AMD's Future Chips & SoC's: News, Info & Rumours.

Page 118 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.

InvalidError

Titan
Moderator
Well, things are getting interesting. NVIDIA now has a CPU design, by virtue of acquiring ARM. Won't hurt AMD in the short term, but I wonder what long-term effects this is going to have.
Most of the hurting that AMD (and Intel) might see from ARM has already been happening in the form of more client-side computing moving to ARM-powered mobile devices. Who owns ARM won't necessarily make any difference there, especially when you consider that Nvidia already had its own in-house custom ARM cores and low-power IGPs and Nvidia could also have chosen to license ARM's stuff without paying anywhere near 40G$ for it.

I'm still scratching my head over how Nvidia plans to earn its 40G$ back without screwing ARM licensees over... unless screwing ARM licensees over IS the plan.
 
Most of the hurting that AMD (and Intel) might see from ARM has already been happening in the form of more client-side computing moving to ARM-powered mobile devices. Who owns ARM won't necessarily make any difference there, especially when you consider that Nvidia already had its own in-house custom ARM cores and low-power IGPs and Nvidia could also have chosen to license ARM's stuff without paying anywhere near 40G$ for it.

I'm still scratching my head over how Nvidia plans to earn its 40G$ back without screwing ARM licensees over... unless screwing ARM licensees over IS the plan.
Oh, I missed that.

At a first glance, I'd imagine that since now nVidia can "tie in" their own GPU designs to ARM, we'll be probably seeing a lot of new phones and tablets with more nVidia GPUs in them. So the ones that will hurt here will be Qualcomm and Imagination Technologies (PowerVR GPUs). Maybe even the ARM Mali GPUs will stop from being actively developed or will be re-branded as GeForces?

Cheers!
 

InvalidError

Titan
Moderator
At a first glance, I'd imagine that since now nVidia can "tie in" their own GPU designs to ARM, we'll be probably seeing a lot of new phones and tablets with more nVidia GPUs in them.
Nvidia already "tied in" ARM and its own GPU designs with its custom ARM cores in the Tegra line, Nvidia didn't need to buy ARM for that. Nothing other than Nvidia was preventing Nvidia from licensing its GPUs for other SoC designers to use in their own ARM-based SoCs, no need to buy ARM for that either.

Buying ARM does not really enable Nvidia to do anything it couldn't already do before other than soak ARM's licensing profits and potentially screw over the other ARM licensees at some point in the future unless stringent conditions get attached to the deal to ensure Nvidia does not go rogue on ARM licensing and provides a reasonable alternative to Mali if it decides to scrap it or merge with GeForce.
 
Nvidia already "tied in" ARM and its own GPU designs with its custom ARM cores in the Tegra line, Nvidia didn't need to buy ARM for that. Nothing other than Nvidia was preventing Nvidia from licensing its GPUs for other SoC designers to use in their own ARM-based SoCs, no need to buy ARM for that either.

Buying ARM does not really enable Nvidia to do anything it couldn't already do before other than soak ARM's licensing profits and potentially screw over the other ARM licensees at some point in the future unless stringent conditions get attached to the deal to ensure Nvidia does not go rogue on ARM licensing and provides a reasonable alternative to Mali if it decides to scrap it or merge with GeForce.

nVidia has made a few other (less high profile) purchases lately - all related to interconnects and server tech. I think the goal here is for nVidia to make a run at the server market itself with a high performance ARM cpu, custom interconnect fabric and of course nVidia gpu tech. I think an issue ARM has had getting into things like the data centre (aside from compatibility with current software) is that there is yet to be a decent high performance arm based design that can really tackle existing server CPU designs. A few small startups have tried, AMD looked at it but decided to stick with x86 (safer when working with limited resources).

I agree - they could have done this by licencing rather than purchasing - however when you look at the direction everything is moving (tighter and tighter integration), nVidia were looking to be in a rather weak position given they didn't own any CPU tech. Intel have moved into graphics (as well as already offering many other devices and accelerators), AMD have both strong CPU and GPU IP + interconnects and other bits and pieces. nVidia now have a complete portfolio - albeit based on a different ISA.

It will be interesting to see what nVidia does regarding the ARM licencees... they aren't known for sharing their latest tech so I can see them holding on to certain IP moving forward to support their own products.
 

InvalidError

Titan
Moderator
I agree - they could have done this by licencing rather than purchasing - however when you look at the direction everything is moving (tighter and tighter integration), nVidia were looking to be in a rather weak position given they didn't own any CPU tech.
Licensing what? Nvidia already owned all of the ARM licenses it needed to design its own ARM-based CPUs and embed them in its own SoCs, which is exactly what it is doing with its Tegra line. If Nvidia wanted a datacenter-centric ARM-based CPU design, it already owned everything it needed to do so, no need to 'wait' for ARM to put one together. Owning ARM does not really change that either. Nvidia's portfolio was already 'complete' as an ARM licensee too.

