 2 August but partners still not begin mass production RX Vega cards yet until sometime after 2 August. So that mean RX Vega launch at Siggraph will be a paper launch. RX Vega reference cards could hit retailers in September depend on GPU shortage and custom RX Vega cards in October or later just a few months before Volta launch.

Absolutely shamble launch schedule.

There can still be reference cards being made by select partners, like sapphire etc. to go with a siggraph launch. just no custom AIB models.

 

[Chuk_Chuk]

https://www.reddit.com/r/Amd/comments/6bzu05/raja_kaduri_ama_recap/

Sounds more like it'd be a thing for Navi more then Vega. The ideal scenario: IF and HBCC being fully utilised and engaging with every game to their full theoretical advantage, the reality is probably just some slight use to start with or in the case of Frontier apparently no use at all of major features? (at least in games)

THANK YOU! That was all.

http://www.guru3d.com/news-story/amd-vega-die-shot.html

Like I said i'm no expert. But now comparing it to the Ryzen die shot, I can see the lack of defined modular sections.

 

[nashathedog]

https://www.reddit.com/r/Amd/comments/6mfcpp/aibs_get_the_final_source_bios_for_radeon_rx_vega/

AIB partners to get RX Vega final BIOS source on 2 August but partners still not begin mass production RX Vega cards yet until sometime after 2 August. So that mean RX Vega launch at Siggraph will be a paper launch. RX Vega reference cards could hit retailers in September depend on GPU shortage and custom RX Vega cards in October or later just a few months before Volta launch.

Absolutely shamble launch schedule.

That isn't saying it'll be a paper launch, It's saying non reference models will come later,

There can still be reference cards being made by select partners, like sapphire etc. to go with a siggraph launch. just no custom AIB models.

Exactly, If it's reference cards first I hope they have a water cooled model as I'm not interested in a blower.

 

[five8five]

https://www.reddit.com/r/Amd/comments/6mfcpp/aibs_get_the_final_source_bios_for_radeon_rx_vega/

AIB partners to get RX Vega final BIOS source on 2 August but partners still not begin mass production RX Vega cards yet until sometime after 2 August. So that mean RX Vega launch at Siggraph will be a paper launch. RX Vega reference cards could hit retailers in September depend on GPU shortage and custom RX Vega cards in October or later just a few months before Volta launch.

Absolutely shamble launch schedule.

It could end being like the 480 launch.

One positive is it'll keep them out of the hands of the miners

 

[Mauller]

THANK YOU! That was all.



Like I said i'm no expert. But now comparing it to the Ryzen die shot, I can see the lack of defined modular sections.

Ryzen is a single Die, you must be mistaking the two solder pads being used for seperate die.

 

[Chuk_Chuk]

Ryzen is a single Die, you must be mistaking the two solder pads being used for seperate die.

You right I am mistaking something, but its not the solder pads.

 

[Kaapstad]

There you go again, making up circular logic even though older AMD cards show improved performance as well with DX12/vulkan etc. The issue was that the drivers couldnt feed it fast enought, considering its a 4000 core part, when the previous largest core from AMD was ~2500 that is a huge difference.

But even with the above, there are still other shortfalls in the architecture that reduce lower resolution performance.

And the point of DX12 etc is to remove the overhead caused by DX11 in the driver and the shortfall in drawcalls it causes.

You are still trying to tell me that HBM needs DX12 on AMD cards to get the best out of them, there are plenty of NVidia cards that are faster that don't use DX12 or HBM to get the performance.

You and DM really need to admit that you got it wrong with HBM and that there are better memory solutions for gaming cards.

Below is an old GPUZ screenshot of one of my Titan Xp cards with a small overclock, checkout the memory bandwidth using cheap GDDR5X.

With GDDR6 coming soon AMD really need to drop HBM for gaming cards.

 

[TimAhKin]

I was in vacation 3 weeks. Where is VEGA and how much it costs?