-ENDED- AMD Next Horizon Gaming E3 2019 Live Thread

[cj09beira]

5 hours ago, Trik'Stari said:

I'm just saying, they could do both.

 

They've been going with that mindset for a while now.

 

Doesn't seem to be working as they still aren't dominating. Although I'm sure people will just blame Nvidia for this. Not that Nvidia isn't without fault at all, but AMD's mindset seems to be the bigger problem to me.

 

They give Nvidia too much control with this mentality. All they have to do is drop prices a bit and AMD get's shut out.

forgot to mention this is the small die (navi 12), there is still the bigger navi 10, and later on (2020) navi 20, it was interesting that they didn't mention anything about navi 10, i wonder why.

guess amd got tired of selling more for less and not getting results from it (470/570 are an excellent example  of this), 

[Taf the Ghost]

2 hours ago, leadeater said:

Nvidia is only releasing what they need to on the game market, architecture development is happening and performance is increasing as much and as fast as it can. Every generation is not going to be large leaps over the last, that's unrealistic and hasn't actually been a thing.

 

But the insinuation I was replying to was that Nvidia does not have the motive to improve which is quite incorrect. Nobody ever replaces like for like on enterprise, the ever march for more, better, faster, efficient has and will always be a thing.

 

If you go way back to the very roots of dedicated GPUs for computers by Nvidia you can see that each generation isn't always a great leap from the last, there have been leaps and there will be leaps to come but expecting Geforce 900 to Geforce 10 series every time is extremely unrealistic.

 

As for gaming that is over 50% of Nvidia's revenue and market, all their other business sectors are much smaller. However the extreme demands for performance and efficiency from the enterprise sector are also mutually beneficial to the gaming sector, if you want to claim Nvidia is holding back then I would invite you to look at the best Tesla cards Nvidia has and show any disparity between the performance increases there that are not equally found in gaming. If they can't improve performance by 40% on Tesla there is no way it's happening on gaming cards.

 

Pricing sucks, has been for a long time. We can debate that all day but that issue ends when equivalent performance cards from AMD are not any cheaper, not by the amounts that people wish were a thing. GPU development costs are extremely high.

Nvidia shifted to a Data Center Primary design approach a while back, but Turing was the first time we've seen it really start to change the gaming dGPU side. The 1660/TI's die is the only gaming part, the rest are for Data Center and Pro work that was given enough of the proper power to do gaming well. Nvidia is clearly on a glide path to leave the Gaming industry, but it's going to be a while.

 

AMD is thus incentivized to not start pricing wars. Nvidia will price to their margins to grab as much as possible each generation while they invest all of that into their Data Center and other business areas. AMD with split designs will just match for price category and have better margins (since they're running far smaller dies). We're definitely into a status period until Intel shakes things up in 2021.

[Taf the Ghost]

4 minutes ago, cj09beira said:

forgot to mention this is the small die (navi 12), there is still the bigger navi 10, and later on (2020) navi 20, it was interesting that they didn't mention anything about navi 10, i wonder why.

guess amd got tired of selling more for less and not getting results from it (470/570 are an excellent example  of this), 

Polaris failed in the regard of getting more market share. Polaris made AMD a lot of money, but AMD's market share only ticked up a minor amount during the period. And it looks like Big Polaris and a new geometry engine for Vega ended up all getting scrapped for Navi and the next console generation. This appears to have been the right call, too.

 

As a result, we're into duopoly mode rather than competition. Nvidia needs those margins and AMD will just price a small amount lower and probably not really dent market share anyway. Pricing is, thusly, here to stay. 

[Trixanity]

12 minutes ago, cj09beira said:

forgot to mention this is the small die (navi 12), there is still the bigger navi 10, and later on (2020) navi 20, it was interesting that they didn't mention anything about navi 10, i wonder why.

guess amd got tired of selling more for less and not getting results from it (470/570 are an excellent example  of this), 

AMD says it's Navi 10 so it isn't the small die. We're yet to see a 1650 competitor so it stands to reason that it's the job of Navi 12 and they should honestly launch that asap. I hoped the 'one more thing' GPU was Navi 12 instead of the anniversary edition.

[leadeater]

15 minutes ago, Taf the Ghost said:

The 1660/TI's die is the only gaming part, the rest are for Data Center and Pro work that was given enough of the proper power to do gaming well. Nvidia is clearly on a glide path to leave the Gaming industry, but it's going to be a while.

