How much untapped potential does Zen have?

[_mockingbird]

How much untapped potential does Zen have?

With Zen being a completely new architecture, I like to think that Zen is in the same stage as Nehalem was and that the next big revision of the architecture would bring a big performance bump similar to Sandy Bridge.

Intel seems to be stuck on Skylake. Kaby Lake is Skylake rebadged. Coffee Lake is basically a price-cut of Skylake: more cores for about the same price.

Obviously, with Intel's architecture being much more mature than AMD's, there simply isn't much room for Intel to easily squeeze out more performance.

 

[CaptNumbNutz]

I will wait until the foundry's get their act together on the lower nodes. Right now, Ryzen has ZERO extra potential since it seems to be artificially capped at 4.0-4.2ghz range. It's an architecture that holds promise, if it's manufactured correctly.

If they had foundry's as good as Intel, I'm sure they could hit the 4.8-5.0 ghz range when overclocked, much like Intel's K chips. Then of course they need to keep funded to continue improvements.

The hopeful side of me believe's their HPC/server side plus the low end Zen APU will bring them back into profitability due to volume shipments from OEM's and Zen 2 will be where the real catch up game begins. My Sandy Bridge chip is still good, I'm going to wait till next year for more news.

 

[crazycrave]

Your going to see more AMD CPU/GPU vs Intel/Nvidia as AMD has a new chipset driver that allows Ryzen to become more (ONE) paired with Radeon GPU's , I know my 290x loves my Ryzen 1400 as the image quality and just how smooth it plays on WoT 30 slot sever with HD Max 1080p settings @ stock speed is mind blowing for a 65watt cpu.

It reminds me of the x64 days when AMD killed ass with lower clock speeds , the more IPC Zen makes the less Mhz it will need to do the same workload and i am not so sure there done yet with the firmware/drivers for Zen/Radeon.

 

[Andy735]

I will wait until the foundry's get their act together on the lower nodes. Right now, Ryzen has ZERO extra potential since it seems to be artificially capped at 4.0-4.2ghz range. It's an architecture that holds promise, if it's manufactured correctly.

If they had foundry's as good as Intel, I'm sure they could hit the 4.8-5.0 ghz range when overclocked, much like Intel's K chips. Then of course they need to keep funded to continue improvements.

The hopeful side of me believe's their HPC/server side plus the low end Zen APU will bring them back into profitability due to volume shipments from OEM's and Zen 2 will be where the real catch up game begins. My Sandy Bridge chip is still good, I'm going to wait till next year for more news.

Has the process improved? Earlier silicon lottery numbers had only about 1/5th of 1700 series Ryzen chips being able to hit 4.0 GHz. I am cautiously optimistic about Zen+(or 2 or whatever they are calling it) but I am unsure if much gains will be made at this stage. Ryzen's memory speed dependence really sucks considering the cost of ram right now.

 

[Windreaper]

Has the process improved? Earlier silicon lottery numbers had only about 1/5th of 1700 series Ryzen chips being able to hit 4.0 GHz. I am cautiously optimistic about Zen+(or 2 or whatever they are calling it) but I am unsure if much gains will be made at this stage. Ryzen's memory speed dependence really sucks considering the cost of ram right now.

While the process remains the same, as far as I know, it would subjectively appear that the yields have improved to some extent (improving quality down the line). I've read some reports of RMA replacements needing less voltage (while some reported no improvement), my RMA chip was also slightly better.

Intel holds a large process advantage, and don't see that changing any time soon. Intel 14nm is more comparable in logic density to other fabs' 10nm, while Intel 10nm will most likely be close to what other fabs call 7nm. The numbers basically tell nothing about the quality of the process, gate length, etc. You can thank marketing departments for the confusion. wikichip.org is a good resource, if you're interested.

Memory speed dependence is related to the workload - gaming performance benefits from fastery memory speeds while workloads less sensitive to interconnect latency see next to no benefit. Memory compatibility is much improved compared to launch, too, so there's no need to buy expensive RAM.

 

[N4CR]

Has the process improved? Earlier silicon lottery numbers had only about 1/5th of 1700 series Ryzen chips being able to hit 4.0 GHz. I am cautiously optimistic about Zen+(or 2 or whatever they are calling it) but I am unsure if much gains will be made at this stage. Ryzen's memory speed dependence really sucks considering the cost of ram right now.

Yes, yields already have increased with the current process as Windreaper pointed out.
Almost all RMAs use lower volts and clock higher. Threadripper is on average another 200MHz higher already too.

