Another navi leak

[LukeSavenije]

Just now, Neftex said:

radeon 7 would still be meh card for gamers even if it wasnt rushed, thats also why they didnt make ton of supply of it.

 

just saying what makes sense, people will be building new pcs with zen2 coming out and they will need a gpu. with navi nowhere and nvidia beating other amd offers with price/performance it might hurt them quite a lot

the actual reason is that they barely make money, if even lose money on it.

[Zeeus]

4 hours ago, LukeSavenije said:

i don't care who it makes

 

intel, amd, nvidia. just gimme dem nice hardware

 

that's how i think everyone should do

I DONT DISAGREE 

 

the issue is that thought is extremely unrealistic, based on the last 5 years seeing gpu releases from both sides its not hard to see that amd is actually 1-2 generations behind in terms of tech of gpus compared to nvidia. 

 

And intel lmfao yeah maybe in like 7-10 years theyll get gtx 2080 level performance

[LukeSavenije]

5 minutes ago, Zeeus said:

I DONT DISAGREE 

 

the issue is that thought is extremely unrealistic, based on the last 5 years seeing gpu releases from both sides its not hard to see that amd is actually 1-2 generations behind in terms of tech of gpus compared to nvidia. 

 

And intel lmfao yeah maybe in like 7-10 years theyll get gtx 2080 level performance

well, they are working on a gpu

 

and they have some high tier people walking around there...

[Zeeus]

20 minutes ago, LukeSavenije said:

well, they are working on a gpu

 

and they have some high tier people walking around there...

They poached raja kadouri or whatever that dude didnt really put amd ahead of nvidia then so esp considering hes gotta start from scratch its safe to assume intels got nothin atleast for several years. I just want someone with the capital to come in and pour in the extensive capital like nvidia puts into their gpu dept. If someone like samsung bought out amd graphics division and poured in billions which amd cannot do then we might see them compete with nvidia again. As things stand theyve gained lots of revenue thru the cpus and now with the new ryzen theyre sure to gain lots more but then assuming they use that capital PROPERLY and invest it into the gpu division we might see the fruits of their labor like 3 years later. Were essentially stuckk with nvidia for now

[CarlBar]

8 hours ago, Doobeedoo said:

Yeah, for the mid-range part, nothing was said about higher end chip, which is the one that HBM2 would only make sense really. Though, just from more architectural improvements, the so called big Navi should be faster by a good margin compared to Radeon VII for example. 

 

They said no HBM on Navi. They didn't specify which size. Thats ignoring the reason they're moving away from HBM in the first place, (stupid expensive), and the technical factor that you cannot replace GDDR with HBM without also replacing the memory controller which is a pretty hefty thing to do.

 

7 hours ago, Mihle said:

Wasnt it that they had no plans for chiplet GPUs for consumers, but they did have it for some server cards? (As some tasks on servers are different and therefore it can work with some type of tasks?)

 

They just said flat out no to chiplets. They did say it could make sense in the server space, but more in passing. remember, chiplets are not amazing because "muh Chiplets" but because of what chiplets allow in terms of binning, scalability and cross product stack silicon usage. If your producing chiplets for just a single market thats not running into die scaling issues yet chiplets lose all of their reasons for existing.

 

[Trixanity]

8 hours ago, CarlBar said:

 

They said no HBM on Navi. They didn't specify which size. Thats ignoring the reason they're moving away from HBM in the first place, (stupid expensive), and the technical factor that you cannot replace GDDR with HBM without also replacing the memory controller which is a pretty hefty thing to do.

 

 

They just said flat out no to chiplets. They did say it could make sense in the server space, but more in passing. remember, chiplets are not amazing because "muh Chiplets" but because of what chiplets allow in terms of binning, scalability and cross product stack silicon usage. If your producing chiplets for just a single market thats not running into die scaling issues yet chiplets lose all of their reasons for existing.

 

Chiplets may not be on the horizon just yet but they're definitely doing lots of R&D on the subject. So is Nvidia and Intel.

It would make sense to scale up a small die but latency is a killer for anything in real time (gaming is probably the most relevant scenario for this forum). So you'd need to somehow either fix or hide the latency. That's hard. Compute is generally not sensitive (to my knowledge) to latency so there's a lot that chiplets could do for servers. Of course (as you say) the chiplets needs to be able to accomplish something a GPU farm cannot. I think chiplets could play a role in the scaling from mainstream users to enterprise with the same die as we see in Zen but it isn't ready yet. The other obvious solution is scaling mainstream GPUs to get flagship performance. However, as I said: for gaming it's just not possible yet.

 

The tl;dr is that the main goal of a GPU chiplet is to move away from monolithic behemoth chips at 500+ mm^2 but we're not there yet and we can still get by with those chips where needed anyway.

[Doobeedoo]

10 hours ago, CarlBar said:

They said no HBM on Navi. They didn't specify which size. Thats ignoring the reason they're moving away from HBM in the first place, (stupid expensive), and the technical factor that you cannot replace GDDR with HBM without also replacing the memory controller which is a pretty hefty thing to do.

