Intel Core i9-10900K 10-core Processor and Z490 Chipset Arrive April 2020

[GoldenLag]

11 minutes ago, Tony Tony Chopper said:

 

Even more reason to glue 2 cpu's on 1 chip then.

issue.

 

there is this thing called numa nodes. Zen 2 isnt exactly great when it comes to latency, but its pretty fast compared to numa solutions in dualsocket etc. 

 

and they are gluing together 2 chips. just not in the consumerspace. 

 

that 56 core chip? thats 2 28 core chips on the same package. 

[RejZoR]

2 minutes ago, GoldenLag said:

issue.

 

there is this thing called numa nodes. Zen 2 isnt exactly great when it comes to latency, but its pretty fast compared to numa solutions in dualsocket etc. 

 

and they are gluing together 2 chips. just not in the consumerspace. 

 

that 56 core chip? thats 2 28 core chips on the same package. 

Zen1 had problems. Zen2 has this done in CPU logic much better so it automatically picks best option, whether it's staying within one chip or goes into multichip operation. The IO is probably more aware of what chips are processing to pick the right one and sack latency. So, its probably staying within one physical chip for as long as possible and goes to split mode when you ask for more. And latency really isn't as much of an issue given you basically double the core count on the cheap. Something you can't do with big monolithic chips which are always more expensive.

[Mira Yurizaki]

28 minutes ago, Tony Tony Chopper said:

Intel should just do what AMD did and solder 2 cpu chips onto 1 cpu like they started doing with thread ripper if cant beat them copy them.

You should look up the Pentium D Smithfield sometime.

[Letgomyleghoe]

On 12/11/2019 at 8:51 AM, GDRRiley said:

Unless they can pull some big IPC (doubt) or clock speed (intels going to get the heater name) it won't do well against ryzen 2 let alone zen 3 parts.

i was about to say wheres the 12 c0res to keep up with ryzen 9's, they have the clock speed advantage but if ryzen figures out how to get some good silicon and keep it away from them epyc fiends 

[Sauron]

30 minutes ago, pas008 said:

mainstream really doesnt need it though

These are 500$+ chips, they aren't mainstream by any stretch.

[pas008]

On 12/11/2019 at 10:51 AM, GDRRiley said:

Unless they can pull some big IPC (doubt) or clock speed (intels going to get the heater name) it won't do well against ryzen 2 let alone zen 3 parts.

not sure what you mean here

https://www.techspot.com/article/1876-4ghz-ryzen-3rd-gen-vs-core-i9/

intels 14nm looks to be hold its own against tsmc 7nm and chiplet design

with it being held back

[pas008]

1 minute ago, Sauron said:

These are 500$+ chips, they aren't mainstream by any stretch.

but its on mainstream platform

i will agree that they should just combine them all

from both companies or all companies

 

 

[Mira Yurizaki]

24 minutes ago, Tony Tony Chopper said:

ain't that chip from 2005 ? anyway what are we disusing anyway?

Intel doesn't need to copy someone when they've done the solution before.

[Drak3]

Just now, Mira Yurizaki said:

Intel doesn't need to copy someone when they've done the solution before.

Nah, Intel’s just copying the brand of glue.

 

/s

[Mira Yurizaki]

3 minutes ago, Drak3 said:

Nah, Intel’s just copying the brand of glue.

 

/s

Indium solder?

[Sauron]

4 minutes ago, pas008 said:

but its on mainstream platform

Well... that doesn't really mean much anymore. If there weren't a different socket for the 2000$ chips would you call them mainstream? These are niche products for enthusiasts or professionals, there's nothing mainstream about them. The people buying this stuff need more cores.