Not really, RT cores don't actually exist in the datacenter Tesla cards. Tensor is still a case of finding a use for something than developing something to accomplish a solution, unlike RT cores which had a specific requirement and purpose. That's what happens in an enterprise first architecture design model though. Until there is a definitive removal of RT and Tensor cores from gaming GPUs GPUs that feature those are no less or more gaming.

 

1600 series was more to do with a realization that the market isn't as willing to pay as high as Nvidia was expecting. It's also why the RTX 2080 Ti came out so soon, that's not typical and usually comes with a price decrease of existing products. Rather big blunder there I have to say, too expensive with nowhere to move. 

[cj09beira]

1 minute ago, Trixanity said:

AMD says it's Navi 10 so it isn't the small die. We're yet to see a 1650 competitor so it stands to reason that it's the job of Navi 12 and they should honestly launch that asap. I hoped the 'one more thing' GPU was Navi 12 instead of the anniversary edition.

did they were, if they did i missed it 

 

2 hours ago, mr moose said:

I don't have an issue with it's price or performance (need to see actual reviews before I care about that).   What I hate seeing is every card they release being only as good as  3rd or 4th tier down from a year ago (they are better this time though as it's actually only 9 months if they hurry up and get it to shelves without delay).

 

Don't you get sick of seeing AMD bring out a new product only for it to be another XX70 or below competitor?  There are literally three better performing GPU'S above that.

 

 

they need an halo product to gain some mind share, but personally dont care too much as thats way off my price range

1 hour ago, Humbug said:

I don't have an issue with them comparing to it's competitor. But the problem is as you said that they are always late. The recent success of the CPU department is partly due to the successful implementation of their strategy of having leapfrogging design teams in order to iterate from Zen 1 to Zen 2 to Zen 3 etc... They are delivering consistently and predictably on new products, they set the trends now.

 

Whereas on the GPU front they are always chasing Nvidia. Always delays and always playing catch up and launching half way through Nvidia's generation. And when they do launch it is never a complete line up either. Never high end and mid range launching close together. They just don't execute reliably on their GPU roadmap. The products are always competitive but never trendsetting.

 

Hopefully with Lisa Su confirming they have been expanding the GPU design teams and the company's new found profitability things may change. 

they simply have a schedule that lands in july, they have been launching products in july for quite some time now, the lack of complete lineup is just the fact that they are a smaller company now, and there is a gpu die missing still

[cj09beira]

@Taf the Ghost did you notice the block diagram in Wendell's video, at least this small gpu is still on 4 shader engines, but with doubled CUs (they are basically in pairs now)

[Trixanity]

3 minutes ago, cj09beira said:

did they were, if they did i missed it 

 

they need an halo product to gain some mind share, but personally dont care too much as thats way off my price range

they simply have a schedule that lands in july, they have been launching products in july for quite some time now, the lack of complete lineup is just the fact that they are a smaller company now, and there is a gpu die missing still

They didn't say it at the E3 event. As you can see I've lifted it directly from the latest GN video. Given the context I think it was a backstage event at Computex.

[cj09beira]

2 minutes ago, Trixanity said:

They didn't say it at the E3 event. As you can see I've lifted it directly from the latest GN video. Given the context I think it was a backstage event at Computex.

thanks, so i guess the gpu is smaller than expected 

[COTG]

5 hours ago, Trik'Stari said:

I'm just saying, they could do both.

 

They've been going with that mindset for a while now.

 

Doesn't seem to be working as they still aren't dominating. Although I'm sure people will just blame Nvidia for this. Not that Nvidia isn't without fault at all, but AMD's mindset seems to be the bigger problem to me.

 

They give Nvidia too much control with this mentality. All they have to do is drop prices a bit and AMD get's shut out.

as we can see from the cpu sector, AMD's mindset is not at fault here. Obviously they would have brought out a 64 cu part if they could, but because they are a significantly smaller company than either nvidia or intel, obviously their R&D is a small fraction of their competitors. It is already extremely impressive that AMD is outperforming the giant intel in the cpu space, and competition value wise with Nvidia. Additionally, AMD has gotten burned quite a bit from consumers on their previous gpu lineups, such as people still buying the gtx 1050/ti instead of the rx 400/500 series. I think according to the steam survey, however accurate that is, the 1050ti has 30x the amount of user the rx480 has, even though its value is in the absolute gutter. And I already know people will say its because the card doesn't need power cables etc etc, but for real that only affects a tiny minority of the actual users. 
I am completely with AMD on this one. They are bringing out strong midrange performers, and don't cut into their already smaller margins because as we can see from the past, it wouldn't even help them gain a significant amount of marketshare.

