Intel 12-core i9-7920X features a 2.9GHz base clock.

[The Benjamins]

2 minutes ago, MageTank said:

How did you arrive to this conclusion based on the max boost clock? Max boost of the 7700k is 4.5ghz, and yet we run 5.2ghz on them all the time. The max clock speeds of these CPU's are going to be determined by the quality of your motherboard and cooling solution. You can still achieve 4.5ghz+ if you have a great board with a solid VRM and an extremely potent cooling solution. A delid is certainly going to be required to hit higher than 4.5ghz, but it's also not outside of the realm of possibilities, as Der8auer seems quite confident that his tool will be out soon.

 

Either way, AMD's CPU's will require the exact same conditions to be met in order to get all of those cores to 4ghz. Unlike Intel, 4ghz will be the limit for AMD, as even the consumer die's that these are based on, seldom go higher. 

I mean in stock config. I do expect Intels CPU's to win at top OCs

 

I do feel that it will not be hard to hit ~4Ghz on the 1950x and 1920x.

 

I do feel that the 14-18 core Intel CPUs will give a person hell when trying to overclock.

[MageTank]

Just now, The Benjamins said:

I mean in stock config. I do expect Intels CPU's to win at top OCs

 

I do feel that it will not be hard to hit ~4Ghz on the 1950x and 1920x.

 

I do feel that the 14-18 core Intel CPUs will give a person hell when trying to overclock.

It won't be, that's the advantage of using an MCM design over a large monolithic die. With much better yields on multiple smaller dies, you can hit their average clock speeds much easier. Now, the average overclock for Ryzen CPU's (according to my math done using the Ryzen 7 owners club on OCN), they are able to hit 3.89ghz on average. We can round that up to say, 3.9ghz. Now, it's possible AMD might do some binning to use lower leakage die's for their CPU's, given how massive these will be. If that is the case, and they get use binned samples that can hit 4ghz, then that will be pretty easy to achieve, assuming you have a decent cooler. The advantage here being, they are soldered, and won't suffer the "die to IHS" heat transfer bottleneck that Intel currently has on their X299 platform.

 

That being said, Intel CPU's are still very capable of going higher, it's all a matter of using the right motherboard, and a custom loop. A motherboard with 2x8EPS is required, along with a VRM capable of pushing at least 1000w (not that hard of a requirement to meet, given we have boards capable of pushing 1200-1300w through their VRM already, including Intel's engineering board that these are tested on). Active cooling on the VRM might be a must at that point as well. Basically, if you intend to overclock the 12c+ CPU's, stop pretending to be an enthusiast, and actually become one, lol. 

 

Threadripper for the most part, won't be trading that many blows with Intel once the enthusiasts iron out those requirements. EPYC on the other hand, still has the chance to be 4ghz (again, boon of using an MCM design) on 64 threads. I have yet to see a Xeon clock that high, with that many threads. Talk about massive lifting, assuming you could somehow keep it cool. 

[The Benjamins]

1 minute ago, MageTank said:

It won't be, that's the advantage of using an MCM design over a large monolithic die. With much better yields on multiple smaller dies, you can hit their average clock speeds much easier. Now, the average overclock for Ryzen CPU's (according to my math done using the Ryzen 7 owners club on OCN), they are able to hit 3.89ghz on average. We can round that up to say, 3.9ghz. Now, it's possible AMD might do some binning to use lower leakage die's for their CPU's, given how massive these will be. If that is the case, and they get use binned samples that can hit 4ghz, then that will be pretty easy to achieve, assuming you have a decent cooler. The advantage here being, they are soldered, and won't suffer the "die to IHS" heat transfer bottleneck that Intel currently has on their X299 platform.

 

That being said, Intel CPU's are still very capable of going higher, it's all a matter of using the right motherboard, and a custom loop. A motherboard with 2x8EPS is required, along with a VRM capable of pushing at least 1000w (not that hard of a requirement to meet, given we have boards capable of pushing 1200-1300w through their VRM already, including Intel's engineering board that these are tested on). Active cooling on the VRM might be a must at that point as well. Basically, if you intend to overclock the 12c+ CPU's, stop pretending to be an enthusiast, and actually become one, lol. 

