AMD Ryzen 5000 Series / Zen 3 Launch

[atomsymbol]

The reason why Intel backported a 10-nm design to 14-nm, creating Rocket Lake, was precisely to reach higher clock frequency, because the limit for their current SuperFin process is 4.8 GHz.

It is certain that Rocket Lake will reach at least 5.0 GHz (seen in engineering samples). It is not known if it will reach as high frequencies as the Skylake derivatives, i.e. at least 5.3 GHz, but we do not know yet any reason that could prevent this.

Nevertheless even if Rocket Lake would succeed to take back the title of being the fastest CPU in single thread applications, that would be just by a few percent, while the multi-threaded performance will be very low compared to Zen 3 or even Zen 2, so it could be recommended only for hardcore gamers, who use their computer for nothing else but games.

Above, I was not clear enough regarding what atomsymbol said "14nm Rocket Lake (Willow Cove arch) will most likely have lower boost frequency than 14nm Comet Lake (Skylake arch) because the former is a more complex design than the latter".

Being a more complex design means doing more per clock cycle, thus consuming more power at the same frequency. This leads to a lower clock frequency in multi-threaded applications, when the power consumed by the CPU is limited by cooling.

In single-threaded applications however, the active core consumes much less than the TDP, e.g. about 50 W, and the clock frequency is limited by the single-core maximum turbo frequency.

So it does not matter that a Rocket Lake core is more complex. That will limit the base frequency to a lower value compared to a Comet Lake with the same number of cores, but that does not limit the maximum turbo frequency, so there is no reason why Rocket Lake could not reach 5.3 GHZ or even more.

Of course, there might be some details of the microarchitecture that could still limit the maximum frequency to a lower value than for Comet Lake, but we do not know yet anything about that.

Moreover, such a lower limit for the clock frequency due to logical design is very unlikely, because the only motivation for the Rocket Lake project was to reach higher frequencies, to have a chance to compete with Zen 3. Intel would never have spent so much money for this project without hoping to reach very high frequencies. It remains to be seen whether they will succeed.

Maximum CPU frequency (boost frequency) depends on two factors:

• The length of the critical path of a pipeline stage. In a simplified view: max frequency = the speed of light divided by the length of the critical path; the speed of light is a constant.
• An increase in frequency requires an increase in current and/or voltage for the CPU to operate correctly at the higher frequency. This generates additional heat that does not contribute to computation and has to be moved away from the CPU die before it will result in a hardware error.

It is likely that the x86 CPU with the shortest critical path ever designed was Pentium 4 (NetBurst architecture, Prescott core) with 31 pipeline stages. This was able to reach 3.2 [email protected] and 3.73 [email protected] using conventional (metal+air+fan, no water/nitrogen) cooling.

• More pipeline stages enable a shorter critical path (because the CPU's work is being subdivided into smaller units)
  • Excessive focus on shortening the critical path results in a lower IPC (instructions per clock) of the CPU (Pentium 4)

It is likely that Pentium 4 Prescott would be able to easily reach over 6 GHz on conventional cooling if manufactured using 14nm technology.

Skylake and Sunny Cove have the same number of pipeline stages: 14-19.

The length of the critical path depends on the number of transistors used to perform a particular function. Sunny Cove increases the size of the L1D cache to 48 KiB (Skylake: 32 KiB). Sunny Cove L1D latency increases to 5 cycles (Skylake: 4-6 cycles).

It is possible, although unlikely, that Sunny Cove with a much larger CPU core than Skylake has a critical path of the same length as Skylake (because of more advanced (better tuned) algorithms in Sunny Cove).

https://en.wikichip.org/wiki/intel/m...kylake_(client)
https://en.wikichip.org/wiki/intel/m...res/sunny_cove
https://software.intel.com/content/w...ce-manual.html

[AdrianBc]

The reason why Intel backported a 10-nm design to 14-nm, creating Rocket Lake, was precisely to reach higher clock frequency, because the limit for their current SuperFin process is 4.8 GHz.

It is certain that Rocket Lake will reach at least 5.0 GHz (seen in engineering samples). It is not known if it will reach as high frequencies as the Skylake derivatives, i.e. at least 5.3 GHz, but we do not know yet any reason that could prevent this.


Nevertheless even if Rocket Lake would succeed to take back the title of being the fastest CPU in single thread applications, that would be just by a few percent, while the multi-threaded performance will be very low compared to Zen 3 or even Zen 2, so it could be recommended only for hardcore gamers, who use their computer for nothing else but games.

Above, I was not clear enough regarding what atomsymbol said "14nm Rocket Lake (Willow Cove arch) will most likely have lower boost frequency than 14nm Comet Lake (Skylake arch) because the former is a more complex design than the latter".


Being a more complex design means doing more per clock cycle, thus consuming more power at the same frequency. This leads to a lower clock frequency in multi-threaded applications, when the power consumed by the CPU is limited by cooling.

In single-threaded applications however, the active core consumes much less than the TDP, e.g. about 50 W, and the clock frequency is limited by the single-core maximum turbo frequency.

So it does not matter that a Rocket Lake core is more complex. That will limit the base frequency to a lower value compared to a Comet Lake with the same number of cores, but that does not limit the maximum turbo frequency, so there is no reason why Rocket Lake could not reach 5.3 GHZ or even more.

Of course, there might be some details of the microarchitecture that could still limit the maximum frequency to a lower value than for Comet Lake, but we do not know yet anything about that.

