Speculation: AMD's response to Intel's 8-core i9-9900K

[epsilon84]

I still don't understand why people talk about power consumption or perf/watt if we're looking at current CPUs that aren't either low-power/mobile or server/render farms.

Any content creator that invests in either a high end "regular" platform 9900k/2700x OR an HEDT x299/x399 platform will generally not be as concerned with power usage compared to pure performance. These businesses (or people) buy high end chips for the performance. Saving time saves more money than saving power.

So again; why the talk about power consumption relative to performance?

I too find this somewhat odd - it's like people arguing over fuel efficiency between performance cars... it's definitely not the most important metric for 'halo' products that target outright performance over maximum efficiency.

The difference in actual power usage between a fully loaded 9900K and 2700X (or TR/SKL-X) really is quite negligible in the grand scheme of things, especially since the vast majority of people don't run their CPUs at full bore 100% load the majority of the time.

I think it all comes down to the fact that people like arguing for the sake of arguing and to try to find 'weak points' in their 'non preferred' brand.

Ultimately I think the 9900K is a great piece of technology let down by an inflated price and lack of availability. Sure, its power hungry, but thats because it has been pushed to its limits in terms of frequency, and its hard to argue that the extra power is being wasted when it actually has competitive performance/watt to the 2700X.

It hangs well enough with the 1920X/2920X (and 7900X of course) that its legitimately an alternative to those HEDT chips. It's not the outright best performer from that bunch of course, especially in total MT throughput, but it represents an ideal mix of strong ST and MT performance, along with market leading gaming performance. Now if only actual prices would come down towards the MSRP of $488 and you could actually buy the damn CPU, things will look a lot better for the 9900K.

 

[PeterScott]

I think it all comes down to the fact that people like arguing for the sake of arguing and to try to find 'weak points' in their 'non preferred' brand.

It looks like 100% this to me.

9900K is an AMAZING desktop CPU. It dominates gaming and low thread count workloads and is also within spitting distance of Intel/AMD 10-12 core workstation CPUs, even on many embarassingly parallel loads.

It has one significant negative, and that is price and availability right now. No need to bend over backwards coming up with sketchy alternative reasons.

 

[mattiasnyc]

It's not about power saving, it's about power as a normalization factor.

If saving time is more valuable than energy, why bench the 7820X @ 140W versus 9900K @ 165W+? Why not let both CPUs work with the same power limits, especially considering they share the same pedigree?

Because it's a matter of what customers do or don't do. So the normalizing factor is either you OC or you don't, when testing. And then either that is a realistic scenario for the customers in question or it isn't.

My bet is that more customers in the professional content creation space will leave the 7820x at stock than the average 9900, enthusiast user. In other words, for a certain segment of the market it actually does make sense to look at stock performance, not OC, because that's what customers will end up using.

I absolutely understand your point from a purely technical standpoint btw, my only objection is what the practical application is of it.

 

[scannall]

It looks like 100% this to me.

9900K is an AMAZING desktop CPU. It dominates gaming and low thread count workloads and is also within spitting distance of Intel/AMD 10-12 core workstation CPUs, even on many embarassingly parallel loads.

It has one significant negative, and that is price and availability right now. No need to bend over backwards coming up with sketchy alternative reasons.

As far as being a workstation CPU, there are two other points it's short on. No ECC memory, and only 16 PCIe lanes.

Other than that, it's a great gaming CPU. I think though an 8086K would be just as good at pure gaming loads.

 

[Charlie22911]

As far as being a workstation CPU, there are two other points it's short on. No ECC memory, and only 16 PCIe lanes.

Other than that, it's a great gaming CPU. I think though an 8086K would be just as good at pure gaming loads.

What blows my mind is that intel does support ECC on the core i3 parts. Their mindless segmentation is absurd.

https://ark.intel.com/products/126688/Intel-Core-i3-8100-Processor-6M-Cache-3-60-GHz-

 

[PeterScott]

As far as being a workstation CPU, there are two other points it's short on. No ECC memory, and only 16 PCIe lanes.

