MP All Models Is Apple ****ing themselves over the ThreadRipper 3 yet?

[ZombiePhysicist]

I think the review is ultimately spot on. It's made with hindsight of course, and there was no way that apple could have taken the risk when designing the thing back in 2017 to go with AMD. But today, it's clear, the next iteration should have an AMD processor, PCI 4 or 5, etc.

Also, does anyone get the feeling that this machine was supposed to come out in 2018 and something went wrong at apple. Basically from their apology tour till they shipped the machine, it took almost 3 years. Seems way late. Had this machine come out in 2018, all of these arguments would have been moot.

 

[darthaddie]

No one has shown a TR3 system without a water cooled solution so far. AMD recommends WC as well. Maybe I am missing something, but isnt this a deal breaker for workstation class systems shipping commercially?

 

[blackadde]

No one has shown a TR3 system without a water cooled solution so far. AMD recommends WC as well. Maybe I am missing something, but isnt this a deal breaker for workstation class systems shipping commercially?

First of all, there's at least one AMD approved traditional HSF solution -

AMD's Page - https://www.amd.com/en/thermal-solutions-threadripper
Noctua's NH-U14S - https://noctua.at/en/nh-u14s-tr4-sp3

Secondly, why would AIO liquid coolers be a deal breaker for commercial workstation systems? The MTBF for most reputable pumps is something in the ballpark of 50,000 hours.

 

[jinnyman]

Based on what Apple has gone to make MP 7,1 silent and thermally not limited, Apple can surely handle TR3, and definitely EPYC version of TR3 which usually run cooler than TR3.

I keep saying this, but MP 7,1 should have come out in early 2018. 2017~2018 and the similar intel chip in MP 7,1 would be a deal breaker. Due to how Apple prolong each generation of MP 7,1, the current 2019 MP is bit late in the market.

EPYC with TR3 generation chip and TB3 with PCIe 4.0 or 5.0, and of course, with Apple's optimization of AMD chips on Mac OS X, is the monster machine that you all were looking for.

 

[throAU]

Why would there be a shortage of Xeon chips if just about "everybody" wasn't buying them ?

Look up how CPU manufacturing yields work.

TLDR: intel's CPUs need to be massive to fit a competitive number of cores. This reduces yield, increases price.

Because they're on 14nm (still) power consumption and heat is worse than AMD's 7nm at the same core count and clock speed (in the real world, not via intel's repeatedly fudged marketing number TDP rating).
[automerge]1581300309[/automerge]

Development of MacOS Catalina running on AMD APUs, and potentially - CPUs, progresses. Enjoy.

But more than 32 cores on macOS is impossible!
But intel is better for macOS!
AMD doesn't have thunderbolt! Apple would never do that!
[automerge]1581300462[/automerge]

I'm reading 36db on my monitor at head-height while blasting my PC by rendering in Fusion360 (100% CPU) and FurMark (100% GPU) simultaneously. Even if I press up against the fan ducts in the back it's only 42db. I'm not even running anything special - a cheapie 120mm single radiator AIO cooler for my CPU and a bog standard 3-fan layout on an Nvidia GPU. I just set the fan curve to keep things quiet - no different than using smcFanControl in MacOS.

Of all the criticisms to lob at modern PC's, this is the most nonsensical.

BUT ONLY APPLE CAN DO QUIET COOLING


(just don't put a mic next to a macbook pro doing anything of note, we won't talk about that)


My AIO cooled Ryzen is pretty quiet too. The GPU isn't but that's going under water soon as well.
[automerge]1581300611[/automerge]

Also, does anyone get the feeling that this machine was supposed to come out in 2018 and something went wrong at apple. Basically from their apology tour till they shipped the machine, it took almost 3 years. Seems way late. Had this machine come out in 2018, all of these arguments would have been moot.

I'll tell you what went wrong - intel would have promised them CPUs in volume and then failed to deliver. Because their manufacturing process is totally screwed.

The mac pro delay probably isn't Apple's fault per-se. But they should definitely be throwing development at an EPYC (or Theadripper - 256 GB of RAM is likely enough for most mac people) variant ASAP.

 

[throAU]

Based on what Apple has gone to make MP 7,1 silent and thermally not limited, Apple can surely handle TR3, and definitely EPYC version of TR3 which usually run cooler than TR3.

