AMD "Navi 31" Memory Cache Die Has Preparation for 3D Vertical Cache?

[btarunr]

AMD possibly has a straightforward path to increasing the performance of the "Navi 31" RDNA3 GPU to power future high-end SKUs, according to semiconductor engineer Tom Wassick. The GPU's main SIMD machinery is located in the Graphics Compute Die (GCD) built on the 5 nm EUV foundry process, surrounded by six Memory Cache Dies (MCDs) built on 6 nm, which each contain GDDR6 memory controllers, and a 16 MB segment of the GPU's 96 MB Infinity Cache memory.

In microscopic observations, Wassick noticed structures on the MCD which he thinks look like an array of through-silicon vias (TSVs), of the kind used in "Zen 3" and "Zen 4" CCDs, to wire out stacked 3D Vertical Cache memory on the L3D (L3 cache die). If the theory holds up, it could be possible for AMD to increase the L3 cache segment size per MCD from 16 MB, and the GPU's overall Infinity Cache memory size. With its RDNA2 graphics architecture (RX 6000 series), AMD significantly enlarged on-die caches on its GPUs, particularly the last-level L3 cache, even giving them the special branding of "Infinity Cache," claiming that they had a big impact in lubricating the memory sub-system, letting GPUs with 256-bit memory buses compete with NVIDIA GPUs with wider 320-bit to 384-bit interfaces.



View at TechPowerUp Main Site | Source

 

[Space Lynx]

It will be interesting to see what the gpu market looks like two years from now.

 

[Chaitanya]

Interesting though no microscope images have been provided even in Twitter thread.

 

[delshay]

Interesting though no microscope images have been provided even in Twitter thread.

True, but even if one was provided, how many users will understand it.

 

[nguyen]

LOL another hyped up tech that only lead up to disapointment

 

[Space Lynx]

LOL another hyped up tech that only lead up to disapointment

What makes you think that, given the success of x3d on CPU side of things?

 

[nguyen]

What makes you think that, given the success of x3d on CPU side of things?

X3D was only good because Intel was being incompetent for 4 straight years
Nvidia however never sit back and relax

 

[TumbleGeorge]

More interesting will be if AMD add local Xilinx IP for even more intelligent internal management of video card resources.

 

[Fluffmeister]

The cost of entry for the current cards is too high, maybe a future SKU with 3D cache and all the annoying niggles ironed out might be more appealing for the money they are asking.

 

[BlaezaLite]

LOL another hyped up tech that only lead up to disapointment

Well, the same could be said about DLSS or Ray Tracing... But that's Nvidia, so we CAN'T do that.

 

[nguyen]

Well, the same could be said about DLSS or Ray Tracing... But that's Nvidia, so we CAN'T do that.

Even if you are ignoring what are the defining features of modern gaming, ADA is still ahead by a mile...

 

[BlaezaLite]

Even if you are ignoring what are the defining features of modern gaming, ADA is still ahead by a mile...

Nvidia is indeed in the lead. If you love paying over the odds for a monstrous brick, that's fine. AMD seem to like letting people get the upper hand and then making a comeback.

 

[Fluffmeister]

Maybe they will go back to a HBM, that was the future once for consumer cards!

 

[sector-z]

They can already make a 7950XTX style easy without 3Dcache like they have do with the 6900 xt.

1) Switch GDDR6 20 Gbps to GDDR6X 24 Gbps chip

2) Put higher Power limite like 450W + 15% slider

3) take best binned silicone add 0.25V and a OC

That was the magical ~ bonus from a 6900 xt to 6950 xt and gain was correct for 100$ more ?

 

[aciDev]

Even if you are ignoring what are the defining features of modern gaming, ADA is still ahead by a mile...

So you are one of those guys playing bar charts on tech sites and scrutinizing screenshots with a magnifying glass...
Guess what? That's not gaming.
The mile you're talking about doesn't matter a inch to the average gamer because outside of bar charts there is no visible difference for the average gamer in the real world. Mind you, I'm not advocating AMD, I'm just saying that outside of technological fanaticism those percentage differences really matter little.

 

[mama]

They can already make a 7950XTX style easy without 3Dcache like they have do with the 6900 xt.

1) Switch GDDR6 20 Gbps to GDDR6X 24 Gbps chip

2) Put higher Power limite like 450W + 15% slider

3) take best binned silicone add 0.25V and a OC

That was the magical ~ bonus from a 6900 xt to 6950 xt and gain was correct for 100$ more ?

They won't do that again. I think this theory on 3D cache is more likely. AMD like to show how they innovate and change up.

 

[napata]

Nvidia is indeed in the lead. If you love paying over the odds for a monstrous brick, that's fine. AMD seem to like letting people get the upper hand and then making a comeback.

