Intel Announces 9th Generation Core CPUs, Eight-Core Core i9-9900K

[jimmysmitty]

Do you think they plan a MCM design? Is AM4 even designed for MCM Chips? I can't see many ways to put 16 cores on AM4 unless they plan a monolithic die which is something that will push their TDP much higher like Intels 16/18 core CPUs and lower clocks.

EPYC2 is on 7nm and to offer a 64C128T model assuming the same quad-die arrangement would mean 16C32T dies. Rumors peg that hypothetical EPYC at 240W, which would be 60W apiece. At 7nm, a single higher-clocked die could easily fit under 100W.

AMD used the same die from Ryzen 1xxx to EPYC the first time around, it would make sense to expect the same thing again - if it ain't broke, don't fix it.

That's assuming that this 7nm process will fit the envelope that well.

As I said we will see what they do. I would be interested to see a 16 core CPU in the mainstream but I feel like it would be too much too fast. I would expect that even with the die shrink they wouldn't be able to keep the clocks as high as their 8 core parts and the mainstream doesn't need more lower clocked cores, it needs a decent number of higher clocked cores. Until software catches up and starts to utilize multiple cores more efficiently.

 

[TerryLaze]

They will probably reply by saying that proper mainstream processors should have integrated GPUs.

Not that it's a very meaningful distinction, in this price range, but it's one they could successfully make against 8-core Ryzens.

Ryzen are server parts scaled down to fit the desktop.
Intel's design is made for the desktop from the beginning.

 

[jimmysmitty]

They will probably reply by saying that proper mainstream processors should have integrated GPUs.

Not that it's a very meaningful distinction, in this price range, but it's one they could successfully make against 8-core Ryzens.

Ryzen are server parts scaled down to fit the desktop.
Intel's design is made for the desktop from the beginning.

Intels mainstream is. Their HEDT and HPC are not. They are designed for servers and workstations.

 

[bit_user]

Am i right that if I'm not planning going with more than 64gb of ram and utilizing more than 28 lanes i rather go z390&9900k route instead of 9800x route?

You mean 20 lanes? That's how much PCIe bandwidth Intel's mainstream socket has. It's worse than that - x16 lanes are direct-connected, but practically all the other slots & peripherals on your mobo compete for just x4 lanes' worth of bandwidth - in the form of the DMI3 connection.

AMD's AM4 socket gives you an extra x4 lanes.

 

[bit_user]

Ryzen are server parts scaled down to fit the desktop.

Scaled down to fit the desktop, or desktop parts scaled up to fit server.

Given the recent emphasis on cloud, I could see the argument for your perspective, but the real answer could be simply that it's a scalable architecture than can scale up to servers and down to APUs, laptops, and embedded.

I predict Intel will adopt AMD's strategy, before long. It has a number of compelling advantages.

 

[jimmysmitty]

Am i right that if I'm not planning going with more than 64gb of ram and utilizing more than 28 lanes i rather go z390&9900k route instead of 9800x route?

You mean 20 lanes? That's how much PCIe bandwidth Intel's mainstream socket has. It's worse than that - x16 lanes are direct-connected, but practically all the other slots & peripherals on your mobo compete for just x4 lanes' worth of bandwidth - in the form of the DMI3 connection.

AMD's AM4 socket gives you an extra x4 lanes.

Thats not correct. Intels mainstream has 16 PCIe lanes that are from the CPU to the slots, normally the x16 slots. The CPU and PCH communicate over DMI 3.0 which is equivalent to 3.94GB/s, the same speed as a PCIe 3.0 X4 link. DMI 3.0 was launched in 2015 with Skylake. DMI has been in use since 2004 in the 900 series chipsets.

It is the exact same as AMDs AM4 platform. The 2700X has 16 PCIe 3.0 lanes for GPU slots with 4 additional for PCH communication so AM4 is limited to the same 3.94GB/s bandwidth.

As for the chipsets, Intels has up to 24 PCIe 3.0 lanes while AMDs current has 8 PCIe 2.0 lanes with their Z490 (they seriously need to be more creative in the naming as this will confuse people when Intel eventually gets to the 400 series chipsets) is rumored to have the same 8 PCIe 2.0 lanes and an additional 4 PCIe 3.0 lanes, I would surmise for NVMe drives more than anything else.

