AMD Remains Competitive: Zen 3 Announced

[Fasauceome]

Ever since Intel promised a neat 18% IPC improvement for their 10 nanometer CPUs, there was probably some worry that it would take AMD a while to catch up in single threaded once more. However, while many thought that Zen 2 + was going to be the next architecture (following in the steps of Zen and Zen +) AMD has confirmed that a new generational microarchitecture is on the way:

https://wccftech.com/amd-zen-3-new-cpu-architecture-significant-ipc-gains-higher-clocks/amp/

Quote

AMD's Senior Vice President, Forrest Norrod, has unveiled some major details of their next-gen Zen 3 CPU architecture in an interview with The Street

There was no exact performance uplift cited, as that is par for the course for something so far from release. I don't expect the huge IPC improvements that made Zen 2 so incredibly competitive against Intel, but I expect it to be meaningful enough that it will remain the best performance for the dollar (unless Intel decides to change their pricing structure drastically.)

 

These chipswill be manufactured on the 7nm+ node from tsmc which means that they'll have a lot of improved density and whatnot.

Quote

Now according to TSMC themselves, the 7nm+ process node allows an increase of 20% in overall transistor density while increasing power efficiency by 10%. AMD could take full advantage of the process node and we could see a density boost of up to 20% in the Zen 3 architecture while having 10% better power efficiency.

And that's what I love most about Ryzen, the power efficiency.

 

And let's not forget about epyc

Quote

Some of the details we know for EPYC Milan CPUs include:

• 7nm+ Zen 3 cores (~64 core / 128 thread)
• Pin Compatible With SP3 Socket
• 120W-225W TDP SKUs
• PCIe 4.0 Support
• DDR4 Memory Support
• Launch in 2020

 

But let's move on past the stuff we expect, what's new?

Quote

When asked about what kind of performance gain Milan's CPU core microarchitecture, which is known as Zen 3, will deliver relative to the Zen 2 microarchitecture that Rome relies on in terms of instructions processed per CPU clock cycle (IPC), Norrod observed that -- unlike Zen 2, which was more of an evolution of the Zen microarchitecture that powers first-gen Epyc CPUs -- Zen 3 will be based on a completely new architecture.

A very important distinction is being made here. It's being implied that the 15% IPC improvement from Zen 2 was kind of chump change, and that maybe the switch to Zen 3, which is going to be fundamentally different, is going to offer so much more in terms of a performance leap. this doesn't surprise me, as the seven nanometer node probably has a ton more potential yet to come.

Quote

AMD also confirmed that they will be following Intel's Tick-Tock cadence which Intel seems to have completely abandoned

Just had to take a brief moment to shit on Intel real quick, lel.

 

My hope:

They adopt "Zen 5" and skip Zen 4 if they go to a new socket and do Ryzen 5000, that way the leading generation number matches up with the microarchitecture identifier and everything becomes easier.

[Otto_iii]

 

5 minutes ago, Fasauceome said:

My hope:

They adopt "Zen 5" and skip Zen 4 if they go to a new socket and do Ryzen 5000, that way the leading generation number matches up with the microarchitecture identifier and everything becomes easier.

I second this, 4s a crappy number anyway.  4000 is fine, but 4, nah son

[Flying Sausages]

[Fasauceome]

On 11/19/2019 at 9:05 AM, Otto_iii said:

 

I second this, 4s a crappy number anyway.  4000 is fine, but 4, nah son

 

I hope you get the reference

[That Franc]

I'm liking the news today. I feel like for my workload, a 65w 8-core Zen 2 processor is plenty powerful, so I won't be waiting for Zen 3, but it's still good to see them push the industry forward quicker than everyone could expect.

[Flying Sausages]

1 minute ago, That Franc said:

I'm liking the news today. I feel like for my workload, a 65w 8-core Zen 2 processor is plenty powerful, so I won't be waiting for Zen 3, but it's still good to see them push the industry forward quicker than everyone could expect.

Wait till next year or two and get yourself a 3900x for ~$400 usd. 

[That Franc]

5 minutes ago, OlympicAssEater said:

Wait till next year or two and get yourself a 3900x for ~$400 usd. 

I need a 65 watt chip, and I don't need that many cores for what I do, so 3700X is more than enough for me. (I'd rather get a better graphics card because Resolve really loves CUDA cores, apparently)

[Tristerin]

Zen+++++++++++++++

[Zando_]

Will be interesting to see if they can push higher clocks. AFAIK Zen 2's IPC is actually a good bit better than Intel's, Intel chips stay on a level or slightly higher in many things due to a higher clock ceiling, but Ryzens give them an extremely hard fight and take home a few wins at 300-500Mhz lower clocks. If they had the clock headroom Intel CPUs do, there wouldn't really be any competition performance wise for most tasks. 

[WereCat]

15 minutes ago, That Franc said:

I need a 65 watt chip, and I don't need that many cores for what I do, so 3700X is more than enough for me. (I'd rather get a better graphics card because Resolve really loves CUDA cores, apparently)

You can run 3900X/3950X at 65W TDP by either turning ECO Mode ON in BIOS or Ryzen Master

[Fasauceome]

5 minutes ago, Zando Bob said:

Will be interesting to see if they can push higher clocks. AFAIK Zen 2's IPC is actually a good bit better than Intel's, Intel chips stay on a level or slightly higher in many things due to a higher clock ceiling, but Ryzens give them an extremely hard fight and take home a few wins at 300-500Mhz lower clocks. If they had the clock headroom Intel CPUs do, there wouldn't really be any competition performance wise for most tasks. 