[pas008]

15 minutes ago, Sauron said:

Well... that doesn't really mean much anymore. If there weren't a different socket for the 2000$ chips would you call them mainstream? These are niche products for enthusiasts or professionals, there's nothing mainstream about them. The people buying this stuff need more cores.

exactly so what does doubling cores matter on mainstream platform

 

like i said in previous post

me personally I would take 9900k over 3700x or 3800x but over the 3900x or 3950x i'd be in a pickle because i'd be trading small performance in certain areas to gain more performance in an area I really dont need to utilize all those cores often,

 

but they could offer 4 to 32 cores on one platform and call it a day

[Sauron]

9 minutes ago, pas008 said:

exactly so what does doubling cores matter on mainstream platform

Just because you need more cores doesn't mean you need the extra features of the workstation platform, which is significantly more expensive.

[Zando_]

Just now, Sauron said:

Just because you need more cores doesn't mean you need the extra features of the workstation platform, which is significantly more expensive.

^^^ For some people who just need straight CPU render performance, stuff like the 3950X and the 3900X and maybe the uh... the 3950X again because Intel caps out at 8c, are actually useful. If you're using a single GPU and no other PCIe devices besides an SSD or two, and don't need moar RAM bandwidth, it's much cheaper than HEDT for similar/the same performance (depends what you compare it to). 

[pas008]

2 minutes ago, Sauron said:

Just because you need more cores doesn't mean you need the extra features of the workstation platform, which is significantly more expensive.

what does that have to do with anything that can be different sku or up to mobo companies to include or exclude

[Zando_]

2 minutes ago, pas008 said:

what does that have to do with anything that can be different sku or up to mobo companies to include or exclude

That has to do with how HEDT and mainstream compare, lol. Mainstream usually has way less PCIe lanes, dual channel RAM support, some other smaller stuff missing too. HEDT adds moar PCIe lanes, quad or higher channel RAM support, and usually the option for even higher core counts, which is why it's an option for some workloads but not others. If you don't need HEDT features and you're looking for value, a high core count CPU on a more affordable mainstream platform is a good option. 

[Sauron]

Just now, pas008 said:

what does that have to do with anything that can be different sku or up to mobo companies to include or exclude

I'm sorry, I don't see your point... first of all that's not true because the chipsets themselves are much more expensive and, secondly, why is it a problem for anyone that these chips are available on the cheaper platform? Even among prosumers not that many people need 60+ pcie lanes or quad channel ram. The "mainstream" platform has become feature complete to the point that a lot of professionals who need cpu performance don't need to splurge for a 600$ motherboard. Or at least, they don't need to if they're buying from AMD.

[pas008]

5 minutes ago, Zando Bob said:

That has to do with how HEDT and mainstream compare, lol. Mainstream usually has way less PCIe lanes, dual channel RAM support, some other smaller stuff missing too. HEDT adds moar PCIe lanes, quad or higher channel RAM support, and usually the option for even higher core counts, which is why it's an option for some workloads but not others. If you don't need HEDT features and you're looking for value, a high core count CPU on a more affordable mainstream platform is a good option. 

 

2 minutes ago, Sauron said:

I'm sorry, I don't see your point... first of all that's not true because the chipsets themselves are much more expensive and, secondly, why is it a problem for anyone that these chips are available on the cheaper platform? Even among prosumers not that many people need 60+ pcie lanes or quad channel ram. The "mainstream" platform has become feature complete to the point that a lot of professionals who need cpu performance don't need to splurge for a 600$ motherboard. Or at least, they don't need to if they're buying from AMD.

thats why you can offer different chipsets and mobos for said needs

like they already do in mainstream from low end to highend but extend the highend to hedt

most of cpu lanes come from the cpu itself ?

memory contoller is where?

 

[Zando_]

2 minutes ago, pas008 said:

 

thats why you can offer different chipsets and mobos for said needs

like they already do in mainstream from low end to highend but extend the highend to hedt

most of cpu lanes come from the cpu itself ?

memory contoller is where?

