AMD Navi 23 ‘NVIDIA Killer’ GPU Rumored to Support Hardware Ray Tracing, Coming Next Year

[humbug]

https://www.anandtech.com/show/15581/el-capitan-supercomputer-detailed-amd-cpus-gpus-2-exaflops

That supercomputer was detailed today and uses AMD CPUs and GPUs.

This is quite interesting.....

But perhaps the biggest part of today’s reveal is the interconnect. For the first time AMD is naming their 3rd generation Infinity Fabric, which will be used to connect the processors within each blade. Like Frontier, El Capitan will be running in a 4:1 configuration, with four GPUs hooked up to each CPU. For Infinity Fabric 3.0, AMD is promising further improvements to inter-chip bandwidth and latency. However the most interesting claim is that these IF 3.0 device nodes will support unified memory across the CPU and GPU, which is something AMD doesn’t offer today. Indeed even Frontier is only slated to offer coherency between the processors which is a step below a true unified memory model. The devil is in the details of course – a unified memory system does not necessarily mean fast access to other devices’ memory – but this stands to be a major leap for AMD as a unified memory system can improve both the ease in programming such a system, and improving its performance when running heterogeneous workloads.

On a side note Anand Tech should really know better than to write that as if Unified Memory Architecture (UMA) is new to AMD, it isn't, Its what a Heterogeneous System Architecture (HSA) enabled Accelerated Processing Unit (APU) is, sorry for the acronym soup, these are things people may have heard of but don't know their significance.

The CPU and GPU share eachothers FP units connected through their memory cache and IMC, what that means is the architecture can use either the serial processing power of the CPU or the parallel processing power of the GPU for application that could normal only run one or the other, so a workload can run on the GPU where its much faster than it otherwise would be limited to the CPU, the distinction between CPU and GPU no longer exists.

APU's have on die GPU's which makes that interconnect easier, however Zen 2 has an external IMC to the cores, what makes it work as if its a monolithic die is Infinity Fabric, its really not far-fetched to connect the memory systems of CPU and GPU together using IF, if fact its a logical next step.

Examples of HSA UMA APU's in action.

 

[TNA]

 

[pmc25]

Interesting that they're not specifying which node for RDNA3 (2021), whereas Zen4 (2021) is specified as 5nm. I assume they don't yet know if they will have enough 5nm wafer supply to do RDNA3 on it ... so it could be 6nm EUV (shrink of 7nm EUV), or it could be 5nm.

[From AMD Analysts Day]

 

[Grim5]

Interesting that they're not specifying which node for RDNA3 (2021), whereas Zen4 (2021) is specified as 5nm. I assume they don't yet know if they will have enough 5nm wafer supply to do RDNA3 on it ... so it could be 6nm EUV (shrink of 7nm EUV), or it could be 5nm.

[From AMD Analysts Day]

So has it been confirmed that Ryzen 4000 is only coming 2021? Sounds like based on what was said at the INvestor Day

 

[pmc25]

So has it been confirmed that Ryzen 4000 is only coming 2021? Sounds like based on what was said at the INvestor Day

Ryzen 4000 is this summer. Ryzen 5000 is next summer. Ryzen 4000 is Zen 3 (7nm+). Ryzen 5000 is Zen 4 (5nm). Almost certainly Ryzen 5000 will not be the same platform / socket given the major changes to IF and likely 3D stacking.

 

[CAT-THE-FIFTH]

So has it been confirmed that Ryzen 4000 is only coming 2021? Sounds like based on what was said at the INvestor Day

https://videocardz.com/newz/amd-speaks-rdna2-rdna3-zen3-and-zen4-announces-new-roadmaps

Zen3 is late 2020. AMD will have a new cDNA uarch for its commercial GPUs.

A leak of a new AMD card.

 

[CAT-THE-FIFTH]

 

[CAT-THE-FIFTH]

Stream is here:

https://www.youtube.com/watch?v=sRG_Hj2u55s

AMD says that DXR1.1 was co-architected and co-developed between them and MS.

The cards will be seen later this year.

 

[shankly1985]

Well this all came to a surprise I had no idea this was going on today Give me RDNA 2, please Happy I didn't go 5700 XT now

 

[pmc25]

Only real relevant details re: RDNA2 that's new are that they're claiming +50% perf per watt over RDNA1.

7nm+, VRR, RT, coming soon etc we already knew.

We also now know that RDNA3 is due next year.

I doubt they'll tease the RDNA2 products at the end of the presentation ... but Su does like doing that, so it's possible. Think the presentations are only half way through though, as they've now gone to intermission.

 

[CAT-THE-FIFTH]

Only real relevant details re: RDNA2 that's new are that they're claiming +50% perf per watt over RDNA1.

7nm+, VRR, RT, coming soon etc we already knew.

We also now know that RDNA3 is due next year.

I doubt they'll tease the RDNA2 products at the end of the presentation ... but Su does like doing that, so it's possible. Think the presentations are only half way through though, as they've now gone to intermission.

Also increased IPC and higher clockspeeds.

 

[gpuerrilla]

Well this all came to a surprise I had no idea this was going on today Give me RDNA 2, please Happy I didn't go 5700 XT now

Second that, on a positive vibe, currently on a vega; the 5600xt is a leap in power efficiency so the newer cards should offer a decent upgrade instead of these ****** 10% improvements that seem to have stagnated the industry of late.

 

[TonyTurbo78]

Well this all came to a surprise I had no idea this was going on today Give me RDNA 2, please Happy I didn't go 5700 XT now

Preach brother

 

[Random Guy Esq.]

So from that sketch, are they going to go for a dual fan config for reference cards I wonder?

 

[Tute]

So from that sketch, are they going to go for a dual fan config for reference cards I wonder?

Apparantly no blower cards.

scottherkelman

VP & GM Radeon Business UnitScore hidden·25 minutes ago

There will be no blower reference fans for gamers on next gen. So you are correct

 

[Random Guy Esq.]

Good news!

 

[TNA]

Good news!

+1

 

[Grim5]

The new Xbox is looking very efficient too

not as hungry as we first thought

12tflop sounded like a monster

now we know to hit 12tfloo the Xbox RDNA 2 gpu only needs 175w. Compare to 5700xt that needs 225w to hit 9tflop

 

[pmc25]

Interesting that RDNA2 appears to be 7nmP rather than EUV. Several possibilities why:

Cost / benefit of porting to partial EUV before a design with full EUV layers (5nm) wasn't worth it for the much larger (than CPU) GPU dies.

7nmP availability is good and it performs well, and is no doubt cheaper.

7nm EUV availability isn't good enough to fulfil CPU & GPU side atm.

Edit: PS / XB eating all the 7nm EUV wafers.

 

[Panos]

The new Xbox is looking very efficient too

not as hungry as we first thought

12tflop sounded like a monster

now we know to hit 12tfloo the Xbox RDNA 2 gpu only needs 175w. Compare to 5700xt that needs 225w to hit 9tflop

Yup. Is bonkers, also AMD is very conservative to their announcement of 50% (as usual)
If all we know is accurate, including MS announcement, doing the maths 12tf at 175w means 15.7-16tf at 225w a +70% perf over the 5700XT level card!!!!!
A 275-279W card (same power level with 2080Ti) would be 20tf. Over twice as fast than the 5700XT.

----------------
Before someone starts blabbing about Tf isn't indicator blah blah, GCN TF, blah blah, Vega blah.
We talking about RDNA 1 and RDNA 2. Worst case scenario the performance between the two is 1-1 as is the same architecture in it's core. There is absolutely no chance RDNA2 terraflop been slower in real gaming performance than RDNA1 terraflop.....