 

[bit_user]

Thats not correct.

I disagree.

Intels mainstream has 16 PCIe lanes that are from the CPU to the slots, normally the x16 slots. The CPU and PCH communicate over DMI 3.0 which is equivalent to 3.94GB/s, the same speed as a PCIe 3.0 X4 link.

So far, same as I said.

It is the exact same as AMDs AM4 platform. The 2700X has 16 PCIe 3.0 lanes for GPU slots with 4 additional for PCH communication so AM4 is limited to the same 3.94GB/s bandwidth.

You're forgetting the x4 NVMe connection, which is simply another x4 PCIe link.

https://www.amd.com/en/products/chipsets-am4#Specifications

On Intel's LGA 1151, your NVMe would go through the chipset (i.e. sharing the DMI3 link bandwidth with everything else, which also adds latency). I won't claim that single M.2 SSD users would typically notice the performance hit, but in certain circumstances (esp. if you run multiple NVMe drives in RAID).

 

[jimmysmitty]

Thats not correct.

I disagree.

Intels mainstream has 16 PCIe lanes that are from the CPU to the slots, normally the x16 slots. The CPU and PCH communicate over DMI 3.0 which is equivalent to 3.94GB/s, the same speed as a PCIe 3.0 X4 link.

So far, same as I said.

It is the exact same as AMDs AM4 platform. The 2700X has 16 PCIe 3.0 lanes for GPU slots with 4 additional for PCH communication so AM4 is limited to the same 3.94GB/s bandwidth.

You're forgetting the x4 NVMe connection, which is simply another x4 PCIe link.

https://www.amd.com/en/products/chipsets-am4#Specifications

On Intel's LGA 1151, your NVMe would go through the chipset (i.e. sharing the DMI3 link bandwidth with everything else, which also adds latency). I won't claim that single M.2 SSD users would typically notice the performance hit, but in certain circumstances (esp. if you run multiple NVMe drives in RAID).

You need to be clearer then.

all the other slots & peripherals on your mobo compete for just x4 lanes' worth of bandwidth

That is not stating that NVMe has a different part but that ALL devices do which is wrong. If you run an x4 NVMe drive then the rest still use the 8 PCIe 2.0 lanes and communicate across the x4 PCIe lanes between the CPU and PCH. Otherwise you would have SATA and x2 NVMe but who wants that? It is not extra lanes to do with as you please as even your linked information shows it is designated for specific uses.

As for NVMe, I doubt there will be that much bottlenecking. The need to run NVMe in RAID is little to none. The other side is I think Intel is going to bank on Optane once it becomes a more affordable and feasible feature for the mainstream.

 

[AnimeMania]

What date and time will the NDA embargo lift?

 

[bit_user]

You need to be clearer then.

Apology accepted.

 

[mlee 2500]

The i5 9600k is the only chip here that really stands out as a possible purchase for me.

Performance per dollar still means something to some of us, and this would help me eek out a few more FPS @1080p on my aging gtx 980.

I like the specs on the i7 9700K, especially for gaming, but that price is just insane.

Yeah, depends how you look at it. When I built my last PC way back in late 2012, I decided to "splurge" the "insane" amount for the i7-3770K, with the net result of that extra couple hundred $$ being that my CPU lasted me over SIX YEARS. Amortizing that roughly $200 worth of extra performance over six years and thousands of hours of use....well, it seems like a BARGAIN now.

Even today I would probably do better by upgrading my GTX 980 instead of getting a new CPU, but I told myself that when the per-core performance improvement reached 50% then I would upgrade...and only today, with the latest 9th gen processors, is that finally the case. Barely.

 

[InvalidError]

I would be interested to see a 16 core CPU in the mainstream but I feel like it would be too much too fast. I would expect that even with the die shrink they wouldn't be able to keep the clocks as high as their 8 core parts and the mainstream doesn't need more lower clocked cores

We already know that AMD's engineering samples are hitting at least 4.5GHz so clock frequencies on 7nm should be fine, we just don't know the specifics of that chip yet. On Intel's side of things, its i9 chips have been overclocked to 5GHz and beyond, so core count isn't that much of an obstacle to high frequencies there either, the stock clocks are lower mainly to keep thermals and power draw within sane limits for 14nm, stock motherboards and typical cooling.