Yup.

If Apple can cool a 28 core desktop xeon, they can cool ThreadRipper.

 

[koyoot]

It isn't propaganda - it is the best review I have seen on the 7,1.

Have you seen any review that has multiple configurations?

Which is the exact reason I posted it here. It even had a little guide - which configuration of Mac Pro is for who: programmers, video editors, photographers(?).

... links from Koyoot. If I wanted AMD propaganda, I'd follow AMD fan sites.

The fact that you wrote this without even watching the video, kinda tells more about you, than actually about me.

 

[askunk]

@koyoot, your tone has been a problem for others in the past. Now I understand why.

What you say is your opinion. That's it. It counts almost as much as mine since we are talking about a technology that is still evolving and software and OSs that have yet to be written.

Furthermore, the point of my post wasn't on ARM. I was saying that while Apple brings their consumer Macs to ARM, they could make the choice of switching to AMD for the high-performance models.

So please: 1) don't reply on things I never said, 2) try to be more polite.

 

[darthaddie]

First of all, there's at least one AMD approved traditional HSF solution -

AMD's Page - https://www.amd.com/en/thermal-solutions-threadripper
Noctua's NH-U14S - https://noctua.at/en/nh-u14s-tr4-sp3

Secondly, why would AIO liquid coolers be a deal breaker for commercial workstation systems? The MTBF for most reputable pumps is something in the ballpark of 50,000 hours.

Oh! I know that WC systems are reliable then ever. My EKWB setup on the 7900x system was running flawlessly for like 2 years.

What I was indeed wondering... I have not seen a WC workstation ship from any of the major brands. In my experience, this is due to the maintenance required on WC systems, which many customers wouldn't want to do.?

Having said that, Yes, Apple could easily cool the TR3.

 

[koyoot]

@koyoot, your tone has been a problem for others in the past. Now I understand why.

What you say is your opinion. That's it. It counts almost as much as mine since we are talking about a technology that is still evolving and software and OSs that have yet to be written.

Furthermore, the point of my post wasn't on ARM. I was saying that while Apple brings their consumer Macs to ARM, they could make the choice of switching to AMD for the high-performance models.

So please: 1) don't reply on things I never said, 2) try to be more polite.

Why do you bring up again topic that was settled two pages ago, and about what I even forgot?

What I am trying to tell is ARM can only work as a low-power solution, that is working alongside other, higher-performance optimized PCs.

ARM never will work as a CPU replacement for computers like Mac Pro, and we all know this. But where it can work is in computers like Chromebooks, or for example "Apple Services access-only machine" .

Both architectures have to work together, and not be replacement of each other, that was the point I was making.

Am I a fan of ARM, or hater?

I was a fan years ago. Now I am genuinely sceptical...

About my tone. Some people are accusing me, or have openly accused me of being AMD fanboy, reading AMD marketing material only, and that appears to be fine for everybody. Why would I have to trim my tone, then, if I am not being rude to anybody?

Linus Thorvalds have once said, that if somebody tells me to be more polite, why I have to change? Maybe its him who has to be more aggressive?

The mac pro delay probably isn't Apple's fault per-se. But they should definitely be throwing development at an EPYC (or Theadripper - 256 GB of RAM is likely enough for most mac people) variant ASAP.

IMO, 256 GB of RAM is Enough for the likes of iMac Pro. Mac Pro should be "limited" to EPYC CPUs for the RAM capacity and capability reasons.

Threadripper is not enough for Mac Pro, but is enough for iMac Pro.

 

[fuchsdh]

Linus Thorvalds have once said, that if somebody tells me to be more polite, why I have to change? Maybe its him who has to be more aggressive?

Quoting giant *******s as a defense rarely helps your case.

"I'm right, therefor have the right to be a dick" is never going to improve a forum thread.

 

[koyoot]

Quoting giant *******s as a defense rarely helps your case.

"I'm right, therefor have the right to be a dick" is never going to improve a forum thread.

Thats because people are taking themselves way too seriously.

IMO, he is right on the matter because friction makes you grow. You either learn to deal with it, or grow thic skin.

