Apple is an existential threat to the PC

[PSman1700]

Lmao, what are you smoking. There’s no way, with current prices, to build 2 * 3090, 5950X, and 128GB of RAM (plus storage and cooling) for the price of a Studio.

Ye, here we pay over 6/7000 dollars for a 64core ultra. Not everyone is living in the US.

And, again, as documented by Andrei F, Geekbench sucks for M1 benchmarks. It doesn’t even scale properly going from M1 to M1 Pro.

Which is abit strange since its the holy grail of benchmarks on (apple) mobile devices, including A15 and m1. But there they have qualcomm competition…

You’re back to your old tricks again: using Blender and gaming. Why are we not surprised.

Some above started comparing relative gaming performance on the ultra vs a 3080/3090. It seems rather intresting to some as to how it performs in gaming.
I mentioned more than just blender though.

EDIT: Trying to even compare M1 Ultra’s media encoding capabilities to anything from Nvidia/AMD is a sad cope. M1 wipes the floor in actual content creation.

Depends on what apps and codecs are being used. Nvidia/amd could actually be much faster even for those tasks.

 

[Flappy Pannus]

Considering the fact that M1 Ultra has way bigger GPU (when counting both chips) than anything on desktop, is it really that noteworthy?

It's noteworthy to me in that in this case, it's a multi-die GPU solution that's scaling near 2X linearly in a game. I'm saying that it's potentially a good indicator that something resembling a chiplet-style GPU architecture can indeed work without additional coding vs the apps we already know scale in such configs when they're targeted by developers. There are obviously a lot of differences compared to PC GPU's atm and ray tracing throws in additional complexity the Ultra doesn't have to deal with, but in years past getting 70% scaling from something like SLI was a 'big win' in terms of efficiency.

Assuming my post was intended to be platform contention material, it's still noteworthy even compared to that PC due to the minescule noise/power budget by comparison. However that wasn't my intention, either that PC or the Studio are well out of my price range and the Mac is obviously not a great gaming platform for many reasons outside of performance.

 

[Flappy Pannus]

Remember that is running through Rosetta 2.

I know, part of the reason I'm basically ignoring the 1080p results. It's likely a CPU bottleneck at that point where Rosetta's encumbrance is more noticeable.

 

[Silent_Buddha]

I just noticed that the M1 Ultra die area is a massive 840 mm^2. That's even more massive than what I thought was an absurdly large 628 mm^2 RTX 3090 die.

Ryzen 5700G is a paltry 180 mm^2 in comparison. While I doubt AMD would ever make an 840 mm^2 APU/SOC, I do wonder how such a beast would perform using newer CPU and GPU cores on the same 5 nm node as the M1 Ultra.

Regards,
SB

 

[Flappy Pannus]

I just noticed that the M1 Ultra die area is a massive 840 mm^2. That's even more massive than what I thought was an absurdly large 628 mm^2 RTX 3090 die.

Ryzen 5700G is a paltry 180 mm^2 in comparison. While I doubt AMD would ever make an 840 mm^2 APU/SOC, I do wonder how such a beast would perform using newer CPU and GPU cores on the same 5 nm node as the M1 Ultra.

Regards,
SB

Doesn't that die size also account for the space to hold up to 128GB of ram, or are you excluding that?

That's the rub with any APU - what do you use to get the bandwidth necessary to drive it? Apple has a massively wide bus to the DDR5, what would AMD theoretically use?

 

[Pressure]

Doesn't that die size also account for the space to hold up to 128GB of ram, or are you excluding that?

That's the rub with any APU - what do you use to get the bandwidth necessary to drive it? Apple has a massively wide bus to the DDR5, what would AMD theoretically use?

A single M1 Max chip has a die size of 432mm2. RAM is separate.

I just noticed that the M1 Ultra die area is a massive 840 mm^2. That's even more massive than what I thought was an absurdly large 628 mm^2 RTX 3090 die.

Ryzen 5700G is a paltry 180 mm^2 in comparison. While I doubt AMD would ever make an 840 mm^2 APU/SOC, I do wonder how such a beast would perform using newer CPU and GPU cores on the same 5 nm node as the M1 Ultra.

Regards,
SB

Remember that a lot of transistors are also used on I/O (NAND flash controller, Thunderbolt/USB4, legacy USB3.1, ethernet), the sizeable neural engine, and four chunky 128-bits memory channels as well as four large blocks of system-level cache.

I'm not sure if the pictures of the die is true but I see a lot of replicated silicon on there, for example we have two neural engines on the M1 Max.

 

[Gubbi]

It's a big die, Apple will get around 120 out of a 300mm wafer which corresponds roughly to $150 cost per die. I don't know the yield, but there are a lot of salvage bins, so I believe it will be fine overall.