We also know that EPYC2 is launching first on 7nm, which means AMD will have a fair volume of EPYC2 rejects. AMD may be in better shape now than it was two years ago but I doubt it can afford to chuck still usable dies in the bin. If there is a 64T128T EPYC2, there will likely be 16C32T Ryzen CPUs made from functional dies that didn't make the EPYC2 cut to monetize those dies, which is basically what Ryzen 1xxx was.

As for being "mainstream", if AMD launched a 16C32T CPU on AM4, it'd be priced somewhat in line with equivalent ThreadRipper (minus two DDR4 channels and 44 PCIe lanes) and at something like $600, I wouldn't call that mainstream. More like HEDT processing power on mainstream socket for those who need the raw CPU-power but neither the memory size, bandwidth or IO.

None of the above excludes the possibility of AMD having separate "low core count" and "high core count" dies to accommodate different total core count ranges across the product stack (ex.: 4-8/12-16 Ryzen, 8-16/24-32 TR, 16-32/48-64 EPYC) this time around.

 

[feelinfroggy777]

The i5 9600k is the only chip here that really stands out as a possible purchase for me.

Performance per dollar still means something to some of us, and this would help me eek out a few more FPS @1080p on my aging gtx 980.

I like the specs on the i7 9700K, especially for gaming, but that price is just insane.

Yeah, depends how you look at it. When I built my last PC way back in late 2012, I decided to "splurge" the "insane" amount for the i7-3770K, with the net result of that extra couple hundred $$ being that my CPU lasted me over SIX YEARS. Amortizing that roughly $200 worth of extra performance over six years and thousands of hours of use....well, it seems like a BARGAIN now.

Even today I would probably do better by upgrading my GTX 980 instead of getting a new CPU, but I told myself that when the per-core performance improvement reached 50% then I would upgrade...and only today, with the latest 9th gen processors, is that finally the case. Barely.

A 1080ti will give you twice as many fps in games than your 980. A 9900k wont give you twice as many fps in games than your 3770. You can get a used 1080ti for the same price as a 9900k. That does not even cover the cost of a mobo and RAM.

 

[hixbot]

I'm not sure the 9700k is an upgrade from the 8700k. Sure it has more cores, but it has less threads.

And the price has gone up... I thought performance per dollar is supposed to improve?

 

[bit_user]

I told myself that when the per-core performance improvement reached 50% then I would upgrade...and only today, with the latest 9th gen processors, is that finally the case. Barely.

A 1080ti will give you twice as many fps in games than your 980.

This.

I told myself I'd upgrade my GPU when I can get one with twice the horsepower. A RTX 2080 Ti seems like a near doubling of my GTX 980 Ti. So, now I just have to wait for the price to come down to earth (which probably means waiting until the RTX 3000 series launches).

 

[InvalidError]

I thought performance per dollar is supposed to improve?

Performance per dollar is improving, just nowhere near as fast as it used to. Until about 10 years ago, CPUs were getting significantly faster and cheaper every year. Since the Core i-series though, Intel has been sticking with 5-10% better performance between most generations at any given price point.

At least bang-per-buck buyers have the Ryzen option. AMD was quite aggressive with first-gen Ryzen's launch and we can always hope for a repeat with Ryzen 3 to put more pressure on Intel across the board.

 

[jewie27]

What do you guys recommend for 1080P video editing and gaming? I'm using Cyberlink PowerDirector software and planning to play Battlefield 5/Division 2/etc... 9700K or 9900K with Hyperthreading?

 

[spdragoo]

i9 9900k = 8c/16t ==> 488$
r7 2700 = 8c/16t ==> 299$

Is the i9 9900k performance increase worth 60% increase in price over r7 2700?
Only time and benchmarks will tell.

