5,000 Mgh RISC-V CPU....

[whm1974]

So reading the news on my Phone, I discovered that Micro Magic has designed a RISC-V 64-bit Processor with a clockspeed of 5Ghz as report here:

 

https://www.eenewseurope.com/news/worlds-fastest-64bit-risc-v-core-claims-5ghz-speed

 

And operates at 1.1v... All of this without requiring Liquid Cooling...

[AndreiArgeanu]

It's not that impressive, clock speed doesn't mean anything if IPC is bad. This CPU you mentioned achieves 13 000 in core mark. For reference the ryzen 7 1700X scores 309 792. Definitely an achievement but nothing that's like out of this world or anything.

[whm1974]

Just now, AndreiArgeanu said:

It's not that impressive, clock speed doesn't mean anything if IPC is bad. This CPU you mentioned achieves 13 000 in core mark. For reference the ryzen 7 1700X scores 309 792. Definitely an achievement but nothing that's like out of this world or anything.

Yeah I was wondering about the actual Performance. But still having Clocks that high at a mere 1.1 Volts is nothing to sneer at. However the article doesn't say how many Cores there are or if it uses SMT or not.

[AndreiArgeanu]

1 minute ago, whm1974 said:

Yeah I was wondering about the actual Performance. But still having Clocks that high at a mere 1.1 Volts is nothing to sneer at. However the article doesn't say how many Cores there are or if it uses SMT or not.

Judging the score, it's probably a single or dual core at most.

[Dabombinable]

6 minutes ago, whm1974 said:

Yeah I was wondering about the actual Performance. But still having Clocks that high at a mere 1.1 Volts is nothing to sneer at. However the article doesn't say how many Cores there are or if it uses SMT or not.

One thing is for sure - designers in pretty much all companies have learned what not to do after Intel and Netburst (Intel was shooting for 5GHz).

Edit: Intel wanted 5GHz at under 1V by 2005:
https://www.anandtech.com/show/680/6

[Ankh Tech]

8 minutes ago, Dabombinable said:

One thing is for sure - designers in pretty much all companies have learned what not to do after Intel and Netburst (Intel was shooting for 5GHz).

Edit: Intel wanted 5GHz at under 1V by 2005:
https://www.anandtech.com/show/680/6

you mean 10ghz under 1v. They planned to use 7nm by 2005, and they are still on 14nm++++++++++++++++++++++++++++ till now

[whm1974]

Just now, Dabombinable said:

One thing is for sure - designers in pretty much all companies have learned what not to do after Intel and Netburst (Intel was shooting for 5GHz).

Yeah Netburst along with IA-64 were Complete and Utter Failures. The Pentium4 performed slower and hotter then then the previous Pentium III. Not to mention AMD Athlon CPUs.

 

All the Ititanum did was to help drive the the 64-bit RISC CPUs out of the Server and Workstation Market.

[Mateyyy]

1 minute ago, Ankh Tech said:

you mean 10ghz under 1v. They planned to use 7nm by 2005, and they are still on 14nm++++++++++++++++++++++++++++ till now

[Ankh Tech]

3 minutes ago, Mateyyy said:

Well I have a point don't I

[Haro]

2 minutes ago, Ankh Tech said:

Well I have a point don't I

No, no you don't. 

[mahyar]

impressive

although i never liked running cpus higher 4.3ghz

but for a risc v there might be better stability 

 

[whm1974]

33 minutes ago, mahyar said:

impressive

although i never liked running cpus higher 4.3ghz

but for a risc v there might be better stability 

 

Here is an Article with more detail:

 

https://www.eenewsanalog.com/news/risc-v-core-out-clocks-apple-sifive-available-ip

 

[TVwazhere]

-Moved to CPU's, Motherboards and Memory-

 

This does not meet Tech News Posting Guidelines

[whm1974]

I wonder if there is a direct relationship between the number of Cores/Threads and the limit of increasing Clockspeeds? I'm thinking that the more there are, the harder it is to get faster Clocks.

 

Is this True? Yes I'm aware that there are other things in play here as well. Like how fast the System Memory is and the number of Memory Channels, I/O, and BUS Slots...

[Dabombinable]

4 hours ago, whm1974 said:

I wonder if there is a direct relationship between the number of Cores/Threads and the limit of increasing Clockspeeds? I'm thinking that the more there are, the harder it is to get faster Clocks.

 

Is this True? Yes I'm aware that there are other things in play here as well. Like how fast the System Memory is and the number of Memory Channels, I/O, and BUS Slots...

IPC is what really affects clock speed, its a lot harder to design an high IPC high clock speed CPU, than a low IPC high clockspeed one. With core counts you just need to increase the capacity of the cooler, and the efficiency of a boards VRM (to minimise power wasted as heat).

[whm1974]

3 hours ago, Dabombinable said:

IPC is what really affects clock speed, its a lot harder to design an high IPC high clock speed CPU, than a low IPC high clockspeed one. With core counts you just need to increase the capacity of the cooler, and the efficiency of a boards VRM (to minimise power wasted as heat).

I see. Does increasing Clockspeeds also increase Power Consumption as well?

[Dabombinable]

1 hour ago, whm1974 said:

I see. Does increasing Clockspeeds also increase Power Consumption as well?

When you're talking 1 architecture on 1 node, then normally yes. With different architectures though and node shrinks - not necessarily.

AMD for example managed to have lower power consumption with Ryzen 3 despite despite higher clock speeds and being on the same manufacturing node as Ryzen 2.

[igormp]

5 hours ago, Dabombinable said:

IPC is what really affects clock speed, its a lot harder to design an high IPC high clock speed CPU, than a low IPC high clockspeed one. With core counts you just need to increase the capacity of the cooler, and the efficiency of a boards VRM (to minimise power wasted as heat).

Going even more in depth, usually IPC alone won't tell you the whole figure, since RISC based archs have way higher IPC than CISC ones, with they key difference being that a single CISC instruction can do the same thing as many RISC instructions.

 

Clock speed is more limited to how complex your design is, a "simpler" design on an advanced node with tons of pipelining can reach really high clocks since you will be able to avoid tons of combinatorial circuits and have each needed cycle be really short (thus increasing your clock rate) at the cost of less stuff done at each cycle. Here's a nice, short explanation on such limits when designing CPUs.