Other than that, it's a great gaming CPU. I think though an 8086K would be just as good at pure gaming loads.

Ultimately this is a premium home enthusiast CPU. It doesn't need ECC or more PCIe lanes.

It does extremely well at a lot more than just gaming. A while back I polled what people need more cores for. The Top 3 by far were Gaming, Video Encoding, and Virtualization, though virtualization is somewhat less of an issue as it divides resources but doesn't use them. It's still the actual software you run that uses them.

So Gaming and Video encoding are the top use cases, and 9900K crushes both of them, and does very well everything else.

People love to go on and on about rendering with breathless exclamations about Cinebench performance, but the reality is that almost no one actually does rendering. While the most mentioned benchmark, it was actually the least cited use case in the polling.

9900K is an Amazing desktop CPU. It dominates most of what people do and is spitting distance in everything else:
https://www.anandtech.com/show/13400/intel-9th-gen-core-i9-9900k-i7-9700k-i5-9600k-review

But I am sure you can search through the benchmarks and find some esoteric case where it gets beaten by 10-12 core Workstation CPU and make that your rallying cry that this is something important you do all the time.

But reality is for people with a wide range of typical use cases, the 9900k is dominant.

Repeating what I said before. It's really only price/availability that holds this great CPU back.

 

[PotatoWithEarsOnSide]

You are right; I use mine as an oven.

 

[krumme]

It's nice to see 8c as the new norm.

 

[UsandThem]

But reality is for people with a wide range of typical use cases, the 9900k is dominant.

Repeating what I said before. It's really only price/availability that holds this great CPU back.

That's the reason I drive a Toyota instead of a GT-R.

it's a great performing chip, but it just doesn't justify its price (at least to me) compared to CPUs like the 8700k or 2700X. For people who must have the latest/greatest, I get it. But for those of who use $/performance for tech purchases, it's a tough sell (even more so with it being over MSRP).

 

[PeterScott]

That's the reason I drive a Toyota instead of a GT-R.

it's a great performing chip, but it just doesn't justify its price (at least to me) compared to CPUs like the 8700k or 2700X. For people who must have the latest/greatest, I get it. But for those of who use $/performance for tech purchases, it's a tough sell (even more so with it being over MSRP).

The market has plenty of people in both groups. There will probably be a lot of 2080Ti/9900K systems in the hands of people who just want the best and are willing to pay a bit more.

 

[Vattila]

Ok, guys and gals, let's circle this back to the thread topic: AMD's response.

There was no 2800X, and I now do not expect there to be. I think Lisa Su's focus is all on the 7nm products. So when will we have 7nm Ryzen 3000?

With the rumoured 8+1 design for "Rome", the server die isn't as cheaply reusable in the mainstream as before (it will at least require a System Controller die of its own, as well as more expensive MCM packaging). So I am starting to doubt we will even see a chip like that. High-volume mainstream requires a cheap APU to match Intel's offerings, not an expensive solution mostly applicable to the relatively small gaming market.

So here is an alternative hypothesis to consider:

EPYC 2 "Rome" arrives in 2019-Q1 in the cloud segment, with OEM servers for enterprise appearing in 2019-H2. Ryzen Threadripper 3000, reusing the server design, arrives early 2019-Q2 (April). The 7nm APU, which depends on "Navi", comes 2019-Q3 or even later in 2019-Q4/2020-H1. This 7nm APU, an 8-core chip, will be the basis for the Ryzen 3000 series for mainstream desktop and high-end notebook.

If 7nm Zen 2 is as good as I expect (beating 14nm Skylake/CFL on all metrics), Threadripper 3000 will become the leading gaming and enthusiast platform. It will have IPC and boost clocks high enough to win in gaming, while providing ample bandwidth, lots of connectivity and as many cores as you want for your other workloads (8 to 32, depending on model).

In the meantime, AMD's response to i9-9900K will indeed only be pricing, bringing down the cost of entry to the HEDT platform.

