Ryzen, Skylake, and everything that's coming next. (MEGA discussion thread)

[Arachnotronic]

For point number 2. Keep in mind that those will run at much lower speeds than what Intel has to offer in the i9 line. The 18 core they offer is 2.2GHz with max Turbo (not sure if this is all core or single core) of 3GHz. That isn't going to fly their Extreme platform.

How much power does an overclocked 6950X draw, especially when exotic cooling like LN2 is used?

These boards are designed with very robust power delivery systems to handle the extreme wattage -- that's why these HEDT boards are expensive compared to the lower-end mainstream boards.

 

[T1beriu]

The 18 core they offer is 2.2GHz with max Turbo (not sure if this is all core or single core) of 3GHz.

Xeon Gold 6150 (Skylake) - 18C (36T) - 2.7 Ghz/3.7Ghz (base/turbo) in 165W TDP. So it can be done.

If you'll still say that 2.7 Ghz it's not going to fly for their Extreme platform, I'm going to say 2 things:

1. The higher core count CPU always had a low clocked base compared to lower core parts

2. I don't think Intel is crazy enough to show up with a equivalent of Xeon Gold 6154 - 18C (36T) - 3.0 GHz base - 200W TDP - for $1999 that would probably go for a $5000-6000 price bracket.

 

[Topweasel]

How much power does an overclocked 6950X draw, especially when exotic cooling like LN2 is used?

These boards are designed with very robust power delivery systems to handle the extreme wattage -- that's why these HEDT boards are expensive compared to the lower-end mainstream boards.

Well over 200W some nearly capping out at around 300w. Problem is overhead. Just because a vast majority of boards can accept more, as AMD learned with the 9K FX, doesn't mean they will. Intel has to tell these guys what the power envelope of the system is. But they probably haven't done enough testing and tweaking to figure out what a competitively clocked HCC die is going to require. The 18c at anywhere near 3GHz, is probably 250w or higher. Unless Intel has already established that this configuration must be rated for that much power (and then the board guys ratchet that up for overclocking headroom). They might be forced to either pull an AMD and release it and hope the Mobo manufactures built a good enough buffer or inform their customers about compatibility or two require a second version of the socket specific to the HCC connections.

Basically just because a socket can take more power doesn't mean that Intel can just plop a chip in it. AMD learned this the hard way.

 

[Topweasel]

Xeon Gold 6150 (Skylake) - 18C (36T) - 2.7 Ghz/3.7Ghz (base/turbo) in 165W TDP. So it can be done.

If you'll still say that 2.7 Ghz it's not going to fly for their Extreme platform, I'm going to say 2 things:

1. The higher core count CPU always had a low clocked base compared to lower clocked parts

2. I don't think Intel is crazy enough to show up with a equivalent of Xeon Gold 6154 - 18C (36T) - 3.0 GHz base - 200W TDP - for $1999 that would probably go for a $5000-6000 price bracket.

I hadn't had a chance to really analyze the rest of the server Skylake release. 2.7 is probably close enough to fine. I do think they will have problems selling people on the idea of an extreme gaming CPU (even though the users of this if any won't game with it), at under 3GHz. That is my limits its platform Viability at that point almost 40% slower per core than the 7700.

Well it's all about whether Intel really intends to sell any of these or if so what they want the demand to be. This is a tightrope that Intel will have to walk for the next year. Fear of cannibalizing sales vs. loss of sales to AMD. The problem with this whole lineup is that it was relatively easy for AMD to show up with 16c chip. While I would love one and AMD's pricing in particular if true is enticing, these aren't really going to be sold to gamers really. Once you get above 10 or so cores on either brand it's all about home servers, super workstation, and epeen stuff. So in the long run it doesn't matter. So yeah 2.7 is good enough, but it's lower than even a mobile i7. Not sure how that will fly even if it doesn't matter. For the 14c and 16c though clock speed will matter more if the rumors for their competitor are true.

 

[jpiniero]

2. I don't think Intel is crazy enough to show up with a equivalent of Xeon Gold 6154 - 18C (36T) - 3.0 GHz base - 200W TDP - for $1999 that would probably go for a $5000-6000 price bracket.

Not having ECC and 1S only should be a big enough difference.

 

[senseamp]

Where are Intel's margins going if they have to sell 1 near perfect huge die to compete with AMD's 4 small dies? Do they have anything like Infinity fabric on the horizon to replicate AMD's strategy?

 

[jpiniero]

Where are Intel's margins going if they have to sell 1 near perfect huge die to compete with AMD's 4 small dies? Do they have anything like Infinity fabric on the horizon to replicate AMD's strategy?

Yeah, EMIB. Which is (on paper at least) much better than IF.

Yields at 14 nm are probably so good at this point that it's not that big of a problem. But at 10 (and especially 7) it would be.