 

Threadripper for the most part, won't be trading that many blows with Intel once the enthusiasts iron out those requirements. EPYC on the other hand, still has the chance to be 4ghz (again, boon of using an MCM design) on 64 threads. I have yet to see a Xeon clock that high, with that many threads. Talk about massive lifting, assuming you could somehow keep it cool. 

well by ~4GHz I was expecting 3.8-4.1GHz going off of silicon lottery's OC probability on the 1800x (I expect this level of die quality)

 

Again I full expect Intel to when when OCed, just that AMD seems to be bring a better offer at stock for the people that use/need CPU's like this and that are not going to tinker/OC them.

[Drak3]

59 minutes ago, MageTank said:

Talk about massive lifting, assuming you could somehow keep it cool.

Segmented CPU water block, that puts each die on it's own loop.

Not pretty, but should do the jobe without huge compromises on cooling perf.

 

EKWB, for royalties, I accept a year supply of green tea ice cream balls, Naked juice, and wasabi coated peanuts.

[Brooksie359]

1 hour ago, MageTank said:

How did you arrive to this conclusion based on the max boost clock? Max boost of the 7700k is 4.5ghz, and yet we run 5.2ghz on them all the time. The max clock speeds of these CPU's are going to be determined by the quality of your motherboard and cooling solution. You can still achieve 4.5ghz+ if you have a great board with a solid VRM and an extremely potent cooling solution. A delid is certainly going to be required to hit higher than 4.5ghz, but it's also not outside of the realm of possibilities, as Der8auer seems quite confident that his tool will be out soon.

 

Either way, AMD's CPU's will require the exact same conditions to be met in order to get all of those cores to 4ghz. Unlike Intel, 4ghz will be the limit for AMD, as even the consumer die's that these are based on, seldom go higher. 

i don't know tbh. the intel 6 core required a freaking custom loop to hit 4.7 ghz so im thinking the 12 core is going to need a custom loop to even get much past 4ghz. now that is an expensive cooling solution required to overclock these cpus. yeah when overclocked with a nice custom loop they will likely beat amd cpus in some tasks but I honestly think it will turn many people away that you need a custom loop. 

[Brooksie359]

55 minutes ago, The Benjamins said:

well by ~4GHz I was expecting 3.8-4.1GHz going off of silicon lottery's OC probability on the 1800x (I expect this level of die quality)

 

Again I full expect Intel to when when OCed, just that AMD seems to be bring a better offer at stock for the people that use/need CPU's like this and that are not going to tinker/OC them.

or spend a thousand bucks for a custom loop to overclock. i think the fact that you wont be able to get by with a AIO is going to turn people off and make it not sell nearly as well. i mean those that will buy the cpu are probably those who were going to in the first place. 

[Drak3]

1 minute ago, Brooksie359 said:

i don't know tbh. the intel 6 core required a freaking custom loop to hit 4.7 ghz so im thinking the 12 core is going to need a custom loop to even get much past 4ghz. now that is an expensive cooling solution required to overclock these cpus. yeah when overclocked with a nice custom loop they will likely beat amd cpus in some tasks but I honestly think it will turn many people away that you need a custom loop. 

A custom loop, yes. A massive custom loop, not really. Same goes for my 5930K, but once cooling isn't a bottleneck, one can go ham until they can't anymore, which the stop is seemingly in the same neighborhood as the quad core 6700K, it's architectural older brother.

[MageTank]

16 minutes ago, Brooksie359 said:

i don't know tbh. the intel 6 core required a freaking custom loop to hit 4.7 ghz so im thinking the 12 core is going to need a custom loop to even get much past 4ghz. now that is an expensive cooling solution required to overclock these cpus. yeah when overclocked with a nice custom loop they will likely beat amd cpus in some tasks but I honestly think it will turn many people away that you need a custom loop. 

