Intel's massive 18-core Core i9 chip

[bearcatrp]

Didn't see this posted but delete if it is. Was considering AMD but will look hard at this chip!
http://www.pcworld.com/article/31984...siast-pcs.html

[csatahajos]

Well, considering that this chip is rumored to be 2000$ I think the Threadripper 1998 for 1000$ is still a nice option for 90% of the performance. Since ASUS teased the Zenith Extreme mobo this time around that (the mobo side) won't be a limiting factor for the AMD platform either. Besides nobody knows how many CCX-es are under the hood in Threadripper, so we might see AMD doing a hattrick and coming up with a 24 core chip just to have the absolute performance crown in the desktop segment (BTW the Sklyale-X HCC die appears to be holding 20 cores) .

[zanzabar]

Well, considering that this chip is rumored to be 2000$ I think the Threadripper 1998 for 1000$ is still a nice option for 90% of the performance. Since ASUS teased the Zenith Extreme mobo this time around that (the mobo side) won't be a limiting factor for the AMD platform either. Besides nobody knows how many CCX-es are under the hood in Threadripper, so we might see AMD doing a hattrick and coming up with a 24 core chip just to have the absolute performance crown in the desktop segment (BTW the Sklyale-X HCC die appears to be holding 20 cores) .

i think the performance difference will be more than that. they have 18 cores and should be one proper cpu per socket instead of 4 cpus dies per socket (or maybe 2 pairs of dies on 2 sockets on one super socket,) and assuming cooling and pwm should clock much higher. thread ripper should have more bang for the buck but i dont see it being 90% of or even 75% or 66% of the top intel.

[xlink]

Well, considering that this chip is rumored to be 2000$ I think the Threadripper 1998 for 1000$ is still a nice option for 90% of the performance. Since ASUS teased the Zenith Extreme mobo this time around that (the mobo side) won't be a limiting factor for the AMD platform either. Besides nobody knows how many CCX-es are under the hood in Threadripper, so we might see AMD doing a hattrick and coming up with a 24 core chip just to have the absolute performance crown in the desktop segment (BTW the Sklyale-X HCC die appears to be holding 20 cores) .

multithreaded difference

12.5% more cores,
15% higher performance per clock


We're looking at a ~30% difference there assuming identical clocks. I suspect HT is a bit weaker than AMD's SMT but being a monolithic chip with a ringbus architecture will likely give the i9 an edge that nullifies HT being slightly weaker so that's a wash.


For lightly threaded applications it's clocked around 10% higher and has around 15-25% more performance per clock so it's in the order of 30-40% faster for single threaded applications as well.

Yes, it's priced 2x as high for a reason.

[qcmadness]

multithreaded difference

12.5% more cores,
15% higher performance per clock


We're looking at a ~30% difference there assuming identical clocks. I suspect HT is a bit weaker than AMD's SMT but being a monolithic chip with a ringbus architecture will likely give the i9 an edge that nullifies HT being slightly weaker so that's a wash.


For lightly threaded applications it's clocked around 10% higher and has around 15-25% more performance per clock so it's in the order of 30-40% faster for single threaded applications as well.

Yes, it's priced 2x as high for a reason.

I don't expect the new Skylake-X can do the same clock as Threadripper 1998. Otherwise Intel would disclose those clock already.

[Sparky]

I don't think you're buying a 16+ core CPU for its single threaded applications...

[xlink]

I don't expect the new Skylake-X can do the same clock as Threadripper 1998. Otherwise Intel would disclose those clock already.

http://www.zdnet.com/article/intel-u...sumer-pc-chip/

The rumor mill says turbo 3.0 to 4.5GHz for the 10 core version. This is around 10% higher than the 8 core Ryzen.

I think it's a reasonable guess to say that the 18 core version of i9 will clock ~10% higher than the 16 core threadripper. I could be wrong, maybe the extended turbo won't end up very different and both will be around 4.1ish GHz.

[Andi64]

multithreaded difference

12.5% more cores,
15% higher performance per clock


We're looking at a ~30% difference there assuming identical clocks. I suspect HT is a bit weaker than AMD's SMT but being a monolithic chip with a ringbus architecture will likely give the i9 an edge that nullifies HT being slightly weaker so that's a wash.


For lightly threaded applications it's clocked around 10% higher and has around 15-25% more performance per clock so it's in the order of 30-40% faster for single threaded applications as well.

Yes, it's priced 2x as high for a reason.

15% higher IPC? Where do you get that number from? Ryzen is nearly on par per clock if you take gaming out of the equation. 5% IPC average would be my guess.

[Gilgamesh]

http://www.zdnet.com/article/intel-u...sumer-pc-chip/

The rumor mill says turbo 3.0 to 4.5GHz for the 10 core version. This is around 10% higher than the 8 core Ryzen.

I think it's a reasonable guess to say that the 18 core version of i9 will clock ~10% higher than the 16 core threadripper. I could be wrong, maybe the extended turbo won't end up very different and both will be around 4.1ish GHz.

Turbo 3.0 (4.5GHz) on Skylake-X just on 2 cores.
Base core for 10c/20t is 3.3GHz - Turbo 4.3 GHz.
I think that 12-14-16 will have a lesser base clock, 3.0-3.2GHz.

For Threadripper (rumors):
https://www.techpowerup.com/233945/a...dripper-models

1998X 3.50/3.9GHz + XFR 16/32
1977X 3.50/4.0GHz + XFR 14/28
1976X 3.60/4.1GHz + XFR 12/24
1955X 3.60/4.0GHz + XFX 10/20

[kromosto]

Because I needed more cores for my virtual machines I was looking for a proc with more cores for a year. I am glad that I haven't upgraded my 3770k until now. 7900x or 7820x will be my new proc. And I am thanking AMD for making available these processors this cheap... while they can't gain anything from this (on my side of course).

