Intel's X299 Platform to Counter AMD's X399 with 12-core CPUs

[Caring1]

There's nothing to "design". Intel's HEDT processors are binned Xeon CPUs that are gimped to only work with their corresponding chipset (X99, X299 etc.). Intel could make new models over night if they wanted to, but there's no reason for them to do so.

I agree with you, based on what I have previously read about XEON chips being used and partially disabled for consumer use.

 

[efikkan]

How naive and clueless can people be?
The new Xeon "E5s" and "E7s" would have arrived regardless of Ryzen, and obviously with the new Xeon "E5" Intel would also released Skylake-X regardless of Ryzen, since they would not have made another chip for consumers. These releases has nothing to do with Ryzen, and that's a fact. End of discussion.

 

[TheLostSwede]

No-one is arguing that Intel wouldn't release new products regardless of Ryzen or not. The schedule has changed because of Ryzen and Intel has decided to add new SKUs as they feel they need something more impressive to counter AMD. I don't know why you're going all crazy over this.

The 12 core is most likely a response to Ryzen though, as it was not on any roadmaps until just now. And Xeons and HEDT are the same chips, with slightly different chipsets. Intel has the ability to quite easily disable features of its chips to make them fit their product plan, this is why we have i7, i5, 3, Pentium and Celeron chips that are all cut from the same die.

 

[efikkan]

The schedule has changed because of Ryzen and Intel has decided to add new SKUs as they feel they need something more impressive to counter AMD. I don't know why you're going all crazy over this.

No, the schedule has not changed.
There is no evidence of new SKUs, since SKUs are usually not public until the release.

The 12 core is most likely a response to Ryzen though, as it was not on any roadmaps until just now.

The 10 core for Broadwell-E was not on early roadmaps either, was that a response to Ryzen, Bulldozer or something? This is just silly.

this is why we have i7, i5, 3, Pentium and Celeron chips that are all cut from the same die.

That's not even remotely true.
The Xeon E3, E5, and E7 lineups are usually made from 2-3 different dies in each segment. Exact core count and features varies from generation to generation. In total Intel has 6+ dies for desktop, some of which are also used for consumer products.

 

[TheLostSwede]

Yes Sir, you know best...

 

[Vego]

i am happy with my 10 cores

 

[notb]

If you compare Ryzen 1800X to 6900K in single and multi-threaded CB15, you'll notice that the 1800X gains distance to the 6900K in multi-threaded:

A better comparison may actually be with the 1700X instead of the 1800X because it comes from behind in single to be ahead in multi.

CineBench results depend on more factors than just the CPU. I've seen charts showing 6900K with more than 1540 points.

Obviously we haven't seen Naples CB15 as of yet but, depending on how it scales, it's quite possible AMD will be ahead of Intel in this.

Well... it simply must be. High-core-count Xeons are based on Broadwell and will soon be replaced by Skylake-SP.

You are ofc correct: the latency between CCXs will be a key factor and we have yet to see how 4 dies in a single package affect latencies. This will "make or break" the success of AMD's server line of CPUs, IMO.

It could be even worse. I didn't have time to read a lot about Naples, but AFAIK CPUs are going to share RAM via the Infinity Fabric. In the early leaks this was even mentioned as a way to achieve the 8 memory channels (sort of). If that's true, it could be really slow...
But we'll see. I don't believe it'll quickly become a popular server platform in actual business solutions, as moving to a totally different arch is very difficult and expensive. And of course today the primary limitation in servers is in storage, not in processing power. But I'm sure we will see some interesting high-profile implementations (the Top500 list).

 

[matar]

Sweet 12 cores...

 

[Solaris17]

That's just it: they should be @ least wary of the server class versions of AMD's Zen based processors because it appears AMD's multi-thread implementation is working better then Intel's: @ least according to CineBench 15. If so, and barring any unforseen last minute problems, AMD will pose a very serious threat: cheaper (supposedly) and faster processors (in multi-threaded) ...

lol I dont think so, CPUs arent the entire ball game in server land. AMD needs to get decent server platforms boards before Intel has anything to worry about.

Not to mention No one that runs servers takes scores from HEDT or consumer chips seriously.

Sorry when people are doing deep learning or or heavy virtualization im looking at CPU instruction sets, cache size, architecture clock cycle performance and thermal output. Not how fast some kids water cooled skylake CPU is rendering balls in cinebench.

Servers =/= Workstations and the comparisons really shouldn't be such.

 

[HTC]

lol I dont think so, CPUs arent the entire ball game in server land. AMD needs to get decent server platforms boards before Intel has anything to worry about.

Not to mention No one that runs servers takes scores from HEDT or consumer chips seriously.

Sorry when people are doing deep learning or or heavy virtualization im looking at CPU instruction sets, cache size, architecture clock cycle performance and thermal output. Not how fast some kids water cooled skylake CPU is rendering balls in cinebench.