Are you referring to Broadwell-E, or Skylake-E? The Skylake-E CPU's are bottlenecked by the heat transfer from the die to the IHS. @done12many2's water wasn't even getting warm, let alone hot, and his CPU would still hit high temperatures. From what we've seen from Der8auer, a delid was able to net 300mhz while remaining at the same temperatures as a non-delidded SKU.

 

The cooler something runs, the less volts required to make it stable at any given temperature. It should go without saying that once you improve the transfer from the die to the IHS, you will lower the temperatures and improve the overclocking. Once Der8auer releases his delid tool, we might get some interesting results as to what these Skylake-X CPU's are really capable of when proper materials are used between the die and IHS.

[Taf the Ghost]

2 hours ago, The Benjamins said:

afaik it is 4.0 for TB2 and 4.4 for TB3 (2 core boost).

I think TB2 is more then 4 cores, but it really does look like that the 16 core part will be clocked lower then the 1950x in base and boost (except 2 core boost)

 

I did some napkin math and I figure the 14 core need to be at 2.7 or better (base clock) to beat the 1920x, and the 16 core part needs to be at 2.9 or better (base clock) to beat the 1950x. and the 18 core part needs to be 2.6 or better (base clock) to beat the 1950x.

 

so I feel that intels top CPUs will not be able to compete. NOTE: I can be wrong just a shot in the dark.

 

  Reveal hidden contents

 

estimated Cinebench scores for intels parts on the left and estimated OC Cinebench scores of TR on the right.

 

The X299 points to something of a "max" power draw limit at standard BIOS settings. It's going to be interesting to see if they run smack into that over the 12c part.

[cj09beira]

its a shame that the platforms that most need strong vrms are the ones with very little space for it,

if i was a mobo manufacturer right now i would seriously consider using a daughter board for the vrm or moving the first pcie slot down, so that there is space for vrm on the top and bottom

[Brooksie359]

49 minutes ago, MageTank said:

Are you referring to Broadwell-E, or Skylake-E? The Skylake-E CPU's are bottlenecked by the heat transfer from the die to the IHS. @done12many2's water wasn't even getting warm, let alone hot, and his CPU would still hit high temperatures. From what we've seen from Der8auer, a delid was able to net 300mhz while remaining at the same temperatures as a non-delidded SKU.

 

The cooler something runs, the less volts required to make it stable at any given temperature. It should go without saying that once you improve the transfer from the die to the IHS, you will lower the temperatures and improve the overclocking. Once Der8auer releases his delid tool, we might get some interesting results as to what these Skylake-X CPU's are really capable of when proper materials are used between the die and IHS.

i was referring to skylake-x. i was specifically referring to the hardware unboxed video where he goes over temp and power draw and shows that the thing produces alot of heat. yeah you can delid a 1200 dollar part and get a custom loop but I am confident that many will not want to have to do that just to overclock.

[mr moose]

2 hours ago, The Benjamins said:

Again I full expect Intel to when when OCed, just that AMD seems to be bring a better offer at stock for the people that use/need CPU's like this and that are not going to tinker/OC them.

 

That's me, I don't overclock. The only times I ever had were spare machines that I tinkered with for fun, but all the everyday systems in my house are running stock out of the box.

[Taf the Ghost]

40 minutes ago, mr moose said:

 

That's me, I don't overclock. The only times I ever had were spare machines that I tinkered with for fun, but all the everyday systems in my house are running stock out of the box.

There's also going to be a snag in all of this when AMD releases lower-clocked versions (probably with an air-cooler) at some point in the future. We pretty much how all of these cores (both Intel & AMD) are going to perform since they're copies of other CPUs. A 3.0 Ghz base-clocked 12c Threadripper will live just fine on air. Especially if it's like $649 USD.

 

And you'll still have more room for your I/O setup on TR.

 

The reality is that AMD has a better "core" while Intel has a much better "process" in the 2017 generation. We're seeing that play out in the results. Which is great for the consumer, maybe not so great for the companies.