Moreover, such a lower limit for the clock frequency due to logical design is very unlikely, because the only motivation for the Rocket Lake project was to reach higher frequencies, to have a chance to compete with Zen 3. Intel would never have spent so much money for this project without hoping to reach very high frequencies. It remains to be seen whether they will succeed.

[AdrianBc]

10nm i7-1185G7 has a maximum clock of 4.8 GHz when running a single thread. It is therefore unlikely for 14nm Rocket Lake to be able to reach 4.8 GHz in single-threaded mode. Notebook CPUs (28W TDP) have approximately the same single-core boost frequency as desktop CPUs (about 100W TDP, but only when multiple cores are utilized). 14nm Rocket Lake (Willow Cove arch) will most likely have lower boost frequency than 14nm Comet Lake (Skylake arch) because the former is a more complex design than the latter.

The reason why Intel backported a 10-nm design to 14-nm, creating Rocket Lake, was precisely to reach higher clock frequency, because the limit for their current SuperFin process is 4.8 GHz.

It is certain that Rocket Lake will reach at least 5.0 GHz (seen in engineering samples). It is not known if it will reach as high frequencies as the Skylake derivatives, i.e. at least 5.3 GHz, but we do not know yet any reason that could prevent this.


Nevertheless even if Rocket Lake would succeed to take back the title of being the fastest CPU in single thread applications, that would be just by a few percent, while the multi-threaded performance will be very low compared to Zen 3 or even Zen 2, so it could be recommended only for hardcore gamers, who use their computer for nothing else but games.

[smitty3268]

We don't want products which are slower that we already own nor buying the same performance for the same price.

Agreed. But even in the worst case here, you are getting 20%+ performance from the 5600X vs the 3600XT, for a 20% price increase. So it's pretty much flat. And the more expensive chips are better, because you're getting that 20%+ performance boost with a 10% (or whatever) price increase.

I think that's why it seems a little strange to see people going crazy over these price increases. They're still better value than the last gen.

[ms178]

I think if you want a cheaper option AMD will still keep selling Zen 2 chips for a while, and those are a pretty decent option too.

I agree with that part under the assumption that they get cheaper in the coming months, we are still not at Zen 1, Zen + pricing levels for these either.

The other part is a question of perspective: People don't seem to adjust for the time which has passed and only compare to past or current Intel product pricing but we expect progress over time. We don't want products which are slower that we already own nor buying the same performance for the same price. We expect to get more overall value over time. Price-to-Performance is in my view the most important metric to measure this, with current data points. Of course the pricing and product strategy of Intel during the past 10 years was absurd in terms of milking its customers - they had a monopoly in several markets. But that doesn't make AMD any better if they mimick that behavior - we saw what happened when only financial goals dominate their thinking.

[wizard69]

Whatever works for you. Personally, until the last couple of years I wouldn't have picked up an AMD CPU for use cases that require good single core performance.

One last thing people complaining about a $50 bump in pricing are completely nuts. First off I just returned from the grocery and spent $56 for two half filled paper bags. The reality is $50 means little these days. More importantly here is that the price is introductory, eventually the price will adjust to meet demand. So if you don't want to pay that extra $50 wait until initial demand dies and sales or specials, happen.

Beyond that this processor looks to be better than AMD implied in their reveal. I just find it odd that they seemingly ignored the technical or knowledgeable user and basically glossed over what these Zen 3 chips are.

[smitty3268]

I also don't get that people are applauding AMD to become the new Intel. They have earned their place through hard work, sure, but 20% more IPC for a lot more in price is not progress in my eyes (just look at the insane 299 $ for the 5600X).

I don't think anyone loves the higher prices, but comparing them to Intel is where people object. 20% is a pretty massive gain for a CPU. Intel has been pushing out 5% gains for years, and they were also charging hundreds of dollars more than AMD is even after this $50 price hike.

For example, the "insane" 5600X is still $60 cheaper than the i7-8700k launched at 3 years ago and provides massive performance increases. The higher-end CPUs are just more and more extreme, as Intel used to sell 8 core cpus for over $1000.

So yes, they are starting to turn in that direction, but in terms of degree they are nowhere close.

I think if you want a cheaper option AMD will still keep selling Zen 2 chips for a while, and those are a pretty decent option too.

[pmorph]

The fact is, there is no such thing as "you wouldn't even consider purchasing them regardless of the price", and claiming there is brings your credibility into question. It's all about price/performance ratio. Very few consumers actually buy the $1000 halo product, they buy something in the middle. Many more buy whatever costs the least. When a $259 AMD chip performs on par with a $329 intel chip, the AMD chip is a better buy, regardless of who wins the absolute performance crown at the top end of the market.

Whatever works for you. Personally, until the last couple of years I wouldn't have picked up an AMD CPU for use cases that require good single core performance.

[wizard69]

People drink more than $50 on their bar tab, in a single night. Either everyone here is completely broke or lead very sad lives.

This "or lead very sad lives" is probably the truth.

[wizard69]

For those of us into throughput, I'm concerned about memory bandwidth...

Well that will not change dramatically until AM5 supporting new RAM standards. You would be right and if you have a decent system right now I'd wait for DDR5 supporting AM5 socket to be released which I can imagine happening early in 2021. A faster RAM solution is really needed to move APU performance forward. That is likely why no APU's released this go around.