 

[formulav8]

Do they have anything like Infinity fabric on the horizon to replicate AMD's strategy?

That is a weakness for Intel's it seems. Intel need's perfect dice for certain chips. AMD brought a true trump card with IF. Not sure if Intel has the same AMD strategy though in the pipeline. Intel selling Xeon's so cheap will reduce their margins for sure in the HEDT space especially having now to compete in that market. Pretty much their HEDT series are renamed Xeons with segmented features like blocking ECC capabilities.

 

[senseamp]

Yeah, EMIB. Which is (on paper at least) much better than IF.

Yields at 14 nm are probably so good at this point that it's not that big of a problem. But at 10 (and especially 7) it would be.

Interesting, thanks, I was not aware of EMIB. Seems like a good idea. But it looks like it requires Intel to embed extra bridge dies in the package, whereas Infinity fabric is over regular package traces? I understand that EMIB should allow for finer pitch and more bandwidth, but If AMD is getting good enough scaling and power consumption with IF, that seems like a better solution to me, since there is going to be extra cost and yield loss from EMIB silicon and integration. At this point, it looks like x86 CPU is headed for commoditization, so the guy with the lowest cost will win.

 

[senseamp]

That is a weakness for Intel's it seems. Intel need's perfect dice for certain chips. AMD brought a true trump card with IF. Not sure if Intel has the same AMD strategy though in the pipeline. Intel selling Xeon's so cheap will reduce their margins for sure in the HEDT space especially having now to compete in that market. Pretty much their HEDT series are renamed Xeons with segmented features like blocking ECC capabilities.

Yeah, their position seems pretty precarious here. I think even $2K is probably BK's wishful thinking in HEDT, especially with their PCIE shortfall, since the professionals spending that kind of money on CPU are also likely to be using accelerators that the CPUs need to feed over PCIE. It's kind of shocking that Intel doesn't seem to have anything in the back pocket for this moment that everyone knew was coming for years and years.

 

[formulav8]

much better than IF.

What makes it much better? And what is the expected ETA if any?

 

[moonbogg]

It doesn't matter how much overhead X299 boards may have for overclocking. Intel can't offer a 200 watt chip for a lesser watt rated platform, regardless of overhead. It has to be according to spec and by the book. Additional overhead is a feature offered by the vendors. If the 14, 16 and 18 core chips are supported by the current X299 boards, they will run at lower stock clocks in order to remain within the rated TDP. Proper performance can only be achieved by overclocking. This will look terrible for reviews since the numbers are always compared at stock clocks and overclocking is covered separately. Overclocking is usually dependent on quality of cooling capacity, chip lottery etc. The popular numbers that people see on reviews and all over the internet will be stock clocks, and to have decent clocks a socket revision will be needed with a higher official TDP rating, like it or not. Intel is in a tight spot right now and threadripper will mop the floor with them, especially in terms of value.

(Drops mic)

 

[maddie]

Yeah, EMIB. Which is (on paper at least) much better than IF.

Yields at 14 nm are probably so good at this point that it's not that big of a problem. But at 10 (and especially 7) it would be.

How can you compare the two as if they're even close to being the same. IF is way beyond what EMIB attempts to do which is to simply connect two die.

I can see someone arguing that EMIB is better than Silicon Interposers, even though EMIB really uses tiny interposers, but comparing EMIB positively to IF means you haven't a clue as to deep level of integration IF allows.

 

[ddogg]

Wouldn't be surprised if they mop the floor performance wise, especially if infinity fabric scales as well as they are claiming it to be.

 

[jpiniero]

What makes it much better? And what is the expected ETA if any?

Connecting directly instead of wiring in the package should enable much higher bandwidth and lower latency between the dies. Of course, this being Intel, it will allow them to really get creative on segmenting, so watch out. First product would probably be Icelake mainstream in late 2018.

How can you compare the two as if they're even close to being the same. IF is way beyond what EMIB attempts to do which is to simply connect two die.

Yeah. I guess the end result of connecting multiple die is the most important part of it though.

 

[Bouowmx]

What is the default power limit on Intel multiplier-unlocked motherboards? Browsing the HWBot Intel XTU submissions and settings used for the submissions, I find that power limit is set greater than 1 kW. Quite the adherence to specs and by-the-bookness? Idk. I do not see how TDP rating will be relavent for X299, when there is practically almost no restriction on power.

TDP rating will be relavent for multiplier-locked motherboards and Xeon processors, ex. some Supermicro 2-socket LGA 2011-3 motherboards only supporting TDP up to 145 W (not the maximum 160 W).

 

[maddie]

Yeah, their position seems pretty precarious here. I think even $2K is probably BK's wishful thinking in HEDT, especially with their PCIE shortfall, since the professionals spending that kind of money on CPU are also likely to be using accelerators that the CPUs need to feed over PCIE. It's kind of shocking that Intel doesn't seem to have anything in the back pocket for this moment that everyone knew was coming for years and years.