[MageTank]

1 hour ago, Brooksie359 said:

i was referring to skylake-x. i was specifically referring to the hardware unboxed video where he goes over temp and power draw and shows that the thing produces alot of heat. yeah you can delid a 1200 dollar part and get a custom loop but I am confident that many will not want to have to do that just to overclock.

I wouldn't exactly consider Hardware Unboxed's 360mm custom loop a "requirement", given it performs not much better than a 280mm CLC: http://www.thinkcomputers.org/thermaltake-pacific-rl360-d5-rgb-water-cooling-kit-review/4/

 

Again, it doesn't really matter how big of a loop you put these CPU's under, the bottleneck lies within the transfer of heat between the die and IHS. You simply can't cool it fast enough, regardless of the loop. Will you fare better with a more robust custom loop? Certainly, but you will inevitably hit the same limitations sooner or later. 

 

Most AIO's, when compared to custom loops, are bottlenecked by the pump assuming all else is the same (rad size, block size, etc). However, on these CPU's, I doubt the pump on an AIO will bottleneck before the paste between the IHS/die does, but I could be wrong. @TahoeDust is using an AIO, and pushing very respectable numbers with his 7820X, an 8c/16t SKU. Granted, he does use an AVX offset for the more "hotter" workloads, but it's silly to not use the tools provided to you if you know which workloads you intend on using. Gone are the days of setting an all-core overclock, and expect it to work in every scenario. Enter the new age of "smart" or "variable" overclocking. I've even seen people go as far as designating specific clock speeds per core, essentially making their own turbo boost. Smart way to solve this thermal "problem".

 

Here is a link to Tahoe's thread with his Skylake-X chip: 

 

Either way, it's a far cry from "needing" a custom loop to overclock a 6 core CPU to 4.7. Just because hardware unboxed used one (and hit thermal junction under AVX), doesn't mean everyone else will in their intended use case. While I agree that delidding won't be everyone's go-to solution to solve this "problem", enthusiasts that invest in this "enthusiast" platform are more than willing to go to extreme's to get every ounce of performance they want. To them, the end always justifies the means. 

[crysilis]

"don't overclock it to avoid overheating"

-Intel

 

[Zodiark1593]

1 hour ago, mr moose said:

 

That's me, I don't overclock. The only times I ever had were spare machines that I tinkered with for fun, but all the everyday systems in my house are running stock out of the box.

I had fun overclocking my old phone as well.

 

[Brooksie359]

1 hour ago, MageTank said:

I wouldn't exactly consider Hardware Unboxed's 360mm custom loop a "requirement", given it performs not much better than a 280mm CLC: http://www.thinkcomputers.org/thermaltake-pacific-rl360-d5-rgb-water-cooling-kit-review/4/

 

Again, it doesn't really matter how big of a loop you put these CPU's under, the bottleneck lies within the transfer of heat between the die and IHS. You simply can't cool it fast enough, regardless of the loop. Will you fare better with a more robust custom loop? Certainly, but you will inevitably hit the same limitations sooner or later. 

 

Most AIO's, when compared to custom loops, are bottlenecked by the pump assuming all else is the same (rad size, block size, etc). However, on these CPU's, I doubt the pump on an AIO will bottleneck before the paste between the IHS/die does, but I could be wrong. @TahoeDust is using an AIO, and pushing very respectable numbers with his 7820X, an 8c/16t SKU. Granted, he does use an AVX offset for the more "hotter" workloads, but it's silly to not use the tools provided to you if you know which workloads you intend on using. Gone are the days of setting an all-core overclock, and expect it to work in every scenario. Enter the new age of "smart" or "variable" overclocking. I've even seen people go as far as designating specific clock speeds per core, essentially making their own turbo boost. Smart way to solve this thermal "problem".

 

Here is a link to Tahoe's thread with his Skylake-X chip: 

 

Either way, it's a far cry from "needing" a custom loop to overclock a 6 core CPU to 4.7. Just because hardware unboxed used one (and hit thermal junction under AVX), doesn't mean everyone else will in their intended use case. While I agree that delidding won't be everyone's go-to solution to solve this "problem", enthusiasts that invest in this "enthusiast" platform are more than willing to go to extreme's to get every ounce of performance they want. To them, the end always justifies the means. 

