[Rumor, Tweaktown] AMD to launch next-gen Navi graphics cards at E3

[lifeblood]

12th december confirmed.

Hopefully it will be the full Navi 14 with all 24 CU's enabled. That should be an interesting contest.

 

[GodisanAtheist]

I wonder if that was the original plan. Let’s be honest, AMD’s GPU division has been marking time since original Polaris. AMD focused (ie, spent all available R&D money) on ZEN while leaving the GPU division on financial life support. The 500 series was just a zero effort refresh while the RX 590 was a zero effort refresh on a slightly tweaked process. Vega and Radeon VII were an underfunded halfhearted effort for the sole purpose of crying out “We’re not dead yet!”. I suspect Navi has always been the real comeback effort.

What I’m curious about is whether RDNA v1 is just an intermediate step, another “We’re not dead yet!” moment until they can get RDNA v2 out the door, or if v1 was really part of the original plan. Was v1 the end-all-be-all or did they originally plan on v1 being the low to mid architecture and v2 being the mid to high architecture? How did Nvidia’s Ray Tracing effect their plans?

In the end it doesn’t matter, obviously RDNA v1 is a good architecture which, unlike Vega, very definitely is a step forward in performance per watt and in raw performance as well.

-This exactly. AMD is not in any financial position to have throwaway "half step" gpus. RDNA1 is going to be milked like the prize cow and will likely filter down as RDNA2 cards are introduced above them.

I suspect they originally planned on RDNA v1 being the console architecture, but then ray tracing threw a spanner in the works as the console makers demanded in their gpus. At which point there has been a rush to add it onto RDNA.

Given that both AMD and NV we're heavily involved in the DXR development process, it's not like AMD didn't know RT was coming or got caught flat footed, they likely just aren't ready to go with their RT arch thanks to a cash strapped, smaller Dev team.

Likely AMD's cut-off point for acceptable RT performance was at the 5700XT level, while NV tried to push it down a little below that performance line with the 2060 and settled on the cutoff being the 1660ti. The 5700 line was done and waiting on "RDNA 2" and instead of sitting on it, we got a half baked launch.

 

[Stuka87]

I wonder if that was the original plan. Let’s be honest, AMD’s GPU division has been marking time since original Polaris. AMD focused (ie, spent all available R&D money) on ZEN while leaving the GPU division on financial life support. The 500 series was just a zero effort refresh while the RX 590 was a zero effort refresh on a slightly tweaked process. Vega and Radeon VII were an underfunded halfhearted effort for the sole purpose of crying out “We’re not dead yet!”. I suspect Navi has always been the real comeback effort.

What I’m curious about is whether RDNA v1 is just an intermediate step, another “We’re not dead yet!” moment until they can get RDNA v2 out the door, or if v1 was really part of the original plan. Was v1 the end-all-be-all or did they originally plan on v1 being the low to mid architecture and v2 being the mid to high architecture? How did Nvidia’s Ray Tracing effect their plans?

In the end it doesn’t matter, obviously RDNA v1 is a good architecture which, unlike Vega, very definitely is a step forward in performance per watt and in raw performance as well.

Well, this isn't entirely true. RDNA has been in development for many years. AMD strangled money away from the GPU division back when Raj was in charge of things. Its obvious AMD started to push money back into the Radeon Group, otherwise we wouldn't have RDNA. But the life cycle is so long, that it was not obvious until RDNA actually shipped. Polaris was good, but they had to make it last until RDNA could hit production, which is why we ended up with the 500 series. Radeon VII was just them giving a home to chips that could not quite cut it as a Pro series chip, while at the same time helping them learn the nuances of 7nm chips.

RDNA is a mostly new ISA (It does borrow bits from GCN), RDNA v2 is not yet another new ISA, its just an iterative improvement on the original. Much like GCN v1.1 was an improvement on 1.0. But the ISA is not to be confused with the architecture of the chips themselves. Navi is the chip, RDNA is the ISA.

 

[Glo.]

https://twitter.com/x/status/1202385013297184768

 

[beginner99]

In the end it doesn’t matter, obviously RDNA v1 is a good architecture

I wouldn't call matching performance/watt with a node advanatge as "good architecture"...

 

[Stuka87]

I wouldn't call matching performance/watt with a node advanatge as "good architecture"...

Navi has better Perf:Watt than Polaris or Vega. It does not simply match it, its better than it. And it would be even better if AMD didn't overvolt all their chips. Something we as consumers can fix, but a bit annoying for out of the box products.

 

[Glo.]

https://twitter.com/x/status/1202576651135025152

Radeon RX 5500 XT with 1408 ALUs confirmed.

 

[VirtualLarry]

Radeon RX 5500 XT with 1408 ALUs confirmed.

So, gunning for the 1660 Super then? Either cheaper, better Performance/Watt, or what? More VRAM (8GB GDDR6?)

If it's within 10%, maybe 5%, and has 2GB extra VRAM, I could see it succeeding.

Edit: Then again, has the 8GB VRAM on RX 570/580/590 cards, actually DONE ANYTHING (*), compared to a 6GB GTX 1060 or GTX 1660/Super/Ti?

(*) Other than a marketing bullet-point to gamers, that is.

 

[Glo.]

What is also confirmed: RX 5500 XT will have 199$ MSRP.

 

[uzzi38]

What is also confirmed: RX 5500 XT will have 199$ MSRP.

By who?

To my knowledge that has not been confirmed, but is guesswork from people looking at Chinese pricing, who have sales tax, and are directly converting the price.