[BeepBeep2]

multithreaded difference

12.5% more cores,
15% higher performance per clock


We're looking at a ~30% difference there assuming identical clocks. I suspect HT is a bit weaker than AMD's SMT but being a monolithic chip with a ringbus architecture will likely give the i9 an edge that nullifies HT being slightly weaker so that's a wash.


For lightly threaded applications it's clocked around 10% higher and has around 15-25% more performance per clock so it's in the order of 30-40% faster for single threaded applications as well.

Yes, it's priced 2x as high for a reason.

Performance per clock will depend on application - 5-15% most likely.
Forget the Turbo Max 3.0 speeds for multithreading performance, though 4.2-4.5 GHz boost on high core parts will set AMD back a good 20-30% in single thread performance, I think they will be competitive in highly threaded workloads.

Do not forget that AMD has better scaling with Ryzen's SMT than Intel's Broadwell-E HT, though Intel has claimed to improve this on Skylake-X with the new cache structure.

3.5 base clock on 16C ThreadRipper will probably be a great match for Intel's 18C. Why? The 7900X already drops to a 3.3 GHz base clock, and Zeppelin seems to scale better perf/watt than intel - ie. The 7900X is already at 140w TDP @ 3.3 GHz, Intel will hit a TDP limit with more cores and need to lower clocks further.


3.5 16C would go up against 3.0 18C just fine, with the 18C CPU probably just barely edging it out.

If intel keeps clocks up, they win performance no doubt, but at what perf/watt?

Also - if users need high single thread performance, then KBL-X is the right CPU to buy. Users buying high core counts are probably doing high core count things.

PS- You claim 15% higher performance per clock, then up to 25%, then claimed 10%+15-25% = 30-40% single thread...not quite buying the math.

[AliG]

Performance per clock will depend on application - 5-15% most likely.
Forget the Turbo Max 3.0 speeds for multithreading performance, though 4.2-4.5 GHz boost on high core parts will set AMD back a good 20-30% in single thread performance, I think they will be competitive in highly threaded workloads.

Do not forget that AMD has better scaling with Ryzen's SMT than Intel's Broadwell-E HT, though Intel has claimed to improve this on Skylake-X with the new cache structure.

3.5 base clock on 16C ThreadRipper will probably be a great match for Intel's 18C. Why? The 7900X already drops to a 3.3 GHz base clock, and Zeppelin seems to scale better perf/watt than intel - ie. The 7900X is already at 140w TDP @ 3.3 GHz, Intel will hit a TDP limit with more cores and need to lower clocks further.


3.5 16C would go up against 3.0 18C just fine, with the 18C CPU probably just barely edging it out.

If intel keeps clocks up, they win performance no doubt, but at what perf/watt?

Also - if users need high single thread performance, then KBL-X is the right CPU to buy. Users buying high core counts are probably doing high core count things.

PS- You claim 15% higher performance per clock, then up to 25%, then claimed 10%+15-25% = 30-40% single thread...not quite buying the math.

IPC difference could actually be much larger for AVX applications - though AMD does tend to win a few brute int and float compute tests.

I think in the end it'll boil down to application needs. There are some where Skylake/Kabylake absolutely crush Ryzen, but for the rest AMD offers tremendous value.

[drmrlordx]

Will the 18c part be available from day 1 with the rest of the Skylake-X chips? It seems like it won't . . .

[kromosto]

18c part seems like just a counter for amd's 16c cpu just to say we have to most core cpu for desktops

[zanzabar]

18c part seems like just a counter for amd's 16c cpu just to say we have to most core cpu for desktops

they do have 22 core parts just sitting there in the xeon line that could have some switches flipped to be an i7/9. their 22 core part is even a 24 core part that they dont want to sell the 24 version to the mass market since you cannot cool it with air in a 2u.

[Vcc]

My doubt is what will be the stock core clocks for this 7980XE cpu and it's overclocking potential, especially under extreme

[zanzabar]

My doubt is what will be the stock core clocks for this 7980XE cpu and it's overclocking potential, especially under extreme

they did unlink the avx engine from the base, core, and IO multipliers. that should really help with extreme overclocking.

[alpha0ne]

18c part seems like just a counter for amd's 16c cpu just to say we have to most core cpu for desktops

Exactly..................mines bigger than yours

[FlanK3r]

http://www.tweaktown.com/news/57849/...ase/index.html

So, this year not...

[Metroid]

This is a healthy competition and it benefits us all. I wish AMD countered Intel in 2009 or so. Anyway, better late than never.

[Vcc]

http://www.tweaktown.com/news/57849/...ase/index.html

So, this year not...

So they just teased us... ;(
Still good to see that they are going to unleash these "hidden monsters" from Intel's cage

[StAndrew]

The 299 platform is a big mess. AMD changed the rules and Intel xeon domination has been challenged. They need to get over this and get over the fact that their i9's will have to overlap xeon functionality in order to compete.

[Particle]

Exactly..................mines bigger than yours

Intel: "Yeah, we can't deliver it any time soon, but since you're announcing it here is some rain for your parade."

It's just how business goes.

[Sparky]

So Intel got caught slumbering again. Not near as bad as back in the P4 days, but they did get a bit complacent, figuring on AMD never competing again, and got lazy. Now they're like "oh wait we can do that too.. um, hang on a sec..."

At least this does look a heck of a lot better than their first dual core attempt, Pentium D LOL

[Vcc]

Tbh i'm considering to get a threadripper instead of i9 to play on extreme.
My last Amd cpu was an 2006 Athlon 64 X2, loved that cpu