Just as a comparison from PCPartPicker:

i9-9900K, cheapest compatible ATX motherboard, & a $50 cooler, 16GB of RAM, 650W PSU, & a $56 case: $1,046USD stand-alone, $1,426USD with a GTX 1070

R7 2700X, MSI B450 Tomahawk ATX motherboard, same cooler, same RAM, same PSU, same case: $1,006USD with a GTX 1060 (6GB), $1,426USD with a GTX 1080TI

At 1440p, or especially 4K, which do you think is going to be better...the system with the GTX 1070, or the system with the GTX 1080TI?

 

[mlee 2500]

The i5 9600k is the only chip here that really stands out as a possible purchase for me.

Performance per dollar still means something to some of us, and this would help me eek out a few more FPS @1080p on my aging gtx 980.

I like the specs on the i7 9700K, especially for gaming, but that price is just insane.

Yeah, depends how you look at it. When I built my last PC way back in late 2012, I decided to "splurge" the "insane" amount for the i7-3770K, with the net result of that extra couple hundred $$ being that my CPU lasted me over SIX YEARS. Amortizing that roughly $200 worth of extra performance over six years and thousands of hours of use....well, it seems like a BARGAIN now.

Even today I would probably do better by upgrading my GTX 980 instead of getting a new CPU, but I told myself that when the per-core performance improvement reached 50% then I would upgrade...and only today, with the latest 9th gen processors, is that finally the case. Barely.

A 1080ti will give you twice as many fps in games than your 980. A 9900k wont give you twice as many fps in games than your 3770. You can get a used 1080ti for the same price as a 9900k. That does not even cover the cost of a mobo and RAM.

Yeah, that's kind of exactly my point...for six years I was better off upgrading my GPU instead of my six year old PC. Still would be in fact, if all I could afford was one one or the other. Fortunately I can do both, and I think we're sufficiently now into territory where a six year old PC will tangibly throttle the performance of the latest graphics cards. Or maybe that's just something I tell myself so I have an excuse to build a new PC. Who knows?

 

[mlee 2500]

I told myself that when the per-core performance improvement reached 50% then I would upgrade...and only today, with the latest 9th gen processors, is that finally the case. Barely.

A 1080ti will give you twice as many fps in games than your 980.

This.

I told myself I'd upgrade my GPU when I can get one with twice the horsepower. A RTX 2080 Ti seems like a near doubling of my GTX 980 Ti. So, now I just have to wait for the price to come down to earth (which probably means waiting until the RTX 3000 series launches).

Do you play first person shooters, bit_user? If so then you're probably right!

Nice thing about being more strategy-game centric (Warhammer, XCOM) is that there's allot less need or value in having a bleeding edge GPU. In fact, I usually buy higher end and then skip two generations.

 

[doitlikePlay]

Do you think they will release a 65W i7 9700?

 

[TCA_ChinChin]

i9 9900k = 8c/16t ==> 488$
r7 2700 = 8c/16t ==> 299$

Is the i9 9900k performance increase worth 60% increase in price over r7 2700?
Only time and benchmarks will tell.

Just as a comparison from PCPartPicker:

i9-9900K, cheapest compatible ATX motherboard, & a $50 cooler, 16GB of RAM, 650W PSU, & a $56 case: $1,046USD stand-alone, $1,426USD with a GTX 1070

R7 2700X, MSI B450 Tomahawk ATX motherboard, same cooler, same RAM, same PSU, same case: $1,006USD with a GTX 1060 (6GB), $1,426USD with a GTX 1080TI

At 1440p, or especially 4K, which do you think is going to be better...the system with the GTX 1070, or the system with the GTX 1080TI?

Well i guess when you put it that way, the answer seems pretty obvious 😀

 

[spdragoo]

Oh, come on, it's not like we should be applying "logic" to any of these discussions, right? LOL

 

[mlee 2500]

Do you think they will release a 65W i7 9700?

Doesn't look like it, though that's something that people who put a bit more of a premium on low noise and low power rather than overclocking would like to see. Who knows though? They could always make that decision later depending on a variety of factors (sales, yields, etc)..

 

[Tanyac]

Somewhere I predicted that AU retailers would sell this for around $800 AUD ($584 USD).
I'm must be slipping. Prices are released and it's actually at least $859 AUD ($627 USD).