Ryzen 4000 and X670 scheduled for late 2020

[Intrafinesse]

On 1/16/2020 at 4:18 PM, Windows95 said:

I also want to stretch my next build for 10 years except GPU, but having that fan there for 10 years spinning makes me nervous..

My CPU fan is an Arctic Freeze 34 eSports Duo, with a 10 year warranty.

I also plan on looking at the temperature every so often.

[Windows95]

4 minutes ago, Intrafinesse said:

My CPU fan is an Arctic Freeze 34 eSports Duo, with a 10 year warranty.

I also plan on looking at the temperature every so often.

i just hope the fan on the mobo doesnt fuck up, the rest i guess i can trust

[Kisai]

On 1/1/2020 at 4:07 PM, leadeater said:

Same, I have a very strong urge to upgrade but I'm on a even older platform, Ivy Bridge-E.

I'm still on my Haswell system.

 

As I've mentioned before, if you're still using DDR3, it's time to upgrade. But I can totally justify waiting a bit longer as there's not urgent reason to upgrade a desktop since you can just upgrade the GPU to squeeze out more gaming, and max out the ram if you haven't. I originally bought 32GB and I replaced the GPU to a GTX 1080 from a 760 when I got a 4K screen. 

 

Unlike the jump from DDR2 to DDR3, the DDR3 to DDR4 wasn't an immediate bandwidth increase so there's still some time.

[leadeater]

46 minutes ago, Kisai said:

As I've mentioned before, if you're still using DDR3, it's time to upgrade

 

46 minutes ago, Kisai said:

Unlike the jump from DDR2 to DDR3, the DDR3 to DDR4 wasn't an immediate bandwidth increase so there's still some time.

 

Ivy-E is quad channel too so it's not like my 6 cores are wanting more bandwidth, they are just slow to today's standards but like you said just a GPU upgrade is more than enough.

[Sir0Tek]

late 2020... damn, I had hopes it would be pre-summer.... to get some 1st and 2nd gen Ryzens almost for free...

[Division]

25 minutes ago, Sir0Tek said:

late 2020... damn, I had hopes it would be pre-summer.... to get some 1st and 2nd gen Ryzens almost for free...

I just got B350 mobo with 16gb of ram and a R7 1700 for 130e. Its happening! 

[JediFragger]

51 minutes ago, valdyrgramr said:

I'm holding onto my VII for now because I'm not purely focusing on gaming, but pro work too which sadly Navi seems to suck at.  When Nvidia or AMD releases a curb stomp to my VII in blender I might upgrade.   However, I don't really plan to any time soon.  I might get a 3600 as a toy CPU in a few weeks then upgrade later this year to either a 3900x or whatever 4000 series CPU in the 500-600 USD range.

Tbf the VII is still a great card for both gaming and pro work, can see it lasting you a while yet

[AngryBeaver]

On 12/8/2019 at 8:53 PM, TrigrH said:

If Zen 3 can actually get some clock speed, its GG.

There is much more to a powerful chip than just clock speed.

 

If for example AMD gets that 10-15% IPC improvement they keep talking about then they don't need any higher clock speeds.

 

Remember a cores power comes down to frequency X IPC. 

 

Just random example.

 

If you have a chip that is 5ghz, but with 1 ipc. That is 5ghz of computer power

 

If you have a 4.5ghz chip with 1.2 ipc then that is 5.4ghz worth of compute power.

 

So there is more to ST performance than just clocks.

 

The ST performance wouldn't even be as relevant if not for DX draw calls still being very much single threaded. When that changes the entire game changes.

[TrigrH]

4 hours ago, AngryBeaver said:

There is much more to a powerful chip than just clock speed.

Thats true and I know how IPC works in relation to performance however some IPC gains do not always benefit all applications. (Unlike clock speed)

 

AMD already has the IPC advantage over current 14nm intel, they just need better clock speed to close the gap, especially in gaming. 

[Blue4130]

On 1/19/2020 at 11:23 AM, Windows95 said:

i just hope the fan on the mobo doesnt fuck up, the rest i guess i can trust

I really don't get the fear of these fans. Back in the day they were extremely common on Motherboards, they rarely died then. They are currently in every 1u device that powers to world. Switches, router, servers... Guess what? They rarely die. Fan reliability is really high, confidence on the other hand... 

[AngryBeaver]

1 hour ago, TrigrH said:

Thats true and I know how IPC works in relation to performance however some IPC gains do not always benefit all applications. (Unlike clock speed)

 

AMD already has the IPC advantage over current 14nm intel, they just need better clock speed to close the gap, especially in gaming. 

The only real area where AMD is not currently leading is gaming. Which as I said previously is an issue with draw calls being single threaded. 

 

In that scenario IPC does play a big part. So if I had to guess we will see a decent IPC improvement and probably another 200mhz on all core clocks. If they can do both of those things then the gaming side of things should start being in AMDs favor.

 

Intel has basically backed themselves into a frequency corner where 5ghz is keeping them competitive in gaming. They need to get 10nm or even a smaller process running on their mainstream platforms quick or they are going to be hard pressed to hold market share.

 

Even then I'm not sure where they will be on their own IPC inprovements until 10nm+. This isn't the first time AMD caught them with their pants down and if this is even remotely similar we are going to have a good time as consumers.

[Reytime]

What about Ryzen 4000 notebooks? When are those supposed to start rolling out?

[Curufinwe_wins]

I will admit... I'm in the same boat others have mentioned as a haswell-e 980ti sli guy. Like yeah grear gains have been made... but not gamechangingly big ones yet and basically I'll need to buy a whole new PC (I probably wouldnt even bother bringing over storage as the **only thing that can leave without making the old pc literally useless** now that SSD prices are down a lot again.

 

I also want to trust AMD with my next platform CPU (desktop dyi, who gives a crap about laptops as long as they have a warranty... but I don't think I'd ever do it right after a launch, way too