 

[mattspace]

What I was indeed wondering... I have not seen a WC workstation ship from any of the major brands. In my experience, this is due to the maintenance required on WC systems, which many customers wouldn't want to do.?

Catastrophic failure of an air-cooling system, in a rack with a bunch of other gear = system shutting down for thermal protction.
Catastrophic failure in a water cooling loop = potentially lots of other gear ruined, electrical fire etc.

Lower risk to chill the air going into the computer, I imagine.

 

[Ulfric]

AMD is getting ready to launch new Epyc skus optimized for higher frequencies, Probaly getting ready to launch Epyc in workstation market

AMD EPYC 7F32 (3.7ghz base, 8C)

AMD EPYC 7F52 (3.5ghz base, 16C)

AMD EPYC 7F72 (3.2ghz base, 24C)

Geekbench 5 results for the 7F72 (dual socket),

Adding ".gb5" to the url you can see what frequency it's running during the test

It appears it's boosting to 3.7ghz all-core.

AMD will probably launch 32 core & 64 core SKUs as well. That will sort out the RDIMM/LRDIMM issues with Threadrippers.

 

[tommy chen]

and apple has been through enough with the water-cooled G5

 

[09872738]

What I am trying to tell is ARM can only work as a low-power solution, that is working alongside other, higher-performance optimized PCs.

ARM never will work as a CPU replacement for computers like Mac Pro, and we all know this. But where it can work is in computers like Chromebooks, or for example "Apple Services access-only machine" .

Why would that be? Some ARMs (on iPad Pro) beat x86_64 on SUSTAINED loads. I cannot see - and never heard of - a valid reason why this might be true. Anyone (@cmaier, perhaps)?

 

[cmaier]

Why would that be? Some ARMs (on iPad Pro) beat x86_64 on SUSTAINED loads. I cannot see - and never heard of - a valid reason why this might be true. Anyone (@cmaier, perhaps)?

It gets old addressing such silly comments. ARM is just another RISC architecture. It's got everything in it that PA-RISC, PowerPC, MIPS, Sparc, and a dozen other architectures have in it. You can make them fast, you can make them slow, you can make the high TDP and you can make them low TDP.

The pre-baked solutions offered by ARM directly have traditionally been focussed on low power consumption, which is why ARM has that reputation. But Apple is free, under license, to do whatever design they want, and they can easily make ARM chips that are quite comparable to x86-64 in performance/power. In fact, given that they don't need microcode sequencers and ROMs, extra decoding pipe stages, etc., and they don't need to support a bunch of 32bit addressing modes that cause horrible complications in x86 designs, in theory ARM (or any RISC) should be able to beat x86 by 10-20% depending on workload.

 

[goMac]

It gets old addressing such silly comments. ARM is just another RISC architecture. It's got everything in it that PA-RISC, PowerPC, MIPS, Sparc, and a dozen other architectures have in it. You can make them fast, you can make them slow, you can make the high TDP and you can make them low TDP.

Most of the comments I see tend to go towards that RISC is unsuitable for higher end CPUs. Which is a pretty unusual and unsound argument.

 

[cmaier]

Most of the comments I see tend to go towards that RISC is unsuitable for higher end CPUs. Which is a pretty unusual and unsound argument.

Yeah, it's not like DEC Alpha, PA-RISC, PowerPC, UltraSparc, RS/6000, etc. were fast or anything

Market forces, not performance ceiling, are why x86 is king of the desktop still.

 

[09872738]

It gets old addressing such silly comments. ARM is just another RISC architecture. It's got everything in it that PA-RISC, PowerPC, MIPS, Sparc, and a dozen other architectures have in it. You can make them fast, you can make them slow, you can make the high TDP and you can make them low TDP.

The pre-baked solutions offered by ARM directly have traditionally been focussed on low power consumption, which is why ARM has that reputation. But Apple is free, under license, to do whatever design they want, and they can easily make ARM chips that are quite comparable to x86-64 in performance/power. In fact, given that they don't need microcode sequencers and ROMs, extra decoding pipe stages, etc., and they don't need to support a bunch of 32bit addressing modes that cause horrible complications in x86 designs, in theory ARM (or any RISC) should be able to beat x86 by 10-20% depending on workload.

