[Videocardz] 1650 details

[Guru]

Yeah, if the uncut chip + GDDR6 came out, and was a tad faster than the RX 570, then I could see that card justifying a $150 price, possibly. But certainly not the GTX 1650, as it stands now. $120 max, is what that card is worth.

Edit: Though, I do wonder how much AMD is making these days on $130 RX 570 4GB/$140 8GB cards. And isn't the RX 590, based on a 12nm shrink? Why no down-rev cards based on 12nm yet? Maybe we'll see an RX 570 Mark 2, or an RX 575, or something, basically, an RX 570 based on the 12nm shrink GPU die, with faster clocks than the current RX 570.

Maybe AMD is saving that, for when the GTX 1650ti comes out?

Edit: The clocks on my RX 570 8GB cards are around 1280Mhz core clock, 1750Mhz (x4) GDDR5 memory clock. If AMD could release an RX 575 at around the same price point (maybe $160-170 to start), with clocks in the 1500Mhz range for core clocks, then I think they could see a resurgence.

Maybe the reason that they haven't released that in terms of a stop-gap against the GTX 16xx cards, is because Navi is right around the corner, and targeted at the same target market? (And thus, releasing an updated RX 575 mid-range card would just cause confusion?)

Edit: Butbutbut... the WSA! If the 14nm/12nm Polaris dies are made at GF, then possibly the Polaris RX GPU family could live on longer, at least in certain regions of the world, living on a 12nm shrink, due to costs in fulfilling the WSA agreement? Since Navi is going to be based on 7nm dies made at TSMC, I guess, and probably pricey.

Because they are working on their Navi lineup, which is going to be releasing in about 3 months. August is basically the max date for arrival for these cards.

They wouldn't have the time and money to port the RX 570 to 12nm. And ultimately it will essentially be a RX 580, which they already sell for as low as $160, on average go for around $180.

All their focus is on Navi right now. At 7nm I think it is going to be very competitive with Nvidia's offerings. They've also taken their sweet time with it, considering its been in development for 3 years and 7nm has been available since mid last year, so hopefully they've optimized it as much as possible and binned their chips really well in order to decrease power draw.

 

[VirtualLarry]

They wouldn't have the time and money to port the RX 570 to 12nm.

I thought the RX 590 was based on a 12nm shrink? Same GPU chip, basically. Should be trivial to cut down shaders and make an RX 570-equivalent out of them, at higher clock speeds like they did with the 580->590 transition. I'm just suggesting that the 570->575 get the same love.

All their focus is on Navi right now.

I realize that. But they may still have outstanding WSA obligations. I don't know what their allocation between 14nm I/O chiplets for Ryzen 3rd-Gen CPUs versus RX 590 12nm GPU dies looks like.

Maybe they will cease production of Polaris at GF, and dedicate their wafers to the I/O dies for Ryzen 3rd-Gen CPUs. Maybe they really will sell that many, or maybe they're looking at supplying HP and Dell with more product this time around.

 

[maddie]

Yeah, if the uncut chip + GDDR6 came out, and was a tad faster than the RX 570, then I could see that card justifying a $150 price, possibly. But certainly not the GTX 1650, as it stands now. $120 max, is what that card is worth.

Edit: Though, I do wonder how much AMD is making these days on $130 RX 570 4GB/$140 8GB cards. And isn't the RX 590, based on a 12nm shrink? Why no down-rev cards based on 12nm yet? Maybe we'll see an RX 570 Mark 2, or an RX 575, or something, basically, an RX 570 based on the 12nm shrink GPU die, with faster clocks than the current RX 570.

Maybe AMD is saving that, for when the GTX 1650ti comes out?

Edit: The clocks on my RX 570 8GB cards are around 1280Mhz core clock, 1750Mhz (x4) GDDR5 memory clock. If AMD could release an RX 575 at around the same price point (maybe $160-170 to start), with clocks in the 1500Mhz range for core clocks, then I think they could see a resurgence.

Maybe the reason that they haven't released that in terms of a stop-gap against the GTX 16xx cards, is because Navi is right around the corner, and targeted at the same target market? (And thus, releasing an updated RX 575 mid-range card would just cause confusion?)

Edit: Butbutbut... the WSA! If the 14nm/12nm Polaris dies are made at GF, then possibly the Polaris RX GPU family could live on longer, at least in certain regions of the world, living on a 12nm shrink, due to costs in fulfilling the WSA agreement? Since Navi is going to be based on 7nm dies made at TSMC, I guess, and probably pricey.

The IO die on Zen2 willl probably almost completely fill the WSA. I see Zen2 selling much much more than Zen1, and I'm believe most agree.