PS. Can't wait for the "Next Horizon" event on Tuesday!

 

[coercitiv]

If 7nm Zen 2 is as good as I expect (beating 14nm Skylake/CFL on all metrics), Threadripper 3000 will become the leading gaming and enthusiast platform.

How exactly are we going from a lean and fast Ryzen 3000 to a TR 3000 being a leading gaming chip?

Threadripper will be about cores & throughput, not latency.

 

[Vattila]

Threadripper will be about cores & throughput, not latency.

Eh. You're too pessimistic. More cache, lower latency, more IPC and higher frequencies. Threadripper 3000 should be good, even for gaming.

PS. Cool fact: Threadripper 2950X has a max boost clock at 4.4 GHz. Ryzen 2700X's max boost is 4.3 GHz. Threadripper with its high boost, many threads and twice the cache already does pretty well in some games.

https://www.youtube.com/watch?v=Ch41ZLBczOQ

 

[maddie]

Ok, guys and gals, let's circle this back to the thread topic: AMD's response.

There was no 2800X, and I now do not expect there to be. I think Lisa Su's focus is all on the 7nm products. So when will we have 7nm Ryzen 3000?

With the rumoured 8+1 design for "Rome", the server die isn't as cheaply reusable in the mainstream as before (it will at least require a System Controller die of its own, as well as more expensive MCM packaging). So I am starting to doubt we will even see a chip like that. High-volume mainstream requires a cheap APU to match Intel's offerings, not an expensive solution mostly applicable to the relatively small gaming market.

So here is an alternative hypothesis to consider:

EPYC 2 "Rome" arrives in 2019-Q1 in the cloud segment, with OEM servers for enterprise appearing in 2019-H2. Ryzen Threadripper 3000, reusing the server design, arrives early 2019-Q2 (April). The 7nm APU, which depends on "Navi", comes 2019-Q3 or even later in 2019-Q4/2020-H1. This 7nm APU, an 8-core chip, will be the basis for the Ryzen 3000 series for mainstream desktop and high-end notebook.

If 7nm Zen 2 is as good as I expect (beating 14nm Skylake/CFL on all metrics), Threadripper 3000 will become the leading gaming and enthusiast platform. It will have IPC and boost clocks high enough to win in gaming, while providing ample bandwidth, lots of connectivity and as many cores as you want for your other workloads (8 to 32, depending on model).

In the meantime, AMD's response to i9-9900K will indeed only be pricing, bringing down the cost of entry to the HEDT platform.

PS. Can't wait for the "Next Horizon" event on Tuesday!

You wrote:
"The 7nm APU, which depends on "Navi", comes 2019-Q3 or even later in 2019-Q4/2020-H1. This 7nm APU, an 8-core chip, will be the basis for the Ryzen 3000 series for mainstream desktop and high-end notebook."

Are you claiming that desktop Ryzen 3000 is based on this and arrives in the stated timeframe?

If yes, some serious bipolar stuff here in this thread.

 

[LTC8K6]

What blows my mind is that intel does support ECC on the core i3 parts. Their mindless segmentation is absurd.

https://ark.intel.com/products/126688/Intel-Core-i3-8100-Processor-6M-Cache-3-60-GHz-

https://ark.intel.com/Search/Featur...ocessors&ECCMemory=true&MarketSegment=Desktop

Nothing new at all about it, though. Been that way for a long time. Presumably the low end chips are used in commercial applications more often. Things like POS systems and signage maybe.

Note that many of the chips, even older ones, are not discontinued.

 

[Vattila]

Are you claiming that desktop Ryzen 3000 is based on this and arrives in the stated timeframe?

I am not claiming. I am just throwing it out there to get some thoughts.

I simply have a hard time believing AMD will go for MCM in the mainstream, due to complexity and cost. On the other hand, a separate monolithic 7nm die without iGPU seems even more unlikely, due to cost. It wouldn't match Intel's feature set, and simply has too narrow focus to be worth it.