Did many really know this was coming? I remember seeing a lot of negative posts, articles, etc, even after early test results were revealed and I can't remember reading much at all about the remarkable scaling ability of IF [A huge advantage].

Intel appears to have payed the price of not having to fight for too long. You see that in all aspects of life. Business, the military, sports, etc. The wrong people [who tend to be politically very astute] get put in charge as even an average leader will still look good. Competitors reorganize, return stronger and pummel the previous champ. This X platform release is an indication of the chaos happening right now. The next couple of years will be very interesting.

 

[Topweasel]

Connecting directly instead of wiring in the package should enable much higher bandwidth and lower latency between the dies. Of course, this being Intel, it will allow them to really get creative on segmenting, so watch out. First product would probably be Icelake mainstream in late 2018.



Yeah. I guess the end result of connecting multiple die is the most important part of it though.

It isn't and that is why you are wrong. It's the interconnect on everything, It's the interconnect for CCX's, it's the interconnect on the GPU dies. It's the interconnect between CPU and GPU on the same die in APU's, it's an interconnect that allows the CPU to be an hbc connector to on-board memory. It's least important connection aspect is CPU dies to CPU dies.

 

[maddie]

It isn't and that is why you are wrong. It's the interconnect on everything, It's the interconnect for CCX's, it's the interconnect on the GPU dies. It's the interconnect between CPU and GPU on the same die in APU's, it's an interconnect that allows the CPU to be an hbc connector to on-board memory. It's least important connection aspect is CPU dies to CPU dies.

A new motto for AMD.

Design once, Use many.

 

[Excessi0n]

What is the default power limit on Intel multiplier-unlocked motherboards? Browsing the HWBot Intel XTU submissions and settings used for the submissions, I find that power limit is set greater than 1 kW. Quite the adherence to specs and by-the-bookness? Idk. I do not see how TDP rating will be relavent for X299, when there is practically almost no restriction on power.

The motherboard needs to be able to deal with the rated TDP. Many will be able to deal with much more. But some lower-end boards will be designed to handle the bare minimum, so it can be problematic if the processor manufacturer releases a later chip with a higher TDP than the platform originally specified.

The power limits listed in HWBot submissions are pretty meaningless. You can set the limit in the BIOS to whatever you want to make sure that power throttling will never happen. I think I have my 6700K's power limit set to 4096 watts, or something silly like that, even though it's never pulled over 185 and the board would spontaneously combust at a few hundred.

 

[imported_ats]

Where are Intel's margins going if they have to sell 1 near perfect huge die to compete with AMD's 4 small dies? Do they have anything like Infinity fabric on the horizon to replicate AMD's strategy?

you mean the exact same strategy that AMD has used for like a decade+?

 

[R0H1T]

What is the default power limit on Intel multiplier-unlocked motherboards? Browsing the HWBot Intel XTU submissions and settings used for the submissions, I find that power limit is set greater than 1 kW. Quite the adherence to specs and by-the-bookness? Idk. I do not see how TDP rating will be relavent for X299, when there is practically almost no restriction on power.

TDP rating will be relavent for multiplier-locked motherboards and Xeon processors, ex. some Supermicro 2-socket LGA 2011-3 motherboards only supporting TDP up to 145 W (not the maximum 160 W).

You mean unlimited power, so when was the last time you saw a 1000W CPU?

OR how about Intel restrict the base clock of 18c & other HCC dies for x299 so that they don't turn into mini Chernobyl?

 

[Arachnotronic]

you mean the exact same strategy that AMD has used for like a decade+?

They used an MCM with Magny Cours and the Bulldozer stuff. This is not a new strategy and it is certainly not revolutionary as some seem to think, IMHO.

 

[Topweasel]

They used an MCM with Magny Cours and the Bulldozer stuff. This is not a new strategy and it is certainly not revolutionary as some seem to think, IMHO.

Zen/Zeppelin are the major accomplishments along with IF. Before even on MCM the dies communicated through FSB. Now AMD has a platform that will allow almost single die loke communication between Dies not only on socket but between them. MCM is old, AMD leveraging MCM to us IF to ramp up the cores at a relatively low cost is revolutionary. AMD can provide almost linear compute growth all the way up 64 cores and 128 threads through $450 in silicon. The big question in EPYC and across sockets what the cache performance (not just latency, but how the cache is handled).

Add the zeppelin efficiency and you have clocks that Intel can't compete with.

The other reason is because it just plainly is a nice option considering the performance monopoly we have been used to. Zen not only coming out ready to fight but fighting above it's weight class is great and that is exciting.

 

[Vaporizer]

It will be interesting to see when AMD transfers this to GPU with Navi. Imaginge how performance and pricewise the fusing of 2 300mm dies will compete with an 600mm die.
Interesting times ahead of us.