I'll say this now: $199 is DoA if MSRP. Simple as.

 

[VirtualLarry]

What is also confirmed: RX 5500 XT will have 199$ MSRP.

Might be worth it... for mining Grin... if it has 8GB GDDR6. Would be the cheapest card with 8GB GDDR6 on the market.

 

[Stuka87]

By who?

To my knowledge that has not been confirmed, but is guesswork from people looking at Chinese pricing, who have sales tax, and are directly converting the price.

I'll say this now: $199 is DoA if MSRP. Simple as.

If it is $199, and it matches the 1660 Super, how is it DoA? Typically a card that matches performance and is cheaper is a good thing.

 

[uzzi38]

If it is $199, and it matches the 1660 Super, how is it DoA? Typically a card that matches performance and is cheaper is a good thing.

Because it's not going to match the 1660 Super. Performance wise it's up against the 1650 Super.

 

[Glo.]

Because it's not going to match the 1660 Super. Performance wise it's up against the 1650 Super.

Apart from the outliers, it is around GTX 1660. And those outliers are making it look worse than RX 5500 actually is.

If you will look not only on performance summary on Techpowerup, but genuinely look at every game, you will see that it is closer to GTX 1660, rather than GTX 1650.

TPU is not he best place to draw conclusions about any GPUs performance...

 

[uzzi38]

Apart from the outliers, it is around GTX 1660. And those outliers are making it look worse than RX 5500 actually is.

If you will look not only on performance summary on Techpowerup, but genuinely look at every game, you will see that it is closer to GTX 1660, rather than GTX 1650.

TPU is not he best place to draw conclusions about any GPUs performance...

Where did I say it was matching the 1650?

The 1650 Super is virtually 1660 performance.

 

[mohit9206]

Are there going to be any game bundles? They need to give some good games to attract people away from Nvidia.

 

[Bouowmx]

Radeon RX 5500 XT variant with same core count as standard variant, clocked to hell at 1.9-2.0 GHz, for 1.06x performance. Sounds about right

 

[uzzi38]

Radeon RX 5500 XT variant with same core count as standard variant, clocked to hell at 1.9-2.0 GHz, for 1.06x performance. Sounds about right

And 8GB VRAM vs 4GB.

It'd be worth picking up at $170, but at $199 its pretty bad value.

 

[Glo.]

And 8GB VRAM vs 4GB.

It'd be worth picking up at $170, but at $199 its pretty bad value.

Why don't we wait for release drivers, and the GPU itself to show up to competition, before we call it a bad value, eh?

 

[soresu]

Radeon RX 5500 XT variant with same core count as standard variant, clocked to hell at 1.9-2.0 GHz, for 1.06x performance. Sounds about right

Really confusing considering the larger Navi 10 is used at full count in 5700 XT, yields should be great for a chip only 160 mm2 compared to 251 mm2.

Even if they are saving the absolute best dies for Apple there should still be enough for a PC SKU at 24 CU/12 WGP.

 

[uzzi38]

Why don't we wait for release drivers, and the GPU itself to show up to competition, before we call it a bad value, eh?

Because the 5500 in OEMs has been tested, which should tell you enough, but also it's easy enough to look at the 5700 and 5700XT and extrapolate how the 5500 series will perform.

 

[uzzi38]

Really confusing considering the larger Navi 10 is used at full count in 5700 XT, yields should be great for a chip only 160 mm2 compared to 251 mm2.

Even if they are saving the absolute best dies for Apple there should still be enough for a PC SKU at 24 CU/12 WGP.

Those 2 extra CUs are deadweights with the bandwidth available to the 5500.

But in any case, this is artificial segmentation just to keep Apple happy more than anything else. It's been done in the past, it was always going to happen again.

 

[Glo.]

Because the 5500 in OEMs has been tested, which should tell you enough, but also it's easy enough to look at the 5700 and 5700XT and extrapolate how the 5500 series will perform.

And the differences bwetween Reference and AIB cards were big enough to say they were exactly the same GPU, eh?

 

[uzzi38]

And the differences bwetween Reference and AIB cards were big enough to say they were exactly the same GPU, eh?

The major differences between reference and AIB models are in the performance of the coolers in terms of acoustics and temps, the difference in performance is minor at best.

In case you hadn't payed attention to reviews there, core overclocks are mostly useless, if nothing else, they were very far from being worth it. Memory overclocking was limited via the drivers.

Now consider one thing: the 5500 has less bandwidth per CU than the 5700XT thanks to more CUs in the singular Shader Engine, only the same bandwidth for that Shader Engine that both SEs in the 5700XT would get individually.

Oh, and less cache per CU as well. Would be silly of me to forget that.

So yeah, you can estimate pretty easily where the 5500XT will perform. It doesn't take a genius to do so.

 

[DeathReborn]

Why don't we wait for release drivers, and the GPU itself to show up to competition, before we call it a bad value, eh?

Because the 5500 in OEMs has been tested, which should tell you enough, but also it's easy enough to look at the 5700 and 5700XT and extrapolate how the 5500 series will perform.

And the differences bwetween Reference and AIB cards were big enough to say they were exactly the same GPU, eh?

The OEM cards are AMD Reference models, the drivers were provided by AMD (so if they aren't good enough blame AMD) & the performance difference between Reference & AIB is what, 3-5%? That will take it to 1650S/1660 performance levels & not 1660S levels but for more power and a higher price (if $179+ is the price).