 

[Yotsugi]

The IO die on Zen2 willl probably almost completely fill the WSA

Better, there's multiple of them.
Probably three.

 

[amrnuke]

Better, there's multiple of them.
Probably three.

Multiple IO dies? Multiple WSAs? Multiple Zens?

 

[Mopetar]

Multiple IO dies? Multiple WSAs? Multiple Zens?

Multiple IO dies. We already know that they have separate ones for server and desktop, and it's almost a certainty that they'll make one for their APUs in the future. I wouldn't be surprised if both the next Sony and Microsoft have custom IO dies that tie everything together. Maybe they eventually go that route for their GPUs as well.

 

[Yotsugi]

Multiple IO dies?

Multiple NBs, yes.

I wouldn't be surprised if both the next Sony and Microsoft have custom IO dies that tie everything together

The GPU would be the NB in consoles, for X360 worked that way, see Xenos (aka the grandaddy of all unified shader parts, ever).

 

[AtenRa]

OT but I dont believe they need more than two different I/O Chips, one for Server and one for Desktop is enough.
APUs will be SoCs, that means monolithic design both for Desktop and Mobile.

 

[Yotsugi]

APUs will be SoCs, that means monolithic design both for Desktop and Mobile.

Careful with the predictions, for AMD has two APUs for the next-gen.

 

[maddie]

OT but I dont believe they need more than two different I/O Chips, one for Server and one for Desktop is enough.
APUs will be SoCs, that means monolithic design both for Desktop and Mobile.

Why will APUs have to be monolithic? Curious.

 

[Mopetar]

The more I look at this card, the more it looks like it's the memory bandwidth that's holding it back. If you look at a 1080 vs a 2070, the 2070 has better real world performance, even though a 1080 has better theoretical GFLOPs. Some of this can be attributed to things like Turing supporting FP16 at double rate and picking up advantages there, but the ratio in GFLOPs between the 1650 and the 1060 (at base clocks) is about the same as the ratio between the 2070 and 1080.

The big difference seems to be the bandwidth (there are other possible bottlenecks, but this seems to be it) where the 2070 has 40% more bandwidth than a 1080, but the 1650 only has 66% of (or 33% less) bandwidth the 1060 3 GB. Faster memory probably help considerably.

I think we might eventually get a 1650 Ti as a refresh part that has the full 1,024 cores and faster memory. There's too much performance on the table there and I'm sure that AMD is going to continue offering good value for dollar in the $100 - $250 range with Navi. It does make a little bit of sense for NVidia to hold it back in response.

 

[jpiniero]

Why will APUs have to be monolithic? Curious.

Well AMD's said it won't be using a chiplet.

Edit: So it's either monolithic or they are creating a 7 nm IO die which has the IGP included, sorta like Arrendale.

 

[ZGR]

Any GTX 750 ti owners looking for this as an upgrade?

 

[Yotsugi]

Edit: So it's either monolithic or they are creating a 7 nm IO die which has the IGP included, sorta like Arrendale.

Sorta like Xenos.
ATi did that trick before.

 

[ozzy702]

Any GTX 750 ti owners looking for this as an upgrade?

Not the 1650, but if there is an uncut 1650TI with GDDR6 I'd consider it for one of my use cases at work. I hope that AMD comes out with 7nm NAVI GPU that only runs of the PCIE slot that has 580 performance but I think we'll have to wait until NVIDIA's 3000 series for that, that's too much to ask out of NAVI/AMD. I have an endless supply of HP business towers that I could slap a decent GPU in and sell as a cheap Fortnite gaming rig, but the 1650 just doesn't look very attractive in any aspect so meh.

 

[Yotsugi]

I hope that AMD comes out with 7nm NAVI GPU that only runs of the PCIE slot

Smaller Navi is pretty much that, literally their mobile fodder, P11 replacement.

that's too much to ask out of NAVI/AMD

The opposite.
You're not asking enough.

 

[ozzy702]

Smaller Navi is pretty much that, literally their mobile fodder, P11 replacement.

The opposite.
You're not asking enough.

The increase in efficiency AMD would need to pull off moving from Polaris levels of power consumption to a PCI-E based replacement equal to a 580 are enormous. We're talking a jump far greater than anything we've seen out of AMD in a single generation. Is it possible? Sure, but I'm not going to hold my breath. Radeon VII gives us a sneak peak of 7nm and while nice, it wasn't amazing, so yeah, I have my doubts as to whether a refined 7nm process and a tweaked architecture is going to do it.

I hope I'm wrong and if so I'll buy a ton of them.

 

[Yotsugi]

The increase in efficiency AMD would need to pull off moving from Polaris levels of power consumption to a PCI-E based replacement equal to a 580 are enormous

They're not enormous given they have veeeeeeeeeeeeeeeeeeeeeeeeeeeeery obvious homework to do.