They said they have GPUs in mind with both GDDR6 and HBM2 though. GDDR6 was priced higher, while HDM2 is still more expensive than that, the tech is supposed to get cheaper over time too. Potentially with HBM3 but we'll see. So only makes sense for the flagship.

[CarlBar]

23 minutes ago, Doobeedoo said:

They said they have GPUs in mind with both GDDR6 and HBM2 though. GDDR6 was priced higher, while HDM2 is still more expensive than that, the tech is supposed to get cheaper over time too. Potentially with HBM3 but we'll see. So only makes sense for the flagship.

 

You have a source on that because i've only heard about HBM on 7nm in relation to the server compute cards.

[Doobeedoo]

14 minutes ago, CarlBar said:

 

You have a source on that because i've only heard about HBM on 7nm in relation to the server compute cards.

Hm I remember them mentioning that Navi is being made to work with both GDDR6 and HBM2 as well. It would make sense considering they're trying to cove the entire range with it as we know. 

[CarlBar]

4 hours ago, Doobeedoo said:

Hm I remember them mentioning that Navi is being made to work with both GDDR6 and HBM2 as well. It would make sense considering they're trying to cove the entire range with it as we know. 

 

Saying you've heard that isn't a source...

[ThePD]

On 3/14/2019 at 2:08 PM, CTR640 said:

Whenever WCCFTech releases a news article about nVidia, Intel and AMD, I simply skip the article and read the comments.

They've got the most creative gifs, images and more of the shitposting They are too hilarious.

This is the exact reason I go to WCCFTech. That comment section is gold.

[Doobeedoo]

9 minutes ago, CarlBar said:

 

Saying you've heard that isn't a source...

Oh I didn't put it. https://www.pcgamesn.com/amd/amd-navi-gpu-release-date-performance

 

[CarlBar]

4 minutes ago, Doobeedoo said:

Oh I didn't put it. https://www.pcgamesn.com/amd/amd-navi-gpu-release-date-performance

 

 

You might want to read your own link. All AMD has said is what i repeated above. Namely it makes sense in compute/workstation uses but not a desktop one. Everything else is the writers of those articles speculating basd of what was said.

[Doobeedoo]

1 hour ago, CarlBar said:

 

You might want to read your own link. All AMD has said is what i repeated above. Namely it makes sense in compute/workstation uses but not a desktop one. Everything else is the writers of those articles speculating basd of what was said.

I was just pointing that Navi is prepped for both memory techs and and GDDR6 making definite sens for mainstream while HBM2 for workstation cards and high end like Radeon VII is.

[Hobox]

Hasn't AMD said, multiple times, that Navi is still based on GCN and is their last one that will be? Why are people saying it isn't now, I haven't seen any indication of this anywhere.

[Hellion]

On Thursday, March 14, 2019 at 3:01 PM, Neftex said:

radeon 7 would still be meh card for gamers even if it wasnt rushed, thats also why they didnt make ton of supply of it.

 

just saying what makes sense, people will be building new pcs with zen2 coming out and they will need a gpu. with navi nowhere and nvidia beating other amd offers with price/performance it might hurt them quite a lot

So do you consider a 2080 "meh" as well then?

 

You realize AMD and Radeon are actually two seperate entities now, right? They have very different release cycles...

[mr moose]

On 3/15/2019 at 8:01 AM, Neftex said:

radeon 7 would still be meh card for gamers even if it wasnt rushed, thats also why they didnt make ton of supply of it.

 

just saying what makes sense, people will be building new pcs with zen2 coming out and they will need a gpu. with navi nowhere and nvidia beating other amd offers with price/performance it might hurt them quite a lot

That makes sense,  I am waiting for Zen2,  I am not in a rush and if they can offer something that makes Nvidia just look expensive at the same time then they'll win another sale.    In then end they'll release at the most tactical time for them. 

[Neftex]

6 hours ago, Hellion said:

So do you consider a 2080 "meh" as well then?

2080 is same price with overall better performance, it was also released earlier - that makes radeon 7 kinda pointless for gamers? why wait longer for a card thats worse? then theres the rtx stuff which isnt in a good shape now but could be in near future.

 

you could even compare it to 1080ti since both cards dont have rtx stuff, then it looks even worse doesnt it? still slower and even waiting longer

 

6 hours ago, Hellion said:

You realize AMD and Radeon are actually two seperate entities now, right? They have very different release cycles...

you realize noone thats gonna build a new pc actually cares about that, right? they need a gpu and navi wont be there

[Stefan Payne]

On 3/15/2019 at 3:58 AM, CarlBar said:

emember, chiplets are not amazing because "muh Chiplets" but because of what chiplets allow in terms of binning, scalability and cross product stack silicon usage.

That too 

but the main reason is that 7nm is more expensive and the I/O stuff just doesn't scale anymore.

And a GPU doesn't have a lot of I/O and needs high bandwith. You can't do that with an I/O die, that would actually increase the Powerconsumption and reduce the performance dramatically.