For the 20 core/64GB/1TB Mac Studio, you have $300 worth of Si in the main SOC, less than $100 for 1TB flash and $320 worth of DDR5 (long term contract price with a 20% premium over DDR4 prices), and they sell it for $4000. The margins are solid.

The Mac Studio is a workstation in a consumer electronics form factor; high performance, compact and, relatively, low power. It is going to kill in that market.

Yeah, you can build a faster workstation from TR 5000 Pros and 3090s, it won't be cheaper and it will double up as a space heater with 500-1KW of power consumption.

By carving out the low end of the workstation market, Apple is undermining the economics of the workstation TR and Xeon models: AMD and Intel have nowhere to get rid of their salvage SKUs, they can dump the price on lower end SKUs, but the motherboards for these SKUs, and other components, are so costly that the total cost of a system is hardly affected.

IMO, Intel or AMD either follow suit or they give up the market.

Cheers

 

[Pressure]

It's a big die, Apple will get around 120 out of a 300mm wafer which corresponds roughly to $150 cost per die. I don't know the yield, but there are a lot of salvage bins, so I believe it will be fine overall.

For the 20 core/64GB/1TB Mac Studio, you have $300 worth of Si in the main SOC, less than $100 for 1TB flash and $320 worth of DDR5 (long term contract price with a 20% premium over DDR4 prices), and they sell it for $4000. The margins are solid.

The Mac Studio is a workstation in a consumer electronics form factor; high performance, compact and, relatively, low power. It is going to kill in that market.

Yeah, you can build a faster workstation from TR 5000 Pros and 3090s, it won't be cheaper and it will double up as a space heater with 500-1KW of power consumption.

By carving out the low end of the workstation market, Apple is undermining the economics of the workstation TR and Xeon models: AMD and Intel have nowhere to get rid of their salvage SKUs, they can dump the price on lower end SKUs, but the motherboards for these SKUs are so costly that the total cost of a system is hardly affected.

IMO, Intel or AMD either follow suit or they give up the market.

Cheers

It's damn smart in the sense that you don't pay for all the I/O chips, as they are all integrated in the SoC. Then you are only dependent on RAM and NAND flash prices, which Apple also has locked down tight.

The same thing is happening to the iPhone where it is rumoured they will launch their own modem for cellular data within the next two years.

 

[Wesker]

Ye, here we pay over 6/7000 dollars for a 64core ultra. Not everyone is living in the US.

I don’t live in the US either.

There is no way you can use any mainstream retailer — US or otherwise — and build the rig you stated for less than a Studio. You’re talking complete fud.

Which is abit strange since its the holy grail of benchmarks on (apple) mobile devices, including A15 and m1. But there they have qualcomm competition…

Says who?
You just haven’t been paying attention.

Some above started comparing relative gaming performance on the ultra vs a 3080/3090. It seems rather intresting to some as to how it performs in gaming.
I mentioned more than just blender though.

One game — one of the three games that actually runs on macOS.

Depends on what apps and codecs are being used. Nvidia/amd could actually be much faster even for those tasks.

Such as?

 

[Pressure]

Larian Studios have done a really great job of optimising Baldur's Gate 3 for Metal and Arm.

It could be a reference point for comparing game performance?

Honestly if my primary goal where to play games I would buy a console or PC.

 

[PSman1700]

I don’t live in the US either.

There is no way you can use any mainstream retailer — US or otherwise — and build the rig you stated for less than a Studio. You’re talking complete fud.

No, i am going after MSRP, not the (still) inflated ETH pricings, as that skews things up. Where i live we pay 7000 dollars, at the least, for the 64core model. I dont know if 128gb makes a huge difference but its going to add to that, aswell if you want the full nvme speeds, then you need go for the bigger drives.

Says who?

Geekbench is generally being used, by virtually every youtuber out there to gauge mobile phone performance. In Geekbench Apple almost always wins there. In other benchmarks, they may or may not win. Whatever that means in mobile (or any) space anyways. its real world performance that matters, people arent buying hardware based on synthetic benchmarks and promises i hope.

One game — one of the three games that actually runs on macOS.

Which means theres zero evidence of the M1/mac being competitive with any AMD/NV gpu in its class. It might happen oneday, it might not. Remember, you never buy hardware based on what it might perform like in the future, you buy it based on what it does now. If you start comparing games/3d rendering, then yeah, we can do that.

Such as?

In Proress, Apple's media accelerators will perform better. In apps that support AV1, a tiger lake setup will do better. Theres quite many videos out there where a RTX setup is faster for content creation than the M1 Max/ultra, and vice versa (for content creation). It all depends on if you use Apple's (optimized) software or not.

Yeah, you can build a faster workstation from TR 5000 Pros and 3090s, it won't be cheaper and it will double up as a space heater with 500-1KW of power consumption.