Is it possible?

Yes.
nVidia did it with Maxwell, and everyone drooled about it.

Radeon VII gives us a sneak peak of 7nm and while nice

Vega20 is like a GK210 of sorts, old uArch for HPC because the new one isn't exactly the FP64 champion@mm^2.

tweaked architecture is going to do it

tweaked

Do you all genuinely think they spent a ~4 years long R&D cycle (probably the longest since Southern Islands) for mere tweaks?
Why do you all think only nVidia can do their homework?
Any vendor can, both big and mobile.
Just look at Qualcomm and their jumps from Adreno 200 to Adreno 300 (and 320 to 330).

 

[maddie]

The increase in efficiency AMD would need to pull off moving from Polaris levels of power consumption to a PCI-E based replacement equal to a 580 are enormous. We're talking a jump far greater than anything we've seen out of AMD in a single generation. Is it possible? Sure, but I'm not going to hold my breath. Radeon VII gives us a sneak peak of 7nm and while nice, it wasn't amazing, so yeah, I have my doubts as to whether a refined 7nm process and a tweaked architecture is going to do it.

I hope I'm wrong and if so I'll buy a ton of them.

Where is this coming from. Radeon VII is roughly Vega on 7nm and it used the entire increase in efficiency to improve performance, equaling 30%. 7nm is 1/2 power. Had they stayed at the same clocks, that equates to close to 100% improvement in power consumption. Two sides of the same coin.

If nothing major changed, we would expect a 2X shader core count as the RX560 (which equals an RX570) at the same clocks to use around the same power (75W).

I would hope that you expect there is at the very least a slight improvement in architecture for Navi, so this allows even more performance/W, so I can't begin to understand this repeated statement.

 

[ozzy702]

Where is this coming from. Radeon VII is roughly Vega on 7nm and it used the entire increase in efficiency to improve performance, equaling 30%. 7nm is 1/2 power. Had they stayed at the same clocks, that equates to close to 100% improvement in power consumption. Two sides of the same coin.

If nothing major changed, we would expect a 2X shader core count as the RX560 (which equals an RX570) at the same clocks to use around the same power (75W).

I would hope that you expect there is at the very least a slight improvement in architecture for Navi, so this allows even more performance/W, so I can't begin to understand this repeated statement.

Is it possible? Sure, but again, we haven't seen AMD execute properly on the GPU side for how long? If NAVI really is this good it will be the first time AMD has executed well on the GPU side in how long? I hope we really do see 580 performance in a 75W package from NAVI, if so it will be huge. "In theory" never seems to pan out and I guess I've just become used to AMD under-delivering and as much as I want to see it happen I'm skeptical.

 

[Dribble]

Do you all genuinely think they spent a ~4 years long R&D cycle (probably the longest since Southern Islands) for mere tweaks?
Why do you all think only nVidia can do their homework?
Any vendor can, both big and mobile.
Just look at Qualcomm and their jumps from Adreno 200 to Adreno 300 (and 320 to 330).

As I understand it AMD fired the GCN dev team and moved GPU dev out to China. I'm guessing they spent several years just trying to get going again.

 

[Yotsugi]

Is it possible?

Easily.

Sure, but again, we haven't seen AMD execute properly on the GPU side for how long?

2016?
Yeah.

I hope we really do see 580 performance in a 75W package from NAVI,

Even a 1:1 Polaris shrink can get close.

As I understand it AMD fired the GCN dev team

I'm pretty sure most people are still there, including Mike Mantor.

I'm guessing they spent several years just trying to get going again.

They used China to make their bigger dies since at least Evergreen.

 

[AtenRa]

Is it possible? Sure, but again, we haven't seen AMD execute properly on the GPU side for how long? If NAVI really is this good it will be the first time AMD has executed well on the GPU side in how long? I hope we really do see 580 performance in a 75W package from NAVI, if so it will be huge. "In theory" never seems to pan out and I guess I've just become used to AMD under-delivering and as much as I want to see it happen I'm skeptical.

People forget that Polaris 10 brought a R9 290X (290W TDP) performance down to 150W. It would not be difficult for AMD to release a 7nm 75W TDP card with close to RX 570 performance. In fact im expecting a VEGA 56/64 performance at 150W with NAVI.

 

[happy medium]

All their focus is on Navi right now.

funny enough so is this 1650 Nvidia thread.

 

[Yotsugi]

People forget that Polaris 10 brought a R9 290X (290W TDP) performance down to 150W

That's FinFETs.

funny enough so is this 1650 Nvidia thread.

1650 is so mediocre of a product that discussing it is a waste of time.
Maybe Navi brings something better 75W.