Servers =/= Workstations and the comparisons really shouldn't be such.

What i was trying to get across was that Zen appears to be better then 6900K @ multi-threaded, as evidenced by the 1700X in the pics i showed, even though it's true that the IPC still favors Intel's 6900K, but that's in the "desktop arena".

 

[Solaris17]

What i was trying to get across was that Zen appears to be better then 6900K @ multi-threaded, as evidenced by the 1700X in the pics i showed, even though it's true that the IPC still favors Intel's 6900K, but that's in the "desktop arena".

I can respect that, It's just important to understand and not singling you out jsut a great example of what the unseen bias appears to be in the forums. That while it may be possible to directly translate consumer or HEDT with workstations for blender and rendering, neither tof these systems builds or parts can be compared to servers.

We don't value the same things. The chips are also different.

 

[HTC]

I can respect that, It's just important to understand and not singling you out jsut a great example of what the unseen bias appears to be in the forums. That while it may be possible to directly translate consumer or HEDT with workstations for blender and rendering, neither tof these systems builds or parts can be compared to servers.

We don't value the same things. The chips are also different.

True: CPU is very important but it's still "only" one variable and there are other variables @ play, such as CCX / memory communication, motherboards (as you have mentioned in the earlier reply).

 

[notb]

True: CPU is very important but it's still "only" one variable and there are other variables @ play, such as CCX / memory communication, motherboards (as you have mentioned in the earlier reply).

Still not that. It's simply that there are other qualities of CPUs beside performance.
@Solaris17 - you care to explain?

 

[Ithanul]

lol I dont think so, CPUs arent the entire ball game in server land. AMD needs to get decent server platforms boards before Intel has anything to worry about.

Not to mention No one that runs servers takes scores from HEDT or consumer chips seriously.

Sorry when people are doing deep learning or or heavy virtualization im looking at CPU instruction sets, cache size, architecture clock cycle performance and thermal output. Not how fast some kids water cooled skylake CPU is rendering balls in cinebench.

Servers =/= Workstations and the comparisons really shouldn't be such.

Agree.

The few servers I worked with are aimed at good stable virtualization and be able to handle a drop in military base infrastructure (in a sandy environment at that). A poor 5960X or 1800X would scream bloody murder with that load.

 

[sergionography]

What i was trying to get across was that Zen appears to be better then 6900K @ multi-threaded, as evidenced by the 1700X in the pics i showed, even though it's true that the IPC still favors Intel's 6900K, but that's in the "desktop arena".

Yes and you are correct. What people seem to be missing is that ryzen does not need to beat skylake in ipc in order to win. Ryzen is broadwell level of ipc but with better thermals/better efficiency, which i suspect is the reason multithread picks up in some benchmarks (ryzen sustaining higher consistent clocks).

Few notes to remember on why ryzen is disruptive that i feel reviews completely missed:

1. Ryzen at this stage is on its first revision which is why it reaches a wall around 4.2ghz or so, but with that in mind its quite impressive.

2. Below that 4ghz mark Ryzen cores scale very well with voltage and run excellent with what i perceived as superior power efficiency than intel broadwell and perhaps even skylake(that remains to be seen)

3. Ryzen offers 8 cores(with broadwell ipc) that scale up to 3.7ghz with 65w tdp, something that i looked for on intels website out of curiosity but could find no match for.

4. Amd cores have a smaller footprint than intel and their chips are manufactured in third party fabs making them cheaper (although intel has finer control over the process but with greater operational cost)

6. Desktop is only the tip of the iceburg with that 4ghz wall being meaninless in mobile and server which are both more profitable and seem like a perfrct fit for ryzen in its current state.

7. With future revisions/iterations of ryzen I would expect much greater clockspeed along with a healthy ipc increase as again ryzen right now is a first iteration making it a worst case scenario with plenty room for modification and ironing out.

8. I skipped point number 5 but you haven't noticed

 

[notb]

2. Below that 4ghz mark Ryzen cores scale very well with voltage and run excellent with what i perceived as superior power efficiency than intel broadwell and perhaps even skylake(that remains to be seen)
7. With future revisions/iterations of ryzen I would expect much greater clockspeed along with a healthy ipc increase as again ryzen right now is a first iteration making it a worst case scenario with plenty room for modification and ironing out.

Basically all CPUs scale well with voltage under their stock speed (or stock boost speed).
What is different about Ryzen is how steep the curve becomes in OC. This is why there is such a noticeable "wall".
It's not like Ryzen lacks power or something, but AMD will have to work hard on the manufacturing process to clock the next generation higher.
It's difficult for your "7" to be called "expectations", as these are usually made on a basis of some analysis or experience. For now they are merely hopes.

3. Ryzen offers 8 cores(with broadwell ipc) that scale up to 3.7ghz with 65w tdp, something that i looked for on intels website out of curiosity but could find no match for.

