64 core AMD Rome CPU

[Fasauceome]

https://www.google.com/amp/s/www.pcworld.com/article/3319296/components-processors/what-amds-64-core-rome-server-cpu-tells-us-about-ryzen-2.amp.html

 

AMD's new CPU for servers shows off a lot of the new goodies Zen 2 has hinted towards. Some of the important things for the CPU come in the form of 8 channel ram, more floating point performance, PCIe 4.0, etc. What I'm excited about is the core count per "chiplet":

Quote

AMD said it reworked the Zen 2 core to offer ... (double) core density and half the energy use per operation of Zen. Compared to Zen, AMD said to expect twice the performance and four times the floating point performance per socket.

This is very exciting for the consumer PC space, because if this improvement arrives on am4, Intel can kiss their precious market share goodbye. I made a thread about this previously, discussing the 7nm shrink's efficient boost, but I didn't expect it to bring this kind of jump.

Also, the 8 core cluster instead of 4 means Zen 2 won't be constrained by infinity fabric, which means budget ram will be a lot more feasible.

[MyName13]

Well crap, hexa core r3s incoming?

[Fasauceome]

Just now, MyName13 said:

Well crap, hexa core r3s incoming?

A quad core, a 6 core, and then a 6 core with SMT? Interesting idea.

[ATFink]

There's lots of speculation now that IPC will improve 20%. Intel single-thread advantage? Gone.

[Fasauceome]

1 minute ago, ATFink said:

There's lots of speculation now that IPC will improve 20%. Intel single-thread advantage? Gone.

I feel like at this point that 20% is a conservative estimate, given the severe size shrink of the process node. it's also combined with a higher clock speed, so we might see an AMD lead on this one.

[Sauron]

2 hours ago, ATFink said:

There's lots of speculation now that IPC will improve 20%. Intel single-thread advantage? Gone.

In the presentation what they demoed looked more like a 25% boost. And it's no wonder considering the new production node. It's also no wonder Intel's 9k series feels so rushed - they need to take as much advantage as possible of whatever edge they have while it lasts.

[Fasauceome]

2 minutes ago, Sauron said:

In the presentation what they demoed looked more like a 25% boost. And it's no wonder considering the new production node. It's no wonder Intel's 9k series feels so rushed - they need to take as much advantage as possible of whatever edge they have while it lasts.

Here's hoping 10nm keeps things competitive

[paddy-stone]

I'm really looking forward to 7nm zen 2, hoping maybe the refresh on the consumer chips might see a bit of trickle down from this. I will definitely be getting a zen 2 chip at some point, and then passing down my current zen cpu into my server/test system. I don't know what pricing will be like, but going by the last 2 gens it might be feasible to expect similar pricing.

[Sauron]

2 minutes ago, fasauceome said:

Here's hoping 10nm keeps things competitive

If that's even still coming... I wouldn't keep my hopes up. Either way Intel needs a full architectural redesign, they can't compete with AMD by rehashing sandy bridge anymore. With their modular design, AMD can just come back in 18 months with twice as many cores without breaking a sweat. Intel has been struggling to add as few as 6 cores per year to their flagship parts due to the limitations of a monolithic design. Their dual chip 48 core is just a small band aid to try and avoid hemorrhaging customers while they work on a proper answer - it's still painfully inferior to Rome. This thing completely devoured two 28 core Xeons on stage, it's coming for the server market like a freight train and Intel can't stop it right now.

[Fasauceome]

2 minutes ago, Sauron said:

If that's even still coming... I wouldn't keep my hopes up. Either way Intel needs a full architectural redesign, they can't compete with AMD by rehashing sandy bridge anymore. With their modular design, AMD can just come back in 18 months with twice as many cores without breaking a sweat. Intel has been struggling to add as few as 6 cores per year to their flagship parts due to the limitations of a monolithic design. 

To be fair, it's not the core count that Intel struggles with, as you can clearly see the ease of adding new cores. The edge of single threaded performance is really what's in jeopardy here.

 

4 minutes ago, Sauron said:

Their dual chip 48 core is just a small band aid to try and avoid hemorrhaging customers while they work on a proper answer - it's still painfully inferior to Rome. This thing completely devoured two 28 core Xeons on stage, it's coming for the server market like a freight train and Intel can't stop it right now.

Yeah, server is swinging heavily in AMD's favor due partly to the 14nm storage and some aggressive marketing from AMD. 

[Sauron]

4 minutes ago, fasauceome said:

To be fair, it's not the core count that Intel struggles with, as you can clearly see the ease of adding new cores. The edge of single threaded performance is really what's in jeopardy here.

Well no - it's easy for AMD to add cores because they can literally just slap on more 8 core modules. Intel can't do that, they need to incorporate those cores on the same chip.

[Froody129]

On the one hand I want CPU power to increase but on the other I don't so that my 4770 lasts for 5 more years 

[Fasauceome]

5 minutes ago, Sauron said:

Well no - it's easy for AMD to add cores because they can literally just slap on more 8 core modules. Intel can't do that, they need to incorporate those cores on the same chip.

While the modular nature of AMD's design is apparently, Intel has not exactly struggled to make their 9900k. Not to mention the fact that 8 cores / 16 threads is Stillwater excessive for the vast majority of users, so they've still got some breathing room. Let's see if they squander it.