[Taf the Ghost]

3 minutes ago, cj09beira said:

@Taf the Ghost did you notice the block diagram in Wendell's video, at least this small gpu is still on 4 shader engines, but with doubled CUs (they are basically in pairs now)

RDNA is a hybrid of GCN and a VLIW setup. In GCN terms, AMD actually did a 2:1 downscale in CUs. It's 40CUs, but it's really 20CUs that can bifurcate themselves. 

 

 

From Reddit, but it makes the point. It's really a math trick, but all of the talk, in the rumor mill, of up to 128CU pointed to this approach. AMD will get more out of their Teraflops than they did before.

[Taf the Ghost]

5 minutes ago, cj09beira said:

thanks, so i guess the gpu is smaller than expected 

The 5700 is the small gaming GPU die. It's the Polaris replacement. There are two bigger dies and probably 2 smaller dies to come.

[cj09beira]

Just now, Taf the Ghost said:

The 5700 is the small gaming GPU die. It's the Polaris replacement. There are two bigger dies and probably 2 smaller dies to come.

the wierd thing is that usually they make the big guy first, its usually polaris 10, then 12, 14 etc, same with vega 10, 12..

so you think the 50ish Cu card is still coming just its navi 12 or something like that?

btw doesn't this mean that amd still hasn't addressed the pixel throughput thing ? (here is not a problem was its lower Cu amounts with full front end but with the 50+ it might come into play)

[Trixanity]

2 minutes ago, cj09beira said:

thanks, so i guess the gpu is smaller than expected 

I honestly think it's right where it should be - perhaps even a bit big within context. I was initially alarmed at the leaked performance/size numbers with the rumors of a 50~ CU design. With 40 CUs I think 251 mm^2 is a decent size given the 7nm process and it leaves room for (perhaps) a sub-200 mm^2 Navi 12 with something like 24 CUs. It would be a killer mobile part that could gain a lot of traction in the 15 inch thin and light segment like the 1050 did.

 

All things considered something on the die is taking up some space when taken into account that based on CUs alone Navi 10 should actually be sub-200 mm^2 so I guess the beefed up geometry and stuff takes up a lot of space.

[cj09beira]

1 minute ago, Trixanity said:

I honestly think it's right where it should be - perhaps even a bit big within context. I was initially alarmed at the leaked performance/size numbers with the rumors of a 50~ CU design. With 40 CUs I think 251 mm^2 is a decent size given the 7nm process and it leaves room for (perhaps) a sub-200 mm^2 Navi 12 with something like 24 CUs. It would be a killer mobile part that could gain a lot of traction in the 15 inch thin and light segment like the 1050 did.

 

All things considered something on the die is taking up some space when taken into account that based on CUs alone Navi 10 should actually be sub-200 mm^2 so I guess the beefed up geometry and stuff takes up a lot of space.

its probably the extra cache, 

40 Cu with the full front end is the sweetstop for gcn, the leaks point to a bigger still die which is fine by me (not that i have the money for it, but might pick it up second hand later on)

[Doobeedoo]

Very exciting indeed. Fitting they did 50y anniversary with new products too. 

But that 16c with that boostclock is amazing oof. As for new cards, great we're moving away from Polaris and Vega now, also I wonder about the blower cooler, decision behind it. Maybe being vapour chamber and designed better, more efficient too, it can keep it cool with OC too I guess. 

[mr moose]

32 minutes ago, ravenshrike said:

It was needed for the AI side of things, not the consumer side, especially since all else being equal AMD was just better at compute than them. But because they had a bunch of chips not suitable for the business side of things, they foisted it off as the latest and greatest gaming tech, even though in order to work properly and have everything actually raytraced in games it needs to be at least 10 times faster than it currently is. As is it makes little sense for developers to sink significant time and money into the systems, and even with the new consoles any tracing APIs will be optimized for the MS spec, not nVidia's RTX.

I'm not sure that makes any sense.   

 

Even though half Nvidia's revenue is from the domestic market, they develop for all markets.  They developed AI for both domestic and commercial, they developed Tensor cores for both,  They definitely developed RT cores for commercial, you only have to look at pixar and al the shit they did with renderman.  Just because they pushed it into domestic doesn't mean they are offloading pointless hardware,  read my post again, no one puts $300M into useless hardware then tries to earn it back out of the domestic market.

 

18 minutes ago, cj09beira said:

they need an halo product to gain some mind share, but personally dont care too much as thats way off my price range

They also need a halo product to bring competition back to the top end of the market and to prove to investors they are capable in that department too.