AMD's TDP is much different to Intel's. Ryzen is in fact very power efficient, but just don't get to attached to those 65W, because it'll draw quite a bit more.

4. Amd cores have a smaller footprint than intel and their chips are manufactured in third party fabs making them cheaper (although intel has finer control over the process but with greater operational cost)

AMD Zen cores are smaller than Intel's because of inferior node. Intel core is a bit better and a bit faster, but also a bit larger. Well... there's always some cost.

6. Desktop is only the tip of the iceburg with that 4ghz wall being meaninless in mobile and server which are both more profitable and seem like a perfrct fit for ryzen in its current state.

Ryzen is not a server CPU (that would be Naples). Mobile Zen platform is still a mystery.
As far as mobile solutions go - Ryzen doesn't have an IGP, so there will always have to be an additional chip included (unless it's the APU you're talking about). Either way, there goes a lot of the price/performance difference.
I'm sure Zen will work well in servers, but I'm just not sure if better than the competition...
As far as HPC - we don't know. In previous I've seen arguments that this arch does not favour HPC (chiefly because of slightly poorer instruction implementation). We'll see how that goes.

 

[sergionography]

Basically all CPUs scale well with voltage under their stock speed (or stock boost speed).
What is different about Ryzen is how steep the curve becomes in OC. This is why there is such a noticeable "wall".
It's not like Ryzen lacks power or something, but AMD will have to work hard on the manufacturing process to clock the next generation higher.
It's difficult for your "7" to be called "expectations", as these are usually made on a basis of some analysis or experience. For now they are merely hopes.

AMD's TDP is much different to Intel's. Ryzen is in fact very power efficient, but just don't get to attached to those 65W, because it'll draw quite a bit more.

AMD Zen cores are smaller than Intel's because of inferior node. Intel core is a bit better and a bit faster, but also a bit larger. Well... there's always some cost.

Ryzen is not a server CPU (that would be Naples). Mobile Zen platform is still a mystery.
As far as mobile solutions go - Ryzen doesn't have an IGP, so there will always have to be an additional chip included (unless it's the APU you're talking about). Either way, there goes a lot of the price/performance difference.
I'm sure Zen will work well in servers, but I'm just not sure if better than the competition...
As far as HPC - we don't know. In previous I've seen arguments that this arch does not favour HPC (chiefly because of slightly poorer instruction implementation). We'll see how that goes.

I kept using the term Ryzen instead of Zen so I apologize for the confusion. So to reiterate, Ryzen is the first take on zen cores, and what i was stating is that zen cores seem to be more efficient that broadwell even though the ipc is similar, but with lower high clock scaling due to zen being a bit more biased towards efficiency/mobile. Remember each design has limitations, so when you design a cpu core you have to choose either high clockspeed or high efficiency at lower clocks. Think of steamroller and excavator cores; do you remember how excavator couldnt reach high clockspeeds compared to steamroller and likewise steamroller compared to piledriver. Why? Because amd designed the cores with super high density allowing much lesser power draw but the result was lower clock headroom. they did so because the core design needed to be scalable across all power budgets which in this case mobile was first. Things arent any different with zen, its a scalable core designed for all power budgets. With server prioritizing multithread and with mobile running on low power budgets its clear where the priority lies for zen1. For that reason I do believe process improvements will give maybe up to 10% clock headroom increase somewhat like with polaris but id love to be wrong in this.

AMD Zen cores are smaller than Intel's because of inferior node

Actually that is incorrect, an inferior node is actually the exact oposite as the transisters and pitch and what not are all larger, however Amd designed zen with certain benefits as well as tradeoffs just like any engineering project.
AMD does use an inferior node but they almost always achieve greater density than intel due to using a gpu like methodology which ends up with zen being smaller and with greater efficiency on lower voltages with the trade off being scaling in higher voltages

 

[R0H1T]

Anything launching this summer was taped out long before the launch of Ryzen.

And you're assuming Intel didn't know of Ryzen from a couple of years back? Intel/Nvidia know more about AMD than any xyz last minute leaker, of course the final performance is usually a closely guarded secret but the fact that Intel is preponing the launch of SKL-X &/or CFL shows that they're into desperation territory, also KBL-X wth is that abomination!

 

[sergionography]

And you're assuming Intel didn't know of Ryzen from a couple of years back? Intel/Nvidia know more about AMD than any xyz last minute leaker, of course the final performance is usually a closely guarded secret but the fact that Intel is preponing the launch of SKL-X &/or CFL shows that they're into desperation territory, also KBL-X wth is that abomination!

they had many of these chips planned for early 2018, now they are being pushed towards second half 2017. also intel has always been able to offer 12cores to the hedt market since the intel extreme chips are practically xeon chips with disabled cores, so the fact intel is upping their core count this gen is basically a response to ryzen and AMD being competitive again.