Still supports my main point that you need to do alot just to be able to overclock the thing which means alot of people are going to opt for the cheaper cpu that likely can overclock well with the run of the mill AIO and doesn't require you to delid the cpu. I mean intel kinda messed up when they made freaking 1200 dollar cpu that requires you to delid it just to overclock well. For 1200 dollars it should come with a proper thermal solution that can transfer the heat properly off the die.

[MageTank]

1 minute ago, Brooksie359 said:

Still supports my main point that you need to do alot just to be able to overclock the thing which means alot of people are going to opt for the cheaper cpu that likely can overclock well with the run of the mill AIO and doesn't require you to delid the cpu. I mean intel kinda messed up when they made freaking 1200 dollar cpu that requires you to delid it just to overclock well. For 1200 dollars it should come with a proper thermal solution that can transfer the heat properly off the die.

Yeah, we've been complaining about their choice to forego soldering the die ever since we saw Der8auer delid it pre-launch. Not only did they forego solder, they made it more difficult to delid this time around with a razor and vice. There are a bank of capacitors on the substrate, so sliding a blade between the IHS and substrate is a no-go. The glue they use, has the potential to leak out on to the capacitors (according to Der8auer) so delidding with a vice and even a delid tool, may still impose a risk depending on whether or not the glue leaked out on to a cap. 

 

It's certainly going to be a chore to "unlock" the real potential of these CPU's, but that is the premium people pay when they want the absolute best performance. Even if they risk destroying the product, they will still take that risk because it's in their nature. Threadripper does indeed offer a better compromise for performance if you want minimal effort, but it still has problems as well. It's memory controller is far weaker than Intel's (something that is extremely important when trying to keep more cores properly fed) and It's AVX performance will be even more lackluster in comparison to the AVX512(F) capable chips, that can do twice the bit-ops/clock as the consumer AVX2 chips, and 4x the bit-ops/clock that AMD can do under AVX. People that intend to make use of AVX, may find themselves looking at Intel regardless of the amount of effort required to tame the heat on those CPU's.

 

Hopefully, when X299 gets it's refresh, Intel decides to go with solder. I also hope to see some slight refinements that makes power consumption a little better, so that cheaper boards become less of a fire risk than what they currently are. EPYC will be interesting though, and I can't wait to see what it looks like when properly cooled. 64 threads at 4ghz is going to be absolutely insane. 

[Taf the Ghost]

2 minutes ago, MageTank said:

Yeah, we've been complaining about their choice to forego soldering the die ever since we saw Der8auer delid it pre-launch. Not only did they forego solder, they made it more difficult to delid this time around with a razor and vice. There are a bank of capacitors on the substrate, so sliding a blade between the IHS and substrate is a no-go. The glue they use, has the potential to leak out on to the capacitors (according to Der8auer) so delidding with a vice and even a delid tool, may still impose a risk depending on whether or not the glue leaked out on to a cap. 

 

It's certainly going to be a chore to "unlock" the real potential of these CPU's, but that is the premium people pay when they want the absolute best performance. Even if they risk destroying the product, they will still take that risk because it's in their nature. Threadripper does indeed offer a better compromise for performance if you want minimal effort, but it still has problems as well. It's memory controller is far weaker than Intel's (something that is extremely important when trying to keep more cores properly fed) and It's AVX performance will be even more lackluster in comparison to the AVX512(F) capable chips, that can do twice the bit-ops/clock as the consumer AVX2 chips, and 4x the bit-ops/clock that AMD can do under AVX. People that intend to make use of AVX, may find themselves looking at Intel regardless of the amount of effort required to tame the heat on those CPU's.

 

Hopefully, when X299 gets it's refresh, Intel decides to go with solder. I also hope to see some slight refinements that makes power consumption a little better, so that cheaper boards become less of a fire risk than what they currently are. EPYC will be interesting though, and I can't wait to see what it looks like when properly cooled. 64 threads at 4ghz is going to be absolutely insane. 