 

What? I'm... confused as to what your point is. Can't extend high end to HEDT, you get all sorts of issues. If you run different sockets it's all split up still and very confusing, if you run the same socket then not only do you have a tiny pentium inside a bigass HEDT shell (which is expensive, driving costs up), now do they all work on all chipsets? On AMD for sure, a TR CPU cannot physically fit in an AM4 socket package, meaning if you wanted to unify the lineup there'd be a 3200G inside a TRX40 socket sized chip, stupidly expensive and annoying to install with low cooler compatability. How are you deciding what chips run what RAM combo and what PCIe slots are active when? Intel's HEDT has been confusing enough with lower PCIe count HEDT chips, let alone trying to blend chips with different RAM channel configs and PCIe lane support. And what if some CPUs support bifurcation and NVMe raid and others don't? How do you differentiate that while keeping a unified setup? If we give all the CPUs all the PCIe lanes and quad channel RAM, the low end will now increase massively in price, no longer being cost effective for the people who need it.

Do you know nothing about chipsets and CPUs and how they work? There's a reason HEDT is usually compartmentalized to it's own socket and chipset/chipsets (Intel HEDT is often cross-compatible with their server platform, IIRC EPYC and TR are not). 

[pas008]

52 minutes ago, Zando Bob said:

What? I'm... confused as to what your point is. Can't extend high end to HEDT, you get all sorts of issues. If you run different sockets it's all split up still and very confusing, if you run the same socket then not only do you have a tiny pentium inside a bigass HEDT shell (which is expensive, driving costs up), now do they all work on all chipsets? On AMD for sure, a TR CPU cannot physically fit in an AM4 socket package, meaning if you wanted to unify the lineup there'd be a 3200G inside a TRX40 socket sized chip, stupidly expensive and annoying to install with low cooler compatability. How are you deciding what chips run what RAM combo and what PCIe slots are active when? Intel's HEDT has been confusing enough with lower PCIe count HEDT chips, let alone trying to blend chips with different RAM channel configs and PCIe lane support. And what if some CPUs support bifurcation and NVMe raid and others don't? How do you differentiate that while keeping a unified setup? If we give all the CPUs all the PCIe lanes and quad channel RAM, the low end will now increase massively in price, no longer being cost effective for the people who need it.

Do you know nothing about chipsets and CPUs and how they work? There's a reason HEDT is usually compartmentalized to it's own socket and chipset/chipsets (Intel HEDT is often cross-compatible with their server platform, IIRC EPYC and TR are not). 

do you realize x299 and z390 are practically the same?lol

most of the other shit is inside the cpu

 

https://www.intel.com/content/www/us/en/products/chipsets/desktop-chipsets/x299.html

 

https://www.intel.com/content/www/us/en/products/chipsets/desktop-chipsets/z390.html

 

[Zando_]

Just now, pas008 said:

do you realize x299 and z390 are practically the same?lol

most of the other shit is inside the cpu

[pas008]

2 minutes ago, Zando Bob said:

point?

not seeing difference on chipsets

 

[MageTank]

4 minutes ago, pas008 said:

do you realize x299 and z390 are practically the same?lol

most of the other shit is inside the cpu

 

https://www.intel.com/content/www/us/en/products/chipsets/desktop-chipsets/x299.html

 

https://www.intel.com/content/www/us/en/products/chipsets/desktop-chipsets/z390.html

 

This is extremely far from true though. We are talking vastly different power delivery (FIVR vs board VRM), dramatically different trace layout due to PCIe lane allocation and DIMM locations, not to mention different wiring to the DIMM slots to handle larger capacity DIMM's due to the sheer difference in memory controllers on the CPU's themselves.

 

What you proposed has already been done on X299 and was deemed a failure. The Kaby-X chips had issues where they would have to disable entire DIMM slots, M.2 slots, PCIe slots and more just because the CPU itself lacked the memory controller and PCIe lanes to drive those board components. If you suggest those features are added across the entire stack, the cost for all consumers will go up.

 

Ignoring that glaring issue, you then run into an issue with EEPROM size limitations and trying to support each microcode/stepping change going forward for each processor, which severely limits future SKU's being used on that platform without removing support for older processor models (something the Kaby-X series suffered from yet again, and happened to Ryzen 1000 series on X570). The easiest solution is to simply use a bigger EEPROM to allow for more data, but again, who eats the cost of that?

