nVidia and AMD to make PC Arm chips

Cmaier

Site Master
Staff Member
Site Donor
Joined
Sep 26, 2021
Posts
5,969
Main Camera
Sony

Article is mostly about nVidia, but they really aren’t very good at making CPUs, and unless something has changed, their whole design methodology is wrong. AMD knows what they are doing, however, and I expect that they will make some formidable chips.
 
I find the prospect of Nvidia pairing Arm’s CPU IP with their GPUs most exciting.

Nvidia doesn’t half-ass the software side. It will no doubt have rock-solid drivers at launch and a wide range of software updated to support it. Expect applications like Blender and popular AAAs flying on these things on day one.

I trust Nvidia to deliver a polished, first class experience… but not so much AMD.
 

Article is mostly about nVidia, but they really aren’t very good at making CPUs, and unless something has changed, their whole design methodology is wrong. AMD knows what they are doing, however, and I expect that they will make some formidable chips.
They also don’t say who will be designing their own cores, just that they’ll be designing their own CPUs which could mean just designing the SOC. While even just using off the shelf ARM designs (at first) would be a massive step for AMD, NVIDIA is already doing that for data center, it’s also not clear from this article yet what products AMD be aiming for. Still massive news if a little vague about details.
 
This article is ridiculous:


Look at the per-core speeds and then look at the headline.
Yeah that’s a pretty dubious assertion given those numbers. It’s got a closer to M2 Pro design than M2. We’ll have to see what it’s power draw is but especially single core score is uhhh … not great which is rather more important for this product segment than Nuvia’s original goal of server design. I’m pretty sure though even then that single core score isn’t what they promised with their “simulated” design back in the day. But you already savaged that in a different post.
 
This article is ridiculous:


Look at the per-core speeds and then look at the headline.
Holy 💩, the Geekbench 5 single core score is awful

There must be a problem with the test system - the performance cores can’t be that bad, surely 😬

It’s only competing with A12X https://browser.geekbench.com/v5/cpu/compare/21859127?baseline=21742445
 
nVidia has that wacky Grace/Hopper datacenter thing, so they have something of a leg up. How Grace architecture would perform at the non-server level, though, is not at all clear.
 
Holy 💩, the Geekbench 5 single core score is awful

There must be a problem with the test system - the performance cores can’t be that bad, surely 😬

It’s only competing with A12X https://browser.geekbench.com/v5/cpu/compare/21859127?baseline=21742445
I’m pretty sure their marketing material from before the buyout was promising better than M1 SC performance.
 
This article is ridiculous:


Look at the per-core speeds and then look at the headline.
One of the worst articles I’ve ever seen. Authors clearly have no idea what they are talking about. They are claiming the gen 4 is slower than the gen 2??? I mean I guess it could be an early prototype but it’s nothing to boast about.

I think Qualcomm has an event tomorrow and amusingly, Gerard Williams has created a forum account at Anandtech seemingly.
 
Re: the Geekbench result

If we assume the reported frequency is correct and scale up to 4.2GHz (the supposed max boost clock), it will just barely match an M1 in single core performance (~1675 points)

That would be pretty embarrassing as Snapdragon 8 Gen 3 with Cortex X4 at 3.2GHz is also coming in roughly the same as M1 https://browser.geekbench.com/v5/cpu/21861210

So… Arm managed to get M1 P-core performance at the same frequency as M1 on a similar process node to M1 with (presumably) the same if not better efficiency as M1… but it looks like Qualcomm/Nuvia only managed to get M1 performance at much higher frequency and (presumably) power draw? 😅 I guess we’ll see soonish.
 
Give that it is almost Halloween, they could call the architecture K12 Resurrected...

If nobody at AMD wants to do an Arm chip, I’m happy to go back there and do it. I bet I could get a few more former team members to join, too, if they promise to leave us alone until the chip tapes out.
 
its running under x86 emumation. Notice the name of the CPU says intel.
Could be … however it says specifically Intel pentium 2/3. Is that what Qualcomm chips read as under emulation? It could also just not be reading the cpu name right as this is pre-release for a brand new processor. With pre-release “leaks” especially there could be a hundred different explanations for a given score. But even so, not a great first impression.
 
One of the worst articles I’ve ever seen. Authors clearly have no idea what they are talking about. They are claiming the gen 4 is slower than the gen 2??? I mean I guess it could be an early prototype but it’s nothing to boast about.

I think Qualcomm has an event tomorrow and amusingly, Gerard Williams has created a forum account at Anandtech seemingly.
I think the article might be also mixing GB6 and GB5 scores … those M-series scores are GB6. This is a GB5 score.
 
Re: the Geekbench result

If we assume the reported frequency is correct and scale up to 4.2GHz (the supposed max boost clock), it will just barely match an M1 in single core performance (~1675 points)

That would be pretty embarrassing as Snapdragon 8 Gen 3 with Cortex X4 at 3.2GHz is also coming in roughly the same as M1 https://browser.geekbench.com/v5/cpu/21861210

So… Arm managed to get M1 P-core performance at the same frequency as M1 on a similar process node to M1 with (presumably) the same if not better efficiency as M1… but it looks like Qualcomm/Nuvia only managed to get M1 performance at much higher frequency and (presumably) power draw? 😅 I guess we’ll see soonish.
Unclear: that is labeled as the base clock but that doesn’t mean that’s the clock it ran at - we don’t know what boost clock it was allowed to go to during the test. Lot’s of unknowns.
 
Could be … however it says specifically Intel pentium 2/3. Is that what Qualcomm chips read as under emulation?
That prettymuch has to be the case. It says "12 cores" which would be a large mbd with a dozen PII/IIIs on it with elaborate bus management, because all the PII/IIIs were single core. Also, those would all be 32-bit processors, which do not really appear at all even in GB5.
 
Back
Top