I hope one of the main points they work to address is the steep voltage curve. If AMD overclocked like Intel they wouldn't stand a chance.

[That Franc]

13 minutes ago, WereCat said:

You can run 3900X/3950X at 65W TDP by either turning ECO Mode ON in BIOS or Ryzen Master

Then I'm pretty sure I'll get performance similar to that of an existing 65w chip because... Well, physics? Besides, getting a powerful chip to then continue to nerf it just doesn't sound like an intelligent move to me.

There is target audience for those processors, I'm just not a part of it.

[Mira Yurizaki]

Quote

AMD also confirmed that they will be following Intel's Tick-Tock cadence which Intel seems to have completely abandoned

So expect 5nm chips in the generation afterwards then, is what they're saying.

 

2 minutes ago, That Franc said:

Then I'm pretty sure I'll get performance similar to that of an existing 65w chip because... Well, physics? Besides, getting a powerful chip to then continue to nerf it just doesn't sound like an intelligent move to me.

I'm not sure of the behavior of Ryzen 3000 chips, but with Ryzen 2000, the X series has an aggressive voltage curve. I lowered the voltage of my Ryzen 2700X and capped the maximum speed to 4GHz from 4.3GHz. While that sounds silly at first glance, the 4.3GHz maximum boost was only possible on a handful of cores and when running a sustained workload test, undervolting and capping the speed allowed the processor to sustain it for over half an hour. At stock settings, the processor was chugging down near base clocks within 20-25 minutes.

[Taf the Ghost]

Nothing much was really shared that wasn't kind of already known. (It should be noted AMD has had working silicon back for like 4 or 5 months at this point, so they know what Zen3 will do.) Epyc is selling really well and will continue to do so because of fairly straight forward TCO reasons.

 

One thing we know, sort of by accident, is that Zen3's core complex will move to a unified design rather than 2 CCXs. That'll open the entire L3 to each chip, which should help IPC on its own and be really useful in certain server applications. 

[soldier_ph]

#Hyped

[williamcll]

It does make me wonder should I by a zen 2 now or wait even further from my 2700X

[WereCat]

5 hours ago, That Franc said:

Then I'm pretty sure I'll get performance similar to that of an existing 65w chip because... Well, physics? Besides, getting a powerful chip to then continue to nerf it just doesn't sound like an intelligent move to me.

There is target audience for those processors, I'm just not a part of it.

No.

Increasing voltage to achieve higher clock speeds increases the power consumption exponentially not linearly so if you lower the TDP by 50% you won't lose 50% of performance.

 

Besides I don't understand your logic there. 

You're basically saying that for some reason a stand alone 65W TDP CPU should perform better than a 105W TDP CPU limited to 65W TDP which is a complete opposite in this case for one particular reason. 

The 3900X/3950X are the best bins in the lineup which means that they not only achieve the best performance but they are also more efficient while doing that. 

You can clearly see that in many reviews the 3950X which is the highest bin matches the 3900X in total power consumption while outperforming it quite significantly. 

Lowering TDP on chips like these will allow you to keep the majority of its performance while heavily cutting down on a total power consumption. 

[DrMacintosh]

Cool, I'm just chilling with my 2600 though, and probably will be for a long time. 

[Taf the Ghost]

39 minutes ago, DrMacintosh said:

Cool, I'm just chilling with my 2600 though, and probably will be for a long time. 

A 4600 in 2021 should be a really good deal & upgrade.

[That Franc]

3 hours ago, WereCat said:

Besides I don't understand your logic there.

There's also an economic argument to be made. Why get a more powerful processor when a less powerful one is already more than enough? I just don't see why I should be spending significantly more money on all that power when it won't make my life substantially better - there is no way I can benefit significantly from this increase.

[Flying Sausages]

18 minutes ago, That Franc said:

There's also an economic argument to be made. Why get a more powerful processor when a less powerful one is already more than enough? I just don't see why I should be spending significantly more money on all that power when it won't make my life substantially better - there is no way I can benefit significantly from this increase.

If you can find 3900x in a year or two at $380 USD on second hands then you will not going to snap it? 3700x to 3900x is a huge performance jump in productivity.

[Drak3]

1 hour ago, OlympicAssEater said:

3700x to 3900x is a huge performance jump in productivity.

That is HEAVILY dependant on usecase.

 

For someone like me (and most people here), the 3900X is irrelevant. Unless you're working with large data sets (like video rendering or large scale number crunching) or multiple instances of smaller data sets (like high traffic file/web/game servers or VM hosting), those extra cores just sit around, doing nothing.

[Otto_iii]

13 hours ago, Fasauceome said:

I hope you get the reference

pity the man who doesnt

[Ratisbona]

15 hours ago, That Franc said:

Then I'm pretty sure I'll get performance similar to that of an existing 65w chip because... Well, physics? Besides, getting a powerful chip to then continue to nerf it just doesn't sound like an intelligent move to me.

There is target audience for those processors, I'm just not a part of it.

Not even close. I did that experiment a few weeks ago on my 3900X. I downclocked it to about 3700-3800MHz fixed and undervolted it to 0.975V or something like that, to make it draw as close as I could get it to 65 W. And it still got over 2800 points in Cinebench.
 

And from what I read about the 3950X, these chips are very much better binned that the 3900x again, so they should perform even better.

[maartendc]

Wait, so are we getting an iteration of Zen 2 like Zen 2+ first?

 

Or are they going straight to a new architecture with Zen 3?

 

Is Zen 3 supposed to be released in 2020?