 

Lastly, maintaining socket compatibility but using different chipsets will open up a nightmare scenario for retail technical support that already have a bad time explaining to customers that it's not a bright idea to use a 9900KS in a budget H310 motherboard. People will complain that their expensive processor isn't working in their cheap board, or worse, complain that their expensive processor killed their budget chipset board, akin to what the FX 9590 did to cheapo AM3 970 boards. 

 

I share your sentiments that a unified product stack would be easier on consumers, but it cannot be done with the current segmentation strategies in mind. Something would have to fundamentally change to make it feasible, which Intel will not do. Besides, enthusiasts will always pay more to differentiate themselves from the rest of the product stack, even if a price premium is involved.

[Zando_]

8 minutes ago, pas008 said:

point?

not seeing difference on chipsets

 

I really don't know how to address this level of just... blinded lunacy? @MageTank Has a good reasoned reply, and if you want a single lineup you need to fix this too:

LGA 1151 CPU vs LGA 2066 CPU. The difference between AM4 and sTR4/sTRX4 is way fuckin bigger, but we'll stick to Intel here. 

Dies:

7980XE

LGA1151 i7s.

You can't physically fit the HEDT Die underneath the LGA1151 IHS, meaning if you want a single lineup, every single CPU needs to be LGA2066, which is a big expensive package. Even Pentiums will need to go under this, driving up their price by a large amount. Will effect mobos too, there's only a single ITX X299 board because the socket is so large, there isn't even an ITX sTR4 board in existence because that package is even bigger. So you can't get mobos as small anymore, half the stuff on your mobos either will or won't work depending on the CPU, and costs for everyone other than HEDT users (it'll stay the same) will go up massively to pay for physically larger and more complicated CPU packages. 

and that's one facet of CPU/socket/chipset differences and why they're split up. Just one, there's a ton more variables. 

Oh, and here's the comparision between sTRX and AM4:

Single TR4 is almost as big as 4 AM4 chips. Oh and no way that die setup fits in the AM4 package:

meaning every single AMD CPU must use the even more expensive, more annoying, less compatible sTR4 package. 

[pas008]

1 minute ago, MageTank said:

This is extremely far from true though. We are talking vastly different power delivery (FIVR vs board VRM), dramatically different trace layout due to PCIe lane allocation and DIMM locations, not to mention different wiring to the DIMM slots to handle larger capacity DIMM's due to the sheer difference in memory controllers on the CPU's themselves.

 

What you proposed has already been done on X299 and was deemed a failure. The Kaby-X chips had issues where they would have to disable entire DIMM slots, M.2 slots, PCIe slots and more just because the CPU itself lacked the memory controller and PCIe lanes to drive those board components. If you suggest those features are added across the entire stack, the cost for all consumers will go up.

 

Ignoring that glaring issue, you then run into an issue with EEPROM size limitations and trying to support each microcode/stepping change going forward for each processor, which severely limits future SKU's being used on that platform without removing support for older processor models (something the Kaby-X series suffered from yet again, and happened to Ryzen 1000 series on X570). The easiest solution is to simply use a bigger EEPROM to allow for more data, but again, who eats the cost of that?

 

Lastly, maintaining socket compatibility but using different chipsets will open up a nightmare scenario for retail technical support that already have a bad time explaining to customers that it's not a bright idea to use a 9900KS in a budget H310 motherboard. People will complain that their expensive processor isn't working in their cheap board, or worse, complain that their expensive processor killed their budget chipset board, akin to what the FX 9590 did to cheapo AM3 970 boards. 

 

I share your sentiments that a unified product stack would be easier on consumers, but it cannot be done with the current segmentation strategies in mind. Something would have to fundamentally change to make it feasible, which Intel will not do. Besides, enthusiasts will always pay more to differentiate themselves from the rest of the product stack, even if a price premium is involved.

 

be easy for intel combine mainstream with their current hedt their damn sockets they pretty damn close in size too

they'd just need to eliminate 115x or lga200 or whatever and go with 2066 or something

then offer different chipsets from there

 

 

its not like amd with am4 and tr sockets

 

fyi you can burn out many boards with 3900x and 9900k