Actually, Threadripper 3000 in the high-end for gamers and enthusiasts running discrete graphics cards, plus an 8-core Ryzen 3000 APU in the mainstream, competing against Intel's 9th generation, would seem to cover the market nicely.

 

[epsilon84]

Eh. You're too pessimistic. More cache, lower latency, more IPC and higher frequencies. Threadripper 3000 should be good, even for gaming.

PS. Cool fact: Threadripper 2950X has a max boost clock at 4.4 GHz. Ryzen 2700X's max boost is 4.3 GHz. Threadripper with its high boost, many threads and twice the cache already does pretty well in some games.

https://www.youtube.com/watch?v=Ch41ZLBczOQ

Unless Threadripper drops to the $300 mark like Ryzen 7, no chance it will be the 'gaming platform of choice' for enthusiasts, especially considering mobos also cost $300 for X399 unlike AM4 which you can get a range of decent B450 mobos for $80 - $100.

The same reason the 9900K has been ridiculed (price/performance vs existing chips) will be even worse for TR 3000 from a gaming perspective.

I'm sure most gamers will choose a $150 Ryzen 5 (like the 2600, or 3600) paired with a sub $100 mobo rather than opt for a $650 CPU + $300 mobo. The ones that go for a 9900K may see a reason to opt for TR depending on how they view gaming vs productivity performance, but I highly doubt many will choose TR as a gaming only platform, the costs of entry are simply too high. You can pay 25% the cost with Ryzen and get similar gaming performance in the vasy majority of titles compared to TR.

 

[Markfw]

Ultimately this is a premium home enthusiast CPU. It doesn't need ECC or more PCIe lanes.

It does extremely well at a lot more than just gaming. A while back I polled what people need more cores for. The Top 3 by far were Gaming, Video Encoding, and Virtualization, though virtualization is somewhat less of an issue as it divides resources but doesn't use them. It's still the actual software you run that uses them.

So Gaming and Video encoding are the top use cases, and 9900K crushes both of them, and does very well everything else.

People love to go on and on about rendering with breathless exclamations about Cinebench performance, but the reality is that almost no one actually does rendering. While the most mentioned benchmark, it was actually the least cited use case in the polling.

9900K is an Amazing desktop CPU. It dominates most of what people do and is spitting distance in everything else:
https://www.anandtech.com/show/13400/intel-9th-gen-core-i9-9900k-i7-9700k-i5-9600k-review

But I am sure you can search through the benchmarks and find some esoteric case where it gets beaten by 10-12 core Workstation CPU and make that your rallying cry that this is something important you do all the time.

But reality is for people with a wide range of typical use cases, the 9900k is dominant.

Repeating what I said before. It's really only price/availability that holds this great CPU back.

For what I use it for, I can buy a 16 core CPU for the price they want, and murder it in performance. But I am sure your rant is for your favorite company, so no sense using facts here....

 

[LTC8K6]

For what I use it for, I can buy a 16 core CPU for the price they want, and murder it in performance. But I am sure your rant is for your favorite company, so no sense using facts here....

Well, you can't buy a 9900K at all.

I suspect Intel is fixing them.

 

[DrMrLordX]

With the rumoured 8+1 design for "Rome", the server die isn't as cheaply reusable in the mainstream as before (it will at least require a System Controller die of its own, as well as more expensive MCM packaging). So I am starting to doubt we will even see a chip like that.

Why? The controller chip would be small, simple, and cheap to produce on GF 12nm. Intel was able to use MCM on Clarkdale years ago at low cost. I see no reason why AMD would shy away from chiplet + controller for something like the 3700x unless latency to the system controller was just that bad (no reason why it would have to be, though).

High-volume mainstream requires a cheap APU to match Intel's offerings, not an expensive solution mostly applicable to the relatively small gaming market.

Lisa Su has said from the beginning of her tenure that AMD has no intention of being the "budget" option. At the present, they are not (good luck buying cheap "mainstream" AMD systems based off Zen/Zen+ tech), and that is unlikely to change with ARM offerings trying to push up into that space to challenge Intel. AMD's own roadmaps place a secondary emphasis on APUs. Matisse will be first to 7nm on the desktop, before any APU products.

