GlobalFoundries abandoning 7nm, 5nm and 3nm

I’ll be honest, 1st gen Zen is going to look a bit amateurish in comparison.
And that’s just process improvements, I didn’t even get to the Zen 2 core architecture improvements.

1 Like

Damn. Are we talking latency improvements?

1 Like

I’ve seen a P4 do 8Ghz before. :laughing:
And that wasn’t even a suicide run.

4 Likes

I’m talking normal workstation performance. A 9590 can run at 5.2GHZ just fine under efficient cooling and run stable. I want to see another 9590 in the next few years.

The pipeline on the FX chips is entirely different though, it was longer and had simpler stages thus making high clocks much easier to attain.

Same goes for the Pentium 4 netburst architecture, intel built that to clock high, but it had a high branch mis-prediction cost as a penality among other things.

It was also a much bigger node with different gate design that could tolerate much higher leakage among various other side effects which allowed higher clocks without loosing stability too much.

That said you will see something much much better than a 9590.
Zen 2 will make a 9590 look like your granpa’s Celeron 233,

3 Likes

All this discussion is a painful reminder that I don’t know very much about processor design.

3 Likes

Well it is a series of tubes

Gives a good intro about AMD’s Pentium 4 moment.
That pipeline was built with the same logic that got intel the fast clocked P4 which could be beat by a much lower clocked Athlon 64.

@FaunCB

The longer pipeline allowed the Bulldozer family of processors to achieve a much higher clock frequency compared to its K10 predecessors. While this increased frequencies and throughput, the longer pipeline also increased latencies and increased branch misprediction penalties.

Piledriver improved little of the critical performance limiters of the Bulldozer architecture, it just pushed clocks even higher to compensate, increased the TDP and added workarounds that did little to address the core problem.

4 Likes

Inb4 a 5-6 ghz 3200X running a router just because.

If I’m correct under 7nm, with silicon only, you begin to meet quantum effects such as tunneling. And you don’t want to happen because an electron could pass even if the transistor is open. So you need to dope the silicon to change its property but even then at some point you will need to change of material like by using graphene to achieve smaller node.

1 Like

Something everyone on this thread should probably read at least once.

2 Likes

This is probably something worth creating it’s own thread for:

2 Likes

Wasn’t there something like a diamond bsed proc a while bck? At least in rumor?

so you think they are staying with 14nm silicon or actually working with those ‘others’, and produce some different kind of chips - maybe not based on silicon anymore?

They’re staying at 12/14nm, along with their other processes.

They do what they do for amd, but with smaller organizations or runs.

5GHz is a nothing burger for a processor. You dont measure a CPU on max Hz or max cores. You test workloads you will use it for.

No one wants a wife’s shopping car to do the quarter mile in 1.4 secs and not have room to bring home beer.

No one wants the wifes semi to not fit at home so parked 1km away either.

Its all about the base, no terrible. Wait I mean real work loads. If it aint better dont buy it.

1 Like

I honestly don’t expect that to happen in the next decade, if ever. You can already live justwith SSDs, a 256/512GB isn’t really that expensive and should be enough for an operating system and a few applications, but if you need real storage it’s long way foward.

Price per GB is still about 6x cheaper on hard drives than solid state. If you consider that even tape is still around it’s hard to see anything displacing magnetic storage in the foreseeable future.

I’m not actively working on ditching HDDs, but several generations of flash have existed that allowed you to cram 10-20TB of flash chips into a 3.5in form factor.

And I’m guessing it’s probably easier to spool up more flash manufacturing than HDD manufacturing. Thus making it more available and cheaper.

I’m just hoping it happens.

1 Like

Sorry wrong area

posted this in the wrong place

I wonder what happens to the arrangements with IBM Research; they have been doing work on 7, 5, and 3 nm:

https://finance.yahoo.com/news/ibm-synopsys-accelerate-3nm-process-130500008.html

I would assume that especially for DOD or other US Government stuff, CPU manufacturing has to take place in the US, so what happens with GlobalFoundries canceling 7nm? The other notable partner in that alliance is Samsung, but I assume all their fabs are in Korea.

Then again, the 14LPP used for Zen/AMD is different from the 14HP process used for POWER9/Z14/IBM. Is it possible this news only affects the AMD processes, and not those planned for use by IBM for (presumably) POWER10 and Z15? Or was there never going to be an LPP/HP split for 7nm?