It's not like things are all that different with AMD. Same overpriced stuff except they're also lacking in a ton of areas.

It's one step at a time but it's clearly RT is the future.

 

[nguyen]

So you are one of those guys playing bar charts on tech sites and scrutinizing screenshots with a magnifying glass...
Guess what? That's not gaming.
The mile you're talking about doesn't matter a inch to the average gamer because outside of bar charts there is no visible difference for the average gamer in the real world. Mind you, I'm not advocating AMD, I'm just saying that outside of technological fanaticism those percentage differences really matter little.

Nah I'm the guy who play game everyday for the last 26 years, and bought lots of high-end GPUs during that period.

Sure I could have fun with whatever GPU i have got, but I would definitely have more fun with faster and more features packed GPU vs slower one.

 

[TumbleGeorge]

Adding new features can come with a new architecture that's designed to deliver them, it can't just be by inflating frequency and power consumption.

 

[gffermari]

Even if the 3D V cache benefit the gpus, I don't think AMDs problem is the performance.
It varies from the gpu prices, lack of competitive features, drivers issues, no CUDA alternative etc.

 

[Vya Domus]

ADA is still ahead by a mile...

Forget about FPS, FLOPS, GB/s, texture fill rate, etc fellas, a new metric just dropped : ahead by a mile.

What does it mean ? I don't know, but it's by a mile, I guess it depends how much of a fanboy you are.

It varies from the gpu prices

In case you forgot, AMD is still on top in terms of performance/price and Nvidia at the bottom, so this is complete nonsense.

lack of competitive features, drivers issues

Such as ?

no CUDA alternative

They have HIP and OpenCL, HIP is not supported on windows but I doubt this is relevant to the vast majority of users.

 

[Daven]

So you are one of those guys playing bar charts on tech sites and scrutinizing screenshots with a magnifying glass...
Guess what? That's not gaming.
The mile you're talking about doesn't matter a inch to the average gamer because outside of bar charts there is no visible difference for the average gamer in the real world. Mind you, I'm not advocating AMD, I'm just saying that outside of technological fanaticism those percentage differences really matter little.

Excellent post!

Nah I'm the guy who play game everyday for the last 26 years, and bought lots of high-end GPUs during that period.

Sure I could have fun with whatever GPU i have got, but I would definitely have more fun with faster and more features packed GPU vs slower one.

I’m curious what features enable you to have ‘more fun’? I’m honestly asking.

 

[Camm]

X3D was only good because Intel was being incompetent for 4 straight years
Nvidia however never sit back and relax

So overhyped & irrelevant that Nvidia going from Ampere to Lovelace went from 6MB of L2 Cache to 72MB. :\

 

[Tomorrow]

They can already make a 7950XTX style easy without 3Dcache like they have do with the 6900 xt.

1) Switch GDDR6 20 Gbps to GDDR6X 24 Gbps chip

2) Put higher Power limite like 450W + 15% slider

3) take best binned silicone add 0.25V and a OC

That was the magical ~ bonus from a 6900 xt to 6950 xt and gain was correct for 100$ more ?

1. Never gonna happen. At this point G6X is something that only Nvidia uses. Besides with G6 getting faster (24-28Gbps on the pipeline) there is little reason to even use G6X as it's very inefficient in terms of power draw vs speed. So AMD can use G6 that is as fast as G6X.

2. It is not the power limit that is limiting performance. Reference cards need 3x8pin instead of 2x8pin and better official software undervolting to get better temps, lower pawer draw and higher clock speeds.

3. Neither AMD or Nvidia has any binned silicon at this point in terms of GPU's. Only some AIB's do binning (EVGA in the past and Galax now). But it's very resource intensive in terms of manpower.

4. I would also double the infinity cache because the option is there but not used on 7900XTX.

5. Does not add much but upgrade to PCIe 5.0 connector instead of PCIe 4.0 allowing to use fewer lanes on AM5 boards.

 

[Daven]

5. Does not add much but upgrade to PCIe 5.0 connector instead of PCIe 4.0 allowing to use fewer lanes on AM5 boards.

This is the most underreported benefit of going to later PCIe generations. PCIe is a serial interface. Unlike parallel, the number of lanes increases bandwidth linearly. If your device saturates at 4 GB/s you only need 1x PCIe 5.0 lane. This can allow an incredible cost savings as more lanes provide no additional benefit. An older PCIe generation would require more lanes and therefore more cost on the connector and motherboard traces.

Mainstream consumer motherboards could come way down in price if we can get rid of the obligatory x16 PCIe connectors. Just four lanes of PCIe 5.0 is enough for the latest GPUs. Less lanes also means less expensive chipsets or no chipset at all.