EPYC 2 "Rome" arrives in 2019-Q1 in the cloud segment, with OEM servers for enterprise appearing in 2019-H2. Ryzen Threadripper 3000, reusing the server design, arrives early 2019-Q2 (April). The 7nm APU, which depends on "Navi", comes 2019-Q3 or even later in 2019-Q4/2020-H1. This 7nm APU, an 8-core chip, will be the basis for the Ryzen 3000 series for mainstream desktop and high-end notebook.

No. Matisse != Threadripper, and it is already slated for Q2 2019. April, if all goes well.

 

[PeterScott]

For what I use it for, I can buy a 16 core CPU for the price they want, and murder it in performance. But I am sure your rant is for your favorite company, so no sense using facts here....

I really don't think you should be throwing around favorite company bias accusations.

The new Intel CPU beats alternate 10-12 core CPUs overall, and costs less, so you leap to more niche use cases with even more expensive 16 core CPU ($900) and motherboard ($300+).

I already acknowledge, that Ryzen is the best perf/$ on the desktop, and Threadripper is the king of high core count rendering tasks.

But point out one niche where the Intel processor succeeds and I am the biased one, that you need to rescue us from.

To me, bias looks a lot more like someone who thinks only one company is the right choice everywhere.

 

[TheGiant]

the problem with 9900K is that..it is too good for the more than average user.
Even enthusiasts will have problems using 16 Threads at @4,7-5GHz.
So it is and allround CPU.
AMD has an allround CPU too, 2700X and even better 2950X (it has higher boost than 2700X).
Response for me from AMD 6C and 8C with performance of 4,5GHz Skylake core is enough, while reducing power which is a problem with ryzen oced above 4GHz generally
9900K when clocked at 4,5-4,7 GHz all core can be quite efficient, at 5GHz the power hits the sky

We have the best allrounder here for a while. The core i7-6950X. Oced to 4.3+GHz will last like next 2 years with no weak spot (except the price lol ). Unbelievable.

 

[Markfw]

I really don't think you should be throwing around favorite company bias accusations.

The new Intel CPU beats alternate 10-12 core CPUs overall, and costs less, so you leap to more niche use cases with even more expensive 16 core CPU ($900) and motherboard ($300+).

I already acknowledge, that Ryzen is the best perf/$ on the desktop, and Threadripper is the king of high core count rendering tasks.

But point out one niche where the Intel processor succeeds and I am the biased one, that you need to rescue us from.

To me, bias looks a lot more like someone who thinks only one company is the right choice everywhere.

No $900 CPU , $680, not much more than the $600 (if you can find one) 9900k. Get your facts straight.

Oh, and I have said many tomes, the 8700k and now the 9900k is the gaming king.

 

[PeterScott]

No $900 CPU , $680, not much more than the $600 (if you can find one) 9900k. Get your facts straight.

I was looking at the price of the 2950x and it was about $900 USD.

I have been checking my local B&M for the past week and they had stock for 9900K at my favored location(though out now), for $669 CDN, they still have stock on the 9700K for $519 CDN.

Locally at my preferred B&M:
9900K is $669 CDN
2920x is $869 CDN
2950x is $1199 CDN.

So a 9900K is $200 cheaper than a 2920x while essentially matching it, and exceeding it at my most common use cases. The 2950x is nearly $500 more.

 

[lixlax]

In my opinion it makes very little sense for AMD to release a 100-200MHz higher clocked version of 2700X. It would close the perfomance gap just a little bit and would probably have much higher power consumtion as well. 2700X already takes 12nm/Zen+ close to the max limits as is.
Hopefully the Ryzen 3000/Zen 2 will be realeased around 6 months timeframe. My estimate is 20-25% speed up vs Zen+ on the same core counts, ofcourse including pure single thread perfomance (whether or not we'll see CPU's with more than 8 cores at launch- I have no idea).