The X-series has been had a 3-year cycle since 2011, so I imagine we'll get some Refreshed CPUs on the platform. My assumption is an Icelake-X in 2019? Broadwell-E came out 2 years later, but we might see the timeline tighter this time around if AMD rolling stronger in the HEDT realm. (Though these are such low-volume parts I'm not sure they care all that much. Just as long as they have more cores. Minus the fact AMD could have TR2 parts 24c/48t in 2019 for all we know.)

[MageTank]

1 minute ago, Taf the Ghost said:

The X-series has been had a 3-year cycle since 2011, so I imagine we'll get some Refreshed CPUs on the platform. My assumption is an Icelake-X in 2019? Broadwell-E came out 2 years later, but we might see the timeline tighter this time around if AMD rolling stronger in the HEDT realm. (Though these are such low-volume parts I'm not sure they care all that much. Just as long as they have more cores. Minus the fact AMD could have TR2 parts 24c/48t in 2019 for all we know.)

With Coffeelake coming off the heels of Kaby so fast, it wouldn't be impossible to see a Coffeelake-X lineup, with basically the only difference being solder, lol. Who knows, perhaps Intel's marketing team will read this and be like "good idea, another way to obfuscate our product line AND profit at the same time?!?!".

 

I should really start charging for these ideas. 

[Taf the Ghost]

1 hour ago, MageTank said:

With Coffeelake coming off the heels of Kaby so fast, it wouldn't be impossible to see a Coffeelake-X lineup, with basically the only difference being solder, lol. Who knows, perhaps Intel's marketing team will read this and be like "good idea, another way to obfuscate our product line AND profit at the same time?!?!".

 

I should really start charging for these ideas. 

Well, it's about design branches. Skylake-X and the previous E CPUs are from the Server Products department. So the refresh is with whatever Xeon upgrade comes next and I believe that's Icelake.  AMD could outright have TR Refresh and TR-Zen2 CPUs out before Intel has brought out the Icelake-X models.  Which explains why the i9-7980XE will exist, actually.

[Joehockey90]

Wow, that's just hard to look at. I expected TR to handle Intel fairly well without too much competition but damn, that's just rough. 75% of the cores and 85% of the clock speed is not even close to being competitive, especially at 120% of the price.

[Brooksie359]

12 hours ago, MageTank said:

Yeah, we've been complaining about their choice to forego soldering the die ever since we saw Der8auer delid it pre-launch. Not only did they forego solder, they made it more difficult to delid this time around with a razor and vice. There are a bank of capacitors on the substrate, so sliding a blade between the IHS and substrate is a no-go. The glue they use, has the potential to leak out on to the capacitors (according to Der8auer) so delidding with a vice and even a delid tool, may still impose a risk depending on whether or not the glue leaked out on to a cap. 

 

It's certainly going to be a chore to "unlock" the real potential of these CPU's, but that is the premium people pay when they want the absolute best performance. Even if they risk destroying the product, they will still take that risk because it's in their nature. Threadripper does indeed offer a better compromise for performance if you want minimal effort, but it still has problems as well. It's memory controller is far weaker than Intel's (something that is extremely important when trying to keep more cores properly fed) and It's AVX performance will be even more lackluster in comparison to the AVX512(F) capable chips, that can do twice the bit-ops/clock as the consumer AVX2 chips, and 4x the bit-ops/clock that AMD can do under AVX. People that intend to make use of AVX, may find themselves looking at Intel regardless of the amount of effort required to tame the heat on those CPU's.

 

Hopefully, when X299 gets it's refresh, Intel decides to go with solder. I also hope to see some slight refinements that makes power consumption a little better, so that cheaper boards become less of a fire risk than what they currently are. EPYC will be interesting though, and I can't wait to see what it looks like when properly cooled. 64 threads at 4ghz is going to be absolutely insane. 

yeah there are definitely those people who already had their hearts set on doing a custom loop on the latest and greatest intel enthusiast cpus and they will of course stick with it regardless of the difficulty and same with the people who actually use avx but for everyone else its probably going to be a huge turnoff from the product.