 

WIth modern CPUs however you have a shit ton of I/O and make the CPU so that it doesn't have to rely too much on the darn slow memory, so you can also go a step back and put a Chipset on the CPU and connect the CPU with some kind of FSB.

 

A GPU has rather limited I/O Stuff. And there already was a "Chiplet" design: nVidia G80 with the I/O Chip...

And that's all you have: a couple of Displays (wich you can put in a seperate die if you wanted to), PCIe x16 and the Memory Controller...

[Hellion]

16 hours ago, Neftex said:

2080 is same price with overall better performance, it was also released earlier - that makes radeon 7 kinda pointless for gamers? why wait longer for a card thats worse? then theres the rtx stuff which isnt in a good shape now but could be in near future.

 

you could even compare it to 1080ti since both cards dont have rtx stuff, then it looks even worse doesnt it? still slower and even waiting longer

 

you realize noone thats gonna build a new pc actually cares about that, right? they need a gpu and navi wont be there

Hahaha...

 

It trades blows with the 2080 within 5% depending on the workload. Take your nvidia goggles off.

 

If you say the Radeon 7 performs average then so does the 2080. Except the Radeon 7 actually has additional tangible benefits with 8GB more vram for those that actually use a machine for productiveity purposes outside of games.

16 hours ago, Neftex said:

you realize noone thats gonna build a new pc actually cares about that, right? they need a gpu and navi wont be there

 

 

You realize the definition of entited is exactly that? Expecting the world to cater around you without taking into consideration that's not how it works in reality.

 

[Neftex]

2 hours ago, Hellion said:

Hahaha...

 

It trades blows with the 2080 within 5% depending on the workload. Take your nvidia goggles off.

 

If you say the Radeon 7 performs average then so does the 2080. Except the Radeon 7 actually has additional tangible benefits with 8GB more vram for those that actually use a machine for productiveity purposes outside of games.

 

 

You realize the definition of entited is exactly that? Expecting the world to cater around you without taking into consideration that's not how it works in reality.

 

it trades blows on a smaller node with 2 years old card . i said its pointless for gamers exactly for the reason that in other tasks it might be good choice. but we are not talking about those workloads here, are we? you need to stop fanboying over amd my friend

 

again, all im saying is that releasing navi 2-3 months later than zen2 will for sure lose them sales. im not demanding from them to release it at the same time, ill just buy whats available when i get my zen2 parts

[Trik'Stari]

As long as they go with GDDR6 instead of fucking HBM.

 

We get it AMD, HBM is neat, but you need to actually have shit in stock for people to buy.

[IntMD]

5 hours ago, Trik'Stari said:

As long as they go with GDDR6 instead of fucking HBM.

 

We get it AMD, HBM is neat, but you need to actually have shit in stock for people to buy.

https://www.overclockers.co.uk/pc-components/graphics-cards/amd/radeon-rx-vega-7

 

https://www.scan.co.uk/shop/computer-hardware/gpu-amd/amd-radeon-vii-graphics-cards

 

They're perfectly in stock in certain countries... but sure.

[CarlBar]

3 hours ago, IntMD said:

https://www.overclockers.co.uk/pc-components/graphics-cards/amd/radeon-rx-vega-7

 

https://www.scan.co.uk/shop/computer-hardware/gpu-amd/amd-radeon-vii-graphics-cards

 

They're perfectly in stock in certain countries... but sure.

 

For that matter the issue with HBM2 was simply ASMD underestimating how fast t would grow and thus price would drop. (also a bit of no one was expecting GDDR6 so soon at the time). GDDR5X and GDDR6 plus other factors around HBM took the legs out from under AMD's table.

 

Also don't assume GDDR6 means HBM is dead. Currently the biggest you can scale the frame buffer to with GDDR6 is 24GB. Where going to have to wait for a new GDDR standard to get more. Ultimately GDDR has a limited lifespan before you just can't scale it anymore in either speed or capacity. Which wall we'll run into first isn't clear but eventually somthing is going to have to step in to replace GDDR and HBM is the most likely candidate ATM.

[cj09beira]

On 3/14/2019 at 9:07 PM, Zeeus said:

I DONT DISAGREE 

 

the issue is that thought is extremely unrealistic, based on the last 5 years seeing gpu releases from both sides its not hard to see that amd is actually 1-2 generations behind in terms of tech of gpus compared to nvidia. 

 

And intel lmfao yeah maybe in like 7-10 years theyll get gtx 2080 level performance

thats not really true, amd simply is limited in their scaling to 64 cus, which means any high end gpu they make will be put outside its sweetspot,

had they released a small vega for example it would have faired quite well,

intel has one thing though, its "Fuck you money", and its a big fuck you, so i would not be surprised if their first gpu was competitive, also had they done this sooner they would also have the foundry advantage, though they might still do quite well given just how good they are at optimizing their design for the node (they also do more manual circuit laying than others if i am not mistaken)