Thanks for pointing this out. Again. I know it gets old, but you are just the better man to explain the matter (and busting myths), at least in comparison to myself. Again, thank you

 

[koyoot]

Why would that be? Some ARMs (on iPad Pro) beat x86_64 on SUSTAINED loads. I cannot see - and never heard of - a valid reason why this might be true. Anyone (@cmaier, perhaps)?

What sustained load was it? Video Editing? When the Video Transcoding was done on dedicated ASIC on the SoC?

How would the comparison between actual CORES performance look?

 

[Quu]

and apple has been through enough with the water-cooled G5

The XEON Apple is using (W-3275M) has a stated TDP of 205 W. The Threadrippers are 280 Watt. It's really not that much more. The cooling solution Apple is already providing could handle 350-400 Watt I'd say.

But also Intel's TDP's are based on their maximum base clocks while AMD's are based on the maximum boost clocks. So you can't compare Intels 205 to AMD's 280 directly. It's likely they're around the same (10-15 Watt difference) in the real-world.

As others have said in the thread though these two chips aren't really pitted against each other in the market. The XEON's with 1.5TB of RAM addressable are more comparable to AMD EPYC's which have 225 Watt TDP's and address 2TB of RAM.

The Threadripper chips with 256GB RAM ceilings do make them a bit of a niche of a niche because if you need 32-64 cores (which Threadripper now offers with the 3990X) you likely also need more than 256GB of RAM unless you're doing something very little people do even within these niche fields already suited to a Mac Pro spec'd computer.

I think AMD mostly just released the 3990X to have a halo product for enthusiasts. And I say that as someone that really likes what they're doing generally. It was about time Intel had some competition. But yeah for me 256GB of RAM on a 64 Core CPU is just not enough. Should have been 768GB-1.5TB somewhere between those I think. Maybe 256GB for the 32-Core, 512GB on the 48 Core and 1TB on the 64 Core but alas they didn't do that.

The higher clocked EPYC's we're expecting them to launch into the workstation category will likely retain the high PCIe lane counts and RAM address capability though, but this is just guessing as we don't have those products yet and the Mac Pro is here today already.

 

[goMac]

Yeah, it's not like DEC Alpha, PA-RISC, PowerPC, UltraSparc, RS/6000, etc. were fast or anything

Market forces, not performance ceiling, are why x86 is king of the desktop still.

It's generally the same chatter RISC naysayers had at the time that RISC requires a bunch of instructions where CISC requires one. Which completely discards instruction cost, pipelining, etc etc...

All of this completely ignores that one of the big gains Intel made was in a transition to an internal RISC core. But that just adds more bloat to Intel's CPU by having to translate all the incoming CISC instructions to a set of internal RISC instructions.

 

[cmaier]

It's generally the same chatter RISC naysayers had at the time that RISC requires a bunch of instructions where CISC requires one. Which completely discards instruction cost, pipelining, etc etc...

”one” instruction that takes N execution clock cycles, vs. N instructions that take 1 clock cycle each.

 

[shaunp]

Threadripper is highly overrated for the compute heavy applications I do in three major areas:

1. AMD's vector core just caught up with Haswell. Zen 1 fakes AVX 256 by doing it in 2 128 bit cycles. Intel has been able to do it in a single cycle since Haswell. New Intel high-end CPUs have dedicated AVX512, so Intel is ahead in vector elements per cycle by 2x.

2. Intel's MKL are highly hand tuned assembly math routines. They're used in nearly every computation program. AMD tried to copy with a drop-in compatible ACML, then quit.

3. Intel has 6 memory channels on Xeon W, Threadripper only has 4. It makes a major difference in bandwidth-limited computation. Plus, Xeon is one die with a 2-D mesh network, not a higher-latency MCP.

If you're serious about memory and IO anyway, you get a Xeon SP dual socket system, not the compromised Xeon W/Threadripper.

In my opinion, Threadripper is a bunch of stuff they threw together to get their core market, gamers and kids at home, excited. That's why people aren't making serious clusters with AMD and why I wouldn't buy an AMD Mac.

Agree with some of what you say, but you should read between the lines. Threadripper is being compared to Xeon by customers who were at the lower end of the Mac Pro customer base because of the price/performance benefit. At the high-end which is where you are looking you should be comparing it to Epyc. There are perhaps some applications where Xeon is still better but times are changing.