By carving out the low end of the workstation market, Apple is undermining the economics of the workstation TR and Xeon models: AMD and Intel have nowhere to get rid of their salvage SKUs, they can dump the price on lower end SKUs, but the motherboards for these SKUs, and other components, are so costly that the total cost of a system is hardly affected.

IMO, Intel or AMD either follow suit or they give up the market.

To be honest, intel/AMD and NV have different hardware for that market. RTX3090 and 5950x are more enthusiast/high end gaming products, not so much for the market we talk about here.
And again, it depends on what workloads are required.

 

[MfA]

EDIT: Trying to even compare M1 Ultra’s media encoding capabilities to anything from Nvidia/AMD is a sad cope. M1 wipes the floor in actual content creation.

Ease of working with 8K sources is definitely a huge plus for Apple, but it's mostly a self inflicted wound on the part of Avid. If Avid got off their asses and started using something like this for their archival level compression they'd have a lot easier time competing, even at 8K 2D video just isn't an impressive data stream to a modern GPU if you use it well ... as I said, it all comes down to bad software ProRes acceleration hardware is definitely nice, but not strictly necessary.

But that's just the ProRes side. For a final encode to H.264/H.265 nothing beats the open source software initially created by some Doom9 amateurs. Hardware is laughable by comparison. That's why handbrake is more relevant than it seems at first glance, there is no alternative. Don't let the amateurish looks deceive you.

 

[Flappy Pannus]

IMO, Intel or AMD either follow suit or they give up the market.

Cheers

I think the rest of the PC industry is moving in this direction, and the proposal of a chiplet interconnect standard which seems to have backing from almost every major player in the industry is further evidence that they recognize the threat, if not from Apple specifically but low-power SOC's in general for use in desktop/datacenter roles. That of course, and just the scalability of chiplets.

I would not expect significant progress, at least for affordable options in the consumer x86 market until more fabs come online around 2025 though.

 

[Ailuros]

They might have moved the entire rasterization stage into unified shader programs, small change, it wouldn't look much different. As Epic proved, hardware triangle rasterizers are relics of the past which have become a boat anchor more than an asset.

I doubt it's the case for Apple for their latest SoCs and of course for future hw we'll now only when its released and analyzed by 3rd parties. If future hw should be IMG IP based then the answer is again no; at least their block diagrams show a clear geometry/raster block below the dedicated RT hw block.

One of the problems Apple's current GPU designs have IMHO right now is a quite awkward GFLOPs/mm2 ratio. I think in that department QCOM's Adreno GPU IP is more than quite a bit ahead, albeit I haven't read about their GPU architectural details for quite some time now. If my dumb math isn't wrong then Apple currently has 64 FP32 FLOPs/clock with 2 TMUs, while the newest IMG generation should be 4x times higher in FLOPs than that. While I don't disagree with your post above, it's also a fact that raster units are relatively small in area. What would it account as a percentage in a tiny mobile single GPU "core" block? 3 or 5%; peanuts IMHO compared to what texture mapping units can account for.

Assuming one would see the Ultra SoCs GPU blocks as one single block, you might have 256 ALU clusters (2 FMACs each) with 512 TMUs and 256 ROPs.

 

[MfA]

I think the rest of the PC industry is moving in this direction, and the proposal of a chiplet interconnect standard

That's not really the problem for PCs. The problem is that Apple is gearing to sell Studio's in massive numbers and the PC workstation market is mostly boutique. You can buy a 24 core TR workstation with a decent GPU around 4K, but not from the big boys ... you have to go for lower margin system assemblers.

Intel and AMD will have to respond by letting completely go of artificial gating of ECC and HP&Co will have to give up margins.

 

[Wesker]

Usual disclaimer: they're not exactly the best source, but they seem to have done a decent comparison here.

No, i am going after MSRP,

So you're in fantasy land. Cool.

Geekbench is generally being used, by virtually every youtuber out there to gauge mobile phone performance. In Geekbench Apple almost always wins there. In other benchmarks, they may or may not win. Whatever that means in mobile (or any) space anyways. its real world performance that matters, people arent buying hardware based on synthetic benchmarks and promises i hope.

Most YouTubers are idiots (see my disclaimer above).

Also, you seem to be confusing GeekBench CPU benchmarks and GeekBench GPU benchmarks. The CPU benchmarks don't have issue scaling across architecures and core counts. The GPU benchmarks are rubbish and have long been disproven by Andrei F.

Which means theres zero evidence of the M1/mac being competitive with any AMD/NV gpu in its class. It might happen oneday, it might not. Remember, you never buy hardware based on what it might perform like in the future, you buy it based on what it does now. If you start comparing games/3d rendering, then yeah, we can do that.