 

 

[ravenshrike]

6 minutes ago, mr moose said:

Just because they pushed it into domestic doesn't mean they are offloading pointless hardware,  read my post again, no one puts $300M into useless hardware then tries to earn it back out of the domestic market.

I'm not saying the hardware is pointless, I'm saying the cards that were pushed for gaming failed qualification for the commercial side of things for one reason or another. It lets them offload otherwise worthless chips onto a market that they have no competition with on the top end.

[Taf the Ghost]

10 minutes ago, cj09beira said:

the wierd thing is that usually they make the big guy first, its usually polaris 10, then 12, 14 etc, same with vega 10, 12..

so you think the 50ish Cu card is still coming just its navi 12 or something like that?

btw doesn't this mean that amd still hasn't addressed the pixel throughput thing ? (here is not a problem was its lower Cu amounts with full front end but with the 50+ it might come into play)

Navi 10 and 12 (we just got shown 12) should be the hybrid design. Navi 20 should be the fully new Architecture. To the point they keep calling it Next-Gen even though we call it Navi 20. Which might have even more geometry. We'll know more since it should be relatively the same design as the Next Xbox's GPU.

 

As for Navi 10, it's probably up to a 64CU design. Maybe only 56, but we'll find out eventually. I imagine it'll also be related to the PS5's GPU.

[cj09beira]

1 minute ago, Taf the Ghost said:

Navi 10 and 12 (we just got shown 12) should be the hybrid design. Navi 20 should be the fully new Architecture. To the point they keep calling it Next-Gen even though we call it Navi 20. Which might have even more geometry. We'll know more since it should be relatively the same design as the Next Xbox's GPU.

 

As for Navi 10, it's probably up to a 64CU design. Maybe only 56, but we'll find out eventually. I imagine it'll also be related to the PS5's GPU.

look a bit up ^ amd claims this is navi 10

[Taf the Ghost]

1 minute ago, cj09beira said:

look a bit up ^ amd claims this is navi 10

Good catch.

[TheDankKoosh]

Overall, this is disappointing due to the fact that navi should have been the product that pushed nvidia to lower their own prices, but instead we get a tweaked Polaris with higher clock speeds and better memory while amd are charging out the ass for a mid range chip. I'll be waiting for nvidia super and the eventual price drops that are coming for the current 20 series. 

[Brooksie359]

12 hours ago, S w a t s o n said:

$750 is too much for the 16 Core.
 

When I made this prediction almost 1 full year ago "economical" meant 1800x pricing or $500. After everyone overhyped the pricing leaks I thought AMD might take liberties and go to $600.

I am not paying $750 for 16 cores on AM4, I'll wait for Threadripper and take the higher pcie lanes and lower heat density of w/e 16 core Zen2 TR they release. Lisa you really let me down by trying to play intel's game so early. You could have at least waited until you had proven yourself.

Seeing as it wont come out until September I would assume they will be highly binned chips so I don't think 750 for a 16 core is that ridiculous. By the time threadripper is out I would expect that the price would drop tbh. 

[Trixanity]

43 minutes ago, TheDankKoosh said:

Overall, this is disappointing due to the fact that navi should have been the product that pushed nvidia to lower their own prices, but instead we get a tweaked Polaris with higher clock speeds and better memory while amd are charging out the ass for a mid range chip. I'll be waiting for nvidia super and the eventual price drops that are coming for the current 20 series. 

That's the problem isn't it? The price is reasonable and the performance likewise for the segment it's in yet you're pretty much saying that it's an Nvidia card you want.

14 minutes ago, Brooksie359 said:

Seeing as it wont come out until September I would assume they will be highly binned chips so I don't think 750 for a 16 core is that ridiculous. By the time threadripper is out I would expect that the price would drop tbh. 

Yeah, I haven't seen anyone else say $750 is expensive. In fact I've mostly heard it's a great price for a 16 core. TR 2950X launched at $900.

[GoldenLag]

1 hour ago, Trixanity said:

With 40 CUs I think 251 mm^2 is a decent size given the 7nm process and it leaves room for (perhaps) a sub-200 mm^2 Navi 12 with something like 24 CUs. It would be a killer mobile part that could gain a lot of traction in the 15 inch thin and light segment like the 1050 did.

expect a part roughly the size of 70-80mm^2. its at this size we will see something that can fit in an AM4 APU. something they will want for both APUs and mobile. though mostly mobile.