Considering Nvidia's anti-competitive history, it is really difficult to imagine this turning out good for consumers.
 
Licensing what? Nvidia already owned all of the ARM licenses it needed to design its own ARM-based CPUs and embed them in its own SoCs, which is exactly what it is doing with its Tegra line. If Nvidia wanted a datacenter-centric ARM-based CPU design, it already owned everything it needed to do so, no need to 'wait' for ARM to put one together. Owning ARM does not really change that either. Nvidia's portfolio was already 'complete' as an ARM licensee too.

Considering Nvidia's anti-competitive history, it is really difficult to imagine this turning out good for consumers.

I fully agree with you there - this move is to benefit nVidia not anyone else.

I think the key though is it's a very different position to be in to hold a license for something than it is to own the IP. Licenses are usually time and / or scope limited - it's possible their license was due for renewal or they wanted / needed access to a larger chunk of ARM IP than they currently have.

I even wonder if it could simply be a protective measure - nVidia do have a complete portfolio thanks to it's ARM license however it is evident ARM were on the table so maybe the move was to ensure that no rival firm could acquire the firm and thus put their licence in jeopardy. I could imagine Qualcom or Apple playing hardball with nVidia if they had acquired ARM. This move secures nVidias access to the IP long term, which is a good thing for long term stability (and thus stock price).
 

InvalidError

Titan
Moderator
I even wonder if it could simply be a protective measure - nVidia do have a complete portfolio thanks to it's ARM license however it is evident ARM were on the table so maybe the move was to ensure that no rival firm could acquire the firm and thus put their licence in jeopardy.
The "license jeopardy" theory is improbable since it applies to the entire ARM ecosystem. I doubt companies like Qualcomm would bet their future on ARM without securing all of the licenses they may possibly require going far enough into the future to cover an eventual transition to something else.

The main way I can imagine Nvidia screwing ARM licensees over is by halting distribution of new HDL and pre-validated silicon IP-cores for future fab processes, thereby forcing ARM licensees to either stick with fab processes that already have ARM IP-cores or design their own ARM CPUs from whatever HDL designs they may have licenses for and have to do the fab-specific silicon optimizations themselves too.
 
nVidia has now complete control of the ARM instruction set. They can tie in whatever they want in terms of "extended operations" (extensions) to shoe in their GPUs or specific logic alongside the reference designs. It's not only about licensing the basic design, but also about controlling the instruction set, which is arguably more important. That was my point.

As for what nVidia will do, no one knows, but closing the ARM ecosystem is not really profitable in the short term or even long term. The licensing money is not small, considering 95% (or around?) of the phone and tablet market is ARM-based.

Cheers!
 
  • Like
Reactions: cdrkf

jaymc

Distinguished
Dec 7, 2007
614
9
18,985
Switching to ARM is more about handling lots of small jobs at very low power rather than using powerful cores to do single tasks or multi-task. So they aren't quite a one-one comparison.

They'll both have their place. AMD even got into the ARM segment themselves some years back and failed spectacularly at adoption.

It's looking like the Chinese may "try" block the ARM deal with Nvidia as part of the escalating trade war:
https://www.eenewseurope.com/news/china-set-block-nvidia-arm-deal

Edit: added the word "try" not sure the Chinese can block this deal they may try an threaten to bar Nvidia products from China but I guess it comes down to how bad Nvidia want ARM right... an I think they want it really bad.. it's a competitor to x86 in general but also in the data center.. imagine where ARM could go with Nvidia's resources pushing the tech. Interesting times.. .big changes if this deal goes through an political backlash as well it seams from China. ARM china may go rogue.
 
Last edited:

jaymc

Distinguished
Dec 7, 2007
614
9
18,985
Level1Ramble: What's Up With Nvidia + ARM? :
View: https://youtu.be/pIthDMsSNGU


"Intel could be depending on AMD to keep ARM at bay.. with Nvidia's resources behind it"

Interesting thought right.. that's if the deal doesn't get blocked I guess which is possible too, where this deal might take us... x86 could be obsolete a lot sooner than expected if Nvidia get it's way that is.

Anyway it's always nice to get Wendel's take on things.
 
Level1Ramble: What's Up With Nvidia + ARM? :
View: https://youtu.be/pIthDMsSNGU


"Intel could be depending on AMD to keep ARM at bay.. with Nvidia's resources behind it"

Interesting thought right.. that's if the deal doesn't get blocked I guess which is possible too, where this deal might take us... x86 could be obsolete a lot sooner than expected if Nvidia get it's way that is.

Anyway it's always nice to get Wendel's take on things.

x86 isn't going anywhere, due to software compatibility reasons. The last chance to abandon it was Itanium, and we know how that ended.