 

[Pieter3dnow]

How much untapped potential does Zen have?
With Zen being a completely new architecture,

I think no one knows maybe they just have better pictures

 

[bobzdar]

Not a whole lot - the untapped potential is more on the software side than the hardware side tbh. I think Zen3 will be the next major bump, Zen2 will be clock bumps and maybe a few minor improvements to edge cases, which is easy to extrapolate, but I think they'll have some more focus on architecture improvements for Zen3. That will be the more interesting refresh, unless they can get a really big clock bump for Zen2. I wouldn't rule that out, though, the 14nmLPP process has a wall at around 3.8ghz above which power and heat goes way up, if they can move that more than the predicted 200-300mhz, it could be a big bump, especially as Intel loses it's clock speed advantage as cores increase.

 

[DuronBurgerMan]

Early 2018 will see clock speed bump. Smarter people than I could speculate as to how much. If I were guessing, though... maybe +400MHz or something. Is what I've heard around the grapevine. But total guess.

Early 2019 will see Zen2 with IPC bump (and clock speed bump too). Heard 5% IPC bump kicked around, no idea if true. With a little more clockspeed boost, I suspect will see close to equal clocks with Intel. Still a bit behind in IPC.

Zen 3... no idea. Their roadmap says 2020, but seems too far out to really speculate on release date or specifics yet.

 

[Dan_D]

We almost never know how much potential a given architecture has until the dust settles on it's replacement. We never knew how much potential the Pentium Pro had at the time it was released. Similarly, we won't know how much legroom Zen has until it's been revised a few times or AMD has to abandon it in favor of something else.

 

[N4CR]

Not a whole lot - the untapped potential is more on the software side than the hardware side tbh. I think Zen3 will be the next major bump, Zen2 will be clock bumps and maybe a few minor improvements to edge cases, which is easy to extrapolate, but I think they'll have some more focus on architecture improvements for Zen3. That will be the more interesting refresh, unless they can get a really big clock bump for Zen2. I wouldn't rule that out, though, the 14nmLPP process has a wall at around 3.8ghz above which power and heat goes way up, if they can move that more than the predicted 200-300mhz, it could be a big bump, especially as Intel loses it's clock speed advantage as cores increase.

Remember current Zen is on a LP process, hence why it is more efficient than prior Intel offerings and has a pretty hard clock limit around 4GHz. 2nd Critical is 3.6GHz and is your 'worst case'. It's basically optimised for datacentre/mobile use.
Zen2 is looking to be a 12 core design. Reason is they have mentioned Epyc 48 core... So if they get close to 5GHz under OC, that's where I pull the trigger on the 1700/1800 equivalent, because any gaming/single thread advantage Intel has is practically nullified.
AVX512 is Intels' only out and they'll push that like they pushed various SSE versions last time AMD got competitive. I hope it doesn't go anywhere and keeps the playing field even, so they can't do the usual proprietary lockout/tick tock re-spin SSE bullshit.

 

[Pieter3dnow]

Remember current Zen is on a LP process, hence why it is more efficient than prior Intel offerings and has a pretty hard clock limit around 4GHz. 2nd Critical is 3.6GHz and is your 'worst case'. It's basically optimised for datacentre/mobile use.
Zen2 is looking to be a 12 core design. Reason is they have mentioned Epyc 48 core... So if they get close to 5GHz under OC, that's where I pull the trigger on the 1700/1800 equivalent, because any gaming/single thread advantage Intel has is practically nullified.
AVX512 is Intels' only out and they'll push that like they pushed various SSE versions last time AMD got competitive. I hope it doesn't go anywhere and keeps the playing field even, so they can't do the usual proprietary lockout/tick tock re-spin SSE bullshit.

The factor of scaling in the next Ryzen cpu is what is appealing rather then architecture differences. AMD can have the fastest options in their cpu but if it requires a much different code path and a fresh compile it tends to go nowhere (FMA4 anyone?).
For the next release it would be a miracle if AMD gets above 15% clock for clock. But I would expect a clock for clock performance below 10% but I would expect scaling to be higher will it reach 5 ghz overclocked , I don't think so (would be a surprise).

 

[N4CR]

The factor of scaling in the next Ryzen cpu is what is appealing rather then architecture differences. AMD can have the fastest options in their cpu but if it requires a much different code path and a fresh compile it tends to go nowhere (FMA4 anyone?).
For the next release it would be a miracle if AMD gets above 15% clock for clock. But I would expect a clock for clock performance below 10% but I would expect scaling to be higher will it reach 5 ghz overclocked , I don't think so (would be a surprise).

Best TRs are at 4.2 already and they are the top 5%. Move to higher power process, few improvements here and there and 5GHz wouldn't be out of question for a decent full loop. I'd expect 4.6-4.8 for average setup and cooling.
I'm not sure on IPC gains. I wouldn't expect too much, Zen is not bad already, smaller R&D team etc, perhaps 5% max? Something good would be if they can fix the memory/HT/IF interlinking requiring cheaper ram.

Edit to add: I called current Ryzen OC capabilities out, along with TR/Epyc predictions as to TDP/clocks and was on the mark. With the above I am referring to Zen2.
Zen+ I would assume 10-15% max, so perhaps 4.2-4.5Ghz.