[Fasauceome]

1 minute ago, Froody129 said:

On the one hand I want CPU power to increase but on the other I don't so that my 4770 lasts for 5 more years 

I feel that. I would still be using my 4790k right now if I didn't win the 8086k giveaway, and would be doing so for a few more years to come.

[Syntaxvgm]

26 minutes ago, fasauceome said:

Here's hoping 10nm keeps things competitive

I really don't. Intel failed to produce major improvements for so long when hey had no competition, I think they need a couple years of being whalloped before coming back. They have the money to lose for a a lot longer than AMD did, and they wouldn't lose for too long. 

Really I think another set of the athlon 64 days would be good for intel. 

Not that I think that will happen. 

[Mira Yurizaki]

2 minutes ago, Sauron said:

Well no - it's easy for AMD to add cores because they can literally just slap on more 8 core modules. Intel can't do that, they need to incorporate those cores on the same chip.

Intel can do that (one could argue Intel did do this 13 some years ago). Intel's been toying with the same basic technology for at least a year AMD has employed: https://www.intel.com/content/www/us/en/foundry/emib.html The problem likely lies with someone in Intel who isn't an engineer making "engineering" decisions.

 

Besides that, I don't even think AMD can just slap on another chiplet and call it a day. You need to feed that beast with data. If you don't have the memory infrastructure to provide it, it's just there for workloads that aren't memory sensitive... which would be an issue for the market that AMD presented this announcement to.

[Fasauceome]

2 minutes ago, Syntaxvgm said:

I really don't. Intel failed to produce major improvements for so long when hey had no competition, I think they need a couple years of being whalloped before coming back. They have the money to lose for a a lot longer than AMD did, and they wouldn't lose for too long. 

Really I think another set of the athlon 64 days would be good for intel. 

Not that I think that will happen. 

I mean, the athlon 200GE did just come out, looks like a real classic style come back.

[jagdtigger]

12 minutes ago, Sauron said:

Well no - it's easy for AMD to add cores because they can literally just slap on more 8 core modules. Intel can't do that, they need to incorporate those cores on the same chip.

Not to mention that AMD can still use those modules even if some cores are unstable, they just sell it as a lower core count chip. On the other hand if intel has this issue then the whole chip is wasted....

[Fasauceome]

5 minutes ago, M.Yurizaki said:

Besides that, I don't even think AMD can just slap on another chiplet and call it a day. You need to feed that beast with data. If you don't have the memory infrastructure to provide it, it's just there for workloads that aren't memory sensitive... which would be an issue for the market that AMD presented this announcement to.

Do you think there's a chance AM4 could have triple channel at some point?

[Sauron]

5 minutes ago, M.Yurizaki said:

Besides that, I don't even think AMD can just slap on another chiplet and call it a day. You need to feed that beast with data. If you don't have the memory infrastructure to provide it, it's just there for workloads that aren't memory sensitive... which would be an issue for the market that AMD presented this announcement to.

Of course it's not that simple, it still requires engineering - but it's a hell of a lot easier than trying to fit more and more cores on a single die. Moreover, you're not bound by manufacturing defects - if one of the 8 blocks that are destined to a Rome cpu isn't up to standard, they can put in another one instead of having to downclass the whole cpu.

 

8 minutes ago, fasauceome said:

While the modular nature of AMD's design is apparently, Intel has not exactly struggled to make their 9900k. Not to mention the fact that 8 cores / 16 threads is Stillwater excessive for the vast majority of users, so they've still got some breathing room. Let's see if they squander it.

I'm not talking about the 9900k, I'm talking about the top end xeons. And their struggles are apparent in the price of the 9900k more than in the core count. The "vast majority of users" have no business dropping 500$ on a cpu.

[ATFink]

53 minutes ago, fasauceome said:

Here's hoping 10nm keeps things competitive

No, I was being conservative since I didn't know if the presentation was referring to 12nm Ryzen or 14nm Ryzen. I think the presentation was referring to 14nm Ryzen since that's the node first generation EPYC is on. If that is the case 7nm Ryzen for consumers will have ~20% IPC gain instead of 25% over latest Ryzen processors for desktop.

 

If Intel ever releases 10nm (at this point I'm not sure anything substantial will come, but I'm keeping an open mind) I think it will be very competitive with Ryzen on a performance metric (and completely ignoring cost).

[Mira Yurizaki]

27 minutes ago, fasauceome said:

Do you think there's a chance AM4 could have triple channel at some point?

No, because that would require more pins and AM4 looks pretty full.

[BuckGup]

31 minutes ago, fasauceome said:

Do you think there's a chance AM4 could have triple channel at some point?

No and why would it? The 2700X and 1900X are the same price with the same number of cores at the same clock speed. The 1900X is for workstation use as it has quad channel and 64PCIE lanes while the 2700X is for mainstream and gaming

[Sauron]

7 minutes ago, VegetableStu said:

wasn't there talk about the 9900Ks still not shipping yet? o_o

Yeah... I'm not sure if that's true or not. It will most likely still be available before zen2 consumer chips land.

[ravenshrike]

11 minutes ago, Sauron said:

Yeah... I'm not sure if that's true or not. It will most likely still be available before zen2 consumer chips land.

Of course, the question is what you mean by available. Will they be out in the wild? Sure. Will there be a stable supply of them? That's a much less likely scenario.