Really it comes down to the same fundamental problem: Emulating the x86/x64 instruction set costs performance, and consumers have already shown (via Itanium) they aren't willing to accept a short-term performance loss while a replacement architecture matures.

ARM will remain powerful in the embedded market (where it's all but forced out PPC at this point) due to Intel's failure to produce an attractive low-power x86 chip, but the server and desktop markets are software-locked to x86-64, and that isn't changing any time soon.
 

InvalidError

Titan
Moderator
Really it comes down to the same fundamental problem: Emulating the x86/x64 instruction set costs performance, and consumers have already shown (via Itanium) they aren't willing to accept a short-term performance loss while a replacement architecture matures.
For many people, this isn't as much of an issue anymore due to languages like C# and Java having platform-agnostic output (at least as an option) that gets locally re-compiled and optimized as native binary so developers can write software without having to worry about platform-specific details.
 
The Ryzen 5000 series announcement was interesting - AMD seem pretty adamant they have now overtaken Intel in gaming and other lightly threaded tasks (e.g. CAD). Looking forward to independent reviews, that 19% average IPC jump sounds impressive on paper.
 
The Zen3 based stuff looks promising alright. Most of those numbers may actually be better in most reviews. I took a look at the fine print for each and using DDR4-3600 for all CPUs, maybe 16GB, was totally standard. I'm sure reviewers can get a bit more performance out of tighter timings in RAM and better cooling; although they did use Noctuas D15's.

Lastly, I found ironic that tease attempt at big Navi with the specific wording they used. Welp, being second is not so bad as long as you're close, I guess? I hope that's the case.

All in all, I think AMD is doing alright. I hope their Zen3 launch is not a bad one, given how scalpers are now using bots to get stock. AMD better take note of nVidia's failure.

Cheers!
 
Lastly, I found ironic that tease attempt at big Navi with the specific wording they used. Welp, being second is not so bad as long as you're close, I guess? I hope that's the case.

Second to which one is the question. I've read rumors that put it anywhere between 'competing with the RTX 3080' to 'just below the RTX 3070'. The former would be awesome. The latter... :(
 
The Zen3 based stuff looks promising alright. Most of those numbers may actually be better in most reviews. I took a look at the fine print for each and using DDR4-3600 for all CPUs, maybe 16GB, was totally standard. I'm sure reviewers can get a bit more performance out of tighter timings in RAM and better cooling; although they did use Noctuas D15's.

Lastly, I found ironic that tease attempt at big Navi with the specific wording they used. Welp, being second is not so bad as long as you're close, I guess? I hope that's the case.

All in all, I think AMD is doing alright. I hope their Zen3 launch is not a bad one, given how scalpers are now using bots to get stock. AMD better take note of nVidia's failure.

Cheers!

Yeah the teaser for the new GPU was interesting - looks like they are in striking distance of the RTX 3080 / 3090 cards - a few people declared it 'DOA' as it's not the fastest, I don't get that attitude. If the card is priced well it should sell and it's nice this gen they are competing in the same tiers rather than having to sidestep the top end as they did with the RX 5700 series.

I'd also bet they might be notably more efficient than the RTX 3000 series, given nVidia is using what appears to be a pretty inefficient Samsung process for them.
 
I doubt Big Navi is going to be more efficient than Ampere at the same performance. Even nVidia had to turn power up by a big amount to keep up with their RTX-shenanigans. If AMD wants to reach parity (not only on performance, but feature-set), they have to play their cards carefully with Big Navi. Just because they have parity in DX12 titles, it won't mean they'll have future parity on RayTracing (when it comes) and DX9-based titles.

I'm being a bit pessimistic, maybe, but it's not without reason. Still, I hope for the best, but expecting the worst as usual.

Cheers!
 
Second to which one is the question. I've read rumors that put it anywhere between 'competing with the RTX 3080' to 'just below the RTX 3070'. The former would be awesome. The latter... :(

They shared a few benchmarks - 61fps in Borderlands 3 at 4k Ultra - a quick search shows this well ahead of the 2080ti (that only gets ~ 40 fps at those settings) but a little behind the 3080 (which scores ~65 fps from what I can find). That would put it nicely ahead of where the 3070 is supposed to land and close (but just behind) the 3080. It probably will depend on the game of course but that doesn't look to be a bad position to be in.

I think it's worth keeping in mind, it's unlikely AMD will release just one SKU - there have been a few performance leaks pointing at different levels, it's likely the 'just below 3070' is a different lower tier card (probably a cut down variant). That's my take anyways, not too long before we find out for certain anyways.
 
Can you post a link to those benches? Must have missed them.

I hope you're right. The GPU market needs to be competitive. Would be great if they have a Big Navi Ultra/Extreme/Super/Whatever under their sleeve that regularly beats the 3080. Of course, NVIDIA will just say, "Hold my beer!" at that point and the market will again be saturated with sub-versions.