no... they're (both/multiple ISPs) still run by clowns.
The fact that the ISP can't deal with 100s of millions of devices on the market (i.e., the real world user base) is the joke here.
Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
MP All Models Mac Pro X,Y…WHAT IF…
- Thread starter ZombiePhysicist
- Start date
- Sort by reaction score
The fact that the ISP can't deal with 100s of millions of devices on the market (i.e., the real world user base) is the joke here.
No, you're wrong. The iPhone is a *client* of the server. The Server sets the procedure and consent conditions, not the client.
The UX problem is that the manual entry option is only available after the automated option has failed. That it trips up the server's defence procedures is a side effect.
Last edited:
A Mac Pro with dual 8490H Xeons and four GPUs and 4.0TB RAM?
I couldn’t imagine it. And Apple would surely charge USD$150K for it.
The ISP is the vendor, the iPhone users are customers that number potentially in the hundreds of millions.
If your customers can't use your product, and fixing it is trivial, you need to fix your product 🤷♂️
Fixing it isn't trivial, that's the point. Cpanel systems have anti-DDOS protections for a reason. The iPhone's beaviour in interrogating a mail server is indistinguishable from an attack, because mail.app is made by clowns.
Being the customer is irrelevant, customers are expected to follow the rules of the establishment, which is why restaurants have bathrooms.
🔮
The NV 5090 isn't really "highest end'. There GPUs that have both higher onboard memory bandwith thresholds and to/from GPU 'card' throughput. They just don't usually come in 1980's vintage form factor standards. Additionally, if use benchmarking metrics that largely consist of software that assumed the PCI-e link is hobbled and slow, then not likely to be able to measure the new improvements all that well. It work 'brute force' most legacy games into running any faster. Software which more substantively used a direct GPU to SDD data transfer mechanism likely would ( if put PCI-e v5 on both side of a primarily direct link... no switches.)
In the Win-World some of the slots are directly hooked to the CPU package PCI-e lanes. In the current Mac Pro system, all the PCI-e slots are provisioned through a switch. Upgrading the backhaul from the CPU package to the switch would all more bandwidth (even if capped at PCI-e v4) to trickle out to the slots. Not necessarily about more bandwidth to a single slot but more aggregate bandwidth.
If ignore MP 2019 slot 1 and 3 the MP 2023 was not a backslide on total bandwidth. Lots of grumbling from the sidelines in these forums didn't want to ignore that 1 and 3 bandwidth and tagged the . The Ultra set-up was a bit asymettrical with x8 and x16 provisioned into the switch. x8 v5 is equivalent bandwidth to x16 v4. If went to two x8 v5 that would be a net gain ( and back to Apple OCD symmetry ) .
That said, I wouldn't hold my breath on PCI-e v5. In the rest of the Mac line up, there isn't much pressure to make the jump. Apple liked using PCI-e v4 in the M series in general to reduce the line outs for PCI-e soaking up edge space on the other dies. v5 might be using to cut down the lanes ( as illustrated above) without hobbling the bandwidth.
That is different objective than Windows workstation world.
Thunderbolt 5 ports and the bandwidth they use might be a good reason for Apple to upgrade to PCI-e5.
- M4 Pro Mac mini
- M4 Pro Macbook Pro
- M4 Max Macbook Pro
PCIe v4 is a feeder to TBv5... It does not 'run on' PCIe v4.
Not only that, but it also runs out of completely different PCI-e root hub than the stuff being provisioned out to the slots. The TB controllers have their own PCIe 'root'. So not tightly coupled at all in implementation. ( other than save effort by not having to do the work to implement the new standard; reusing designs already have for different instances ).
The backhaul internal network on the SoC die is somewhat a zero sum game. The more bandwidth that TBv5 (and new CPU , new GPU , new NPU , faster LPDDRx , etc) want to consume the less 'extra' bandwidth there is going to be for the the PCI-e root that is provisioning out the external backhaul to the slot switch.
The asymmetric mode of TBv5 means the GPUs have to shovel more data to the display controllers (and out the TBv5 ) port than TBv3. That is more than just 'doubling' the traffic (i.e., more competition internal network. ).
