NVidia Ada Speculation, Rumours and Discussion

Status
Not open for further replies.
The majority of governments across the world have sent a clear message about their level of concern with the future of humanity by their response to the climate crisis.

We are not, as a modern civilization, looking down the gun barrel because Obama put a 5 year freeze on Nasa's planetary exploration budget in 2012.
That person has a very, very transparent political agenda in their posts and I'm surprised it's even being allowed given how obvious it is.
 
Yeah, exactly. That is one additional aspect to the one-dimensional chart with Dollars per Gate, right?
That is correct. There are multiple variables on the cost side, but when one of the most important variables shows an unusually disturbing trend it needs to be acknowledged and understood. I am very glad that we have had a healthy conversation about this on the forum and there's a general awareness of what's going on.

Beyond costs, there are multiple variables on the business side as well. Both Nvidia and AMD have grown from being scrappy little companies to behemoths that don't want to mess around with low-margin commoditized products. Although most of their margins come from the datacenter side (see $36K H100), they simply don't seem to be interested in playing low-margin games even on the client side (with a few exceptions). That said, gross margin can, and almost always is, redistributed across different tiers in the product stack. And if demand takes a dump then they may be forced to swallow a significant cut to gross margins. If this happens, it will hit Nvidia way harder than AMD because GeForce is still ~50% of their revenue and may therefore require significant business-level restructuring (layoffs etc.).

Who knows -- they may still be able to come up with a creative way to redistribute gross margins, but -- closing the loop back to the cost equation -- the number of creative choices is dramatically reduced when your cost side is horrible and trending worse.
 
Q: I noticed in the presentation that there was no NVlink connector on the cards. Is that completely gone for Ada?

Huang:
There is no NVlink on Ada. The reason why we took it out is because we needed the I/Os for something else. We used the I/Os and the area to cram in as much AI processing as we could. And also, because Ada is based on PCIe Gen 5, we now have the ability to do peer to peer across Gen 5 that’s sufficiently fast that it was a better tradeoff. That’s the reason.
Gen 5? :unsure:
 
Maybe it's there but isn't certified / enabled yet due to a lack of PCIE5 CPUs? Hopper has it, it is a bit weird that Lovelace doesn't.
May also be because it would have required a PCB with a couple of extra layers with PCIe5's tighter signalling requirements; maybe the silicon supports it but adding some more PCB layers added too much cost to the BoM to make it worth it.

Kind of like how AMD's going in the motherboard space for AM5 - all the chipsets and CPUs support PCIe 5.0, but you've got to pay more to get it enabled/validated.

Edit: Just as an example, a lot of the EE articles about testing/validating for PCIe5 say that you can't even use FR4 as a PCB material with PCIe5 and have to move on to more exotic (and presumably much more expensive) ones, even if you manage to keep the layer count the same:

 
Last edited:
Somehow thought that it only had 5.0 for NVMe.

It's the opposite (well kind of), no dedicated x4 PCIe 5.0 lanes off the CPU, so all PCIe NVMe 5.0 support would be if the motherboard chooses to bifurcate the x16 lane. Chipsets also only support PCIe 4.0 as well.

Along those lines though it will be interesting to see how much RT implementation, direct storage implementation (particularly with GPU decompression), and sampler feedback might affect how how much additional data gets transferred via PCIe and whether or not we hit performance implications faster.
 
Status
Not open for further replies.
Back
Top