Return to

The POWER and PowerPC General Discussion / News Thread



For those interested in the Talos II you can check out some initial benchmarks from Phoronix below


Sounds like Raptor Computing Systems is willing to give Level1 a Talos II mainboard to review!

In response to my question/tweet:

I would love to see Level1Techs do a mainboard review for the #TalosII. Any chance RaptorCompSys would be willing to hook them up with one?

We have this:

I referred them the Level1 contact us page, does that sound about right?

In general I’m just plain excited at the prospect of @wendell playing with a Talos II machine, and thought I would share the news of this possibility.

BTW, shouldn’t this thread be tagged as powerpc?


Why does it cost so much?


It’s relatively low volume, server/workstation hardware, with PCIe 4.0.

It’s also the company’s first mainboard as far as I can tell, although it looks like they have been working with Coreboot for a while.


The Talos 2? It was originally designed to take P7, Tyan stopped making P7 boards and recycled all the ones they had so that killed Talos 1, Tyan made Power 8 boards so they called it Talos 2, then Power 9 came out and they upgraded from one to the other.

Power 8 and 9, even 4 core, are rather high in price because its mostly HPC that buys them. Theres no real desktop use for them even though they would wreck any i7 that tried to compete.

Though I still drool at the 2 core 16 thread P7 chips from time to time…


Raptor originally had plans to make a POWER7 board? I’ve only heard about the Talos I crowdfunding attempt, which would have been POWER8 if successful.

From what I’ve read, (as I have no personal experience buying enterprise/workstation CPUs) the POWER9 chips are actually cheaper than their x86 counterparts, but the mainboard is more expensive.


From what I’ve read, yup. Thats the company’s hardware history. And since they aim at the consumer /[ college markets they’re way at the bottom of the totem pole as far as Tyan and other manufac’s are concerned.

If P9 really is all that much cheaper than the X86 parts I might actually look into putting a POWER machine together for myself. Dream computer right there.


The pricing thing was a comparison with server-grade stuff like Xeon/EPYC, consumer CPUs will obviously be cheaper. With some quick napkin math, the 8 and 18 core chips are the most reasonable price points at the moment:

Cores Price Per-core
4 375 USD 93.75 USD
8 595 USD 74.36 USD
18 1425 USD 79.17 USD
22 2625 USD 119.32 USD

Keep in mind that POWER9 is SMT4, so I don’t know what Xeon/EPYC chips (SMT2) would be equivalent to these.

Adi Gangidi was comparing STREAM benchmarks with POWER9 vs an EPYC 7601 and a Skylake Xeon 8176; but those were tests for memory bandwidth, so I don’t know if those are really the best processors for comparison.

I should note that the Barreleye G2 uses LaGrange modules, rather than the Sforzas in the Talos II. LaGrange has more memory channels:


Interesting. If I get a talos I’ll be sure to pump some benchmarks out. Seems theres no public info anywa?ys.


I would love a Power9 machine. I have several PPC Macs and miss using them. A 867Mhz PowerMac G4 was my first personal computer and it was great for years. Till OSX Snow Leopard and support started to fade away,


Well then this is the thread for you!


Got to love the PPC. There is a PowerMac G5 dual 2.7, my 867 PowerMac G4, 2 1.25Ghz iMac G4s with 2GB RAM and SSDs and a PowerBook G4 12” 1.5Ghz with a SSD on and under my desk.


You should use them! I use my ibook almost every day.

Look up leopard rebirth.


Sadly I’ve gotten to the point where a C2D is the slowest I can handle for day to day tasks. I have a C2D MacBook Pro and Mac Mini that are used daily but the G4 and G5 are just to slow for me anymore.


Well the fact that you think the G4’s are slower than the C2D’s is funny to me. They’re actually rather fast chips, OSX is just a pile of garbage and has a lot of stuff in it that doesn’t need to be there. Look into leopard rebirth and monolingual and let me know what happens.


They are fast on OSX Tiger and earlier but slow past that. Tiger and later were build for Intel machines.


Yes so get monolingual and strip out the intel and G3 code. Tada, leopard runs faster than tiger stock.



What kind of witchcraft did I just install on my G5? It’s fast and youtube works flawlessly on T4F with several tabs open.


Have you done any Linux benchmarks to compare G4/G5 to C2D without changes in OSX?

In general, you would also have to watch out for any architecture-specific optimizations; for example it can make a substantial difference if the code is properly using an architecture’s vector functionality. Luc Trudeau (trudluc) has some rather impressive charts showing gains from tuning VP9 for POWER9 chips.

How does removing language files and executable code improve the speed of OSX?
Both should never be even loaded into RAM unless you use them, and even then, your only benefit would be RAM and therefore potentially less use of swap space. The actual computation speed should not be affected by such things.


So the basic problem with OSX 10.5.8 is that its MEANT to be Intel only. But, at the time of release there was still enough demand to push the PPC version out. So 10.5 was split in two and we got Snow Leopard pretty soon after. They had planned the Intel drop sometime in 04, but there was still PPC demand and better G5 / 970 chips were coming out so they wanted to see where it’d go.

I like Leopard a lot, but it goes through a list of available executeables when you boot and tries to find what chip you have and what executable to run. For the G4 specifically, theres 4 major architectural differences, 3 which apply to the mac, and 2 that apply to powerbooks. But, because its G4 its going to try to run those all anyways till one works. Or at least thats what I can see in Activity Monitor. So with most 7447 and 7450e chips, 7400 code isn’t going to run, at least not well. As is the same with 745/750 (G3) code. It’ll run, just not well. And sometimes I have seen that the app will ignore the whole architecture check and just run whatever runs, resulting in garbage performance.

So, stripping all that shit out actually DOES help, even though it seems like it wouldn’t. I normally have it run every few months on my machines if I have done anything with them on a major scale.

Think of it as like the problem with smartphones. The differences are so stupid and so ingrained that its easier to just do a new architecture. The G4 is actually based on the 603, as is the G3. As well, the G4 is the WORST chip. An 800MHz G3 can run apps and a system BETTER than a G4 both because there isn’t as much shit piled into the processor, and because the cache is faster.

Really powerpc macs are a fucking mess and I can only tell people to use G5’s because its a whole clean slate.

Honestly I could have a whole panel at defcon about how much of a pile of shit the G4 is and I’d still use my ibook every day lol.