The internal network technically isn't completely a zero sum game. They can try cranking up the internal die network to supply even more bandwidth, but that will put more stress on the UltraFusion connector between dies (if there is one). Making UltraFusion faster/better is likely going to be a higher priority that hot rodding a single PCI-e slot with PCI-e v5.
Two x8 PCI-e v5 provisioning would put less bandwidth pressure on the internal network than two x16 PCIe v5 will. Two x8 PCI-e v5 is likely better perf/watt number for the whole die also ( fewer I/O off the die , lower wattage). If Apple puts in the 'extra' work , decent chance more motivated by that , than being "competition" driven by TBv5 on Mac Studio ports.
The other , but likely slower paced, driver would be wanted to get their won SSD modules into a better competitive stance with x4 (and up) PCI-e v5 SSDs performance levels. There is lower competitive pressure though on the primary SSD. ( have to buy Apple's). At some point will move , but again the rest of the line up isn't feeling huge competitive 'heat' there ( 'fast enough' for huge bulk of target market).
[ There is usually some hand waving at a 'real AI data center chip' but fairly good chance that won't be generic MacOS or a chip that will end up in a any "Mac" or on the retail market.
multiple 400GBe ports on a server ... yeah they'd need to up the bandwidth game. Better chance Apple would slap the security enclave and some custom drivers on a 'fork' of something that Broadcomm was already doing. ]
So while I have decreed the death of the Mac Pro...
I did spot one interesting tidbit in Apple's marketing materials when you scroll down on this page a bit:
PCI expansion perhaps leaves some sliver of hope a new mac pro with actual PCI might be forth coming? Am I hoping against hope at this point?
I did spot one interesting tidbit in Apple's marketing materials when you scroll down on this page a bit:
Mac Studio
The ultimate pro desktop. Powered by M4 Max and M3 Ultra for all-out performance and extensive connectivity. Built for Apple Intelligence.
www.apple.com
PCI expansion perhaps leaves some sliver of hope a new mac pro with actual PCI might be forth coming? Am I hoping against hope at this point?
So while I have decreed the death of the Mac Pro...
I did spot one interesting tidbit in Apple's marketing materials when you scroll down on this page a bit:
Mac Studio
The ultimate pro desktop. Powered by M4 Max and M3 Ultra for all-out performance and extensive connectivity. Built for Apple Intelligence.www.apple.com
View attachment 2488792
PCI expansion perhaps leaves some sliver of hope a new mac pro with actual PCI might be forth coming? Am I hoping against hope at this point?
That isn't technically correct. Some marketing folks are confusing Tunderbolt with PCI-e . They are not the same thing. There is no asymmetric PCI-e , so that "up to" is just handwaving. 80Gb/s symmetric is still better than 40Gb/s (TBv4) .
The only thing getting 120Gb/s on TBv5 is outbound video streams. Not PCI-e.
So while I have decreed the death of the Mac Pro...
I did spot one interesting tidbit in Apple's marketing materials when you scroll down on this page a bit:
Mac Studio
The ultimate pro desktop. Powered by M4 Max and M3 Ultra for all-out performance and extensive connectivity. Built for Apple Intelligence.www.apple.com
View attachment 2488792
PCI expansion perhaps leaves some sliver of hope a new mac pro with actual PCI might be forth coming? Am I hoping against hope at this point?
Unless your ideal creative playground involves more displays than your soldered, shared memory GPU is allowed to sustain.
Funny how the amount of VRAM you had in a GPU was what determined how many displays you could drive, then when VRAM became effectively as much memory as your actual ram, you can't drive any more displays than a conveniently market segmenting quantity based on how far up the range you buy.
2023 Mac Pro already came with actual PCI Express slots. The problem is a lack of add-in cards that could meaningfully make use of them. In particular, people's interest in GPUs.
The upcoming Mac Pro should come with higher total bandwidth of the PCIe slots. I think the iGPU performance will be on par if not surpass that of highest achievable in 2019 Mac Pro since Apple ends its OS upgrade next year.
Also, Apple is a member of Ultra Accelerator Link consortium. So don't be surprised if UALink comes in a future Mac Pro. I believe whether that will realise or not depends on the synergy with their internal-use-only AI server hardware.