You're a broken record.
No one is buying any Apple product for gaming. The comparisons aren't even theoretically interesting since the 3D API implementation in macOS is a joke.

In Proress, Apple's media accelerators will perform better. In apps that support AV1, a tiger lake setup will do better. Theres quite many videos out there where a RTX setup is faster for content creation than the M1 Max/ultra, and vice versa (for content creation). It all depends on if you use Apple's (optimized) software or not.

What are the use cases here? Watching YouTube and Netflix?

Post the benchmarks.

To be honest, intel/AMD and NV have different hardware for that market. RTX3090 and 5950x are more enthusiast/high end gaming products, not so much for the market we talk about here.
And again, it depends on what workloads are required.

So it's weird that you keep posting gaming and Blender benchmarks in this thread.

Ease of working with 8K sources is definitely a huge plus for Apple, but it's mostly a self inflicted wound on the part of Avid. If Avid got off their asses and started using something like this for their archival level compression they'd have a lot easier time competing, even at 8K 2D video just isn't an impressive data stream to a modern GPU if you use it well ... as I said, it all comes down to bad software ProRes acceleration hardware is definitely nice, but not strictly necessary.

But that's just the ProRes side. For a final encode to H.264/H.265 nothing beats the open source software initially created by some Doom9 amateurs. Hardware is laughable by comparison. That's why handbrake is more relevant than it seems at first glance, there is no alternative. Don't let the amateurish looks deceive you.

Agreed -- Handbrake is great!

 

[PSman1700]

Most YouTubers are idiots (see my disclaimer above).

Exactly, see your video above, as an example of a true idiot. I think geekbench or any bennchmark doesnt really give the full truth for mobile phone performance since throttling is almost always a factor.

Also, you seem to be confusing GeekBench CPU benchmarks and GeekBench GPU benchmarks. The CPU benchmarks don't have issue scaling across architecures and core counts. The GPU benchmarks are rubbish and have long been disproven by Andrei F.

No im not, in mobile device reviews its usually CPU benchmarks being done by most youtubers. And ofcourse the GPU benchmark (even metal) is rubbish, so are other benchmarks where Apple sillicon wins out. Also, these machines should be compared as a whole, cpu vs cpu or gpu vs gpu might not tell the whole picture.

You're a broken record.
No one is buying any Apple product for gaming. The comparisons aren't even theoretically interesting since the 3D API implementation in macOS is a joke.

Yet people keep coming in with gaming benchmarks against gaming products.

So it's weird that you keep posting gaming and Blender benchmarks in this thread.

In reply to the gaming benchmarks being posted, this time with no settings given and a 4k bench left out for the 3090, the resolution most would use with a gpu like that in tomb raider which isnt really that heavy of a game, considering whats out there. Same for baldurs gate and WoW, they are not that intresting.

Arstechnica has done testing too: https://arstechnica.com/gadgets/2022/03/mac-studio-review-a-nearly-perfect-workhorse-mac/2/#h1

 

[MfA]

While I don't disagree with your post above, it's also a fact that raster units are relatively small in area. What would it account as a percentage in a tiny mobile single GPU "core" block? 3 or 5%; peanuts IMHO compared to what texture mapping units can account for.

I thought you meant the old triangle rasterizers would start to bottleneck them. The easiest way to remove the bottleneck is obviously to just stop using FF hardware for it altogether.

 

[MfA]

Agreed -- Handbrake is great!

And for the moment when using it for highest quality, ie. x264/x265, x86 has the advantage. Maybe with M2 they'll improve on NEON.

 

[Wesker]

Exactly, see your video above, as an example of a true idiot. I think geekbench or any bennchmark doesnt really give the full truth for mobile phone performance since throttling is almost always a factor.

No im not, in mobile device reviews its usually CPU benchmarks being done by most youtubers. And ofcourse the GPU benchmark (even metal) is rubbish, so are other benchmarks where Apple sillicon wins out. Also, these machines should be compared as a whole, cpu vs cpu or gpu vs gpu might not tell the whole picture.



Yet people keep coming in with gaming benchmarks against gaming products.



In reply to the gaming benchmarks being posted, this time with no settings given and a 4k bench left out for the 3090, the resolution most would use with a gpu like that in tomb raider which isnt really that heavy of a game, considering whats out there. Same for baldurs gate and WoW, they are not that intresting.

Arstechnica has done testing too: https://arstechnica.com/gadgets/2022/03/mac-studio-review-a-nearly-perfect-workhorse-mac/2/#h1

You're not making any sense. You're the one investing stock in Geekbench, but now you're trying to reverse your position...?

What is your thesis? Now you're trying to post Chinese benchmarks between a 12900K + 2*3080 Ti workstation vs the Studio? What's that, a 900w TDP system (and at a higher cost) vs a system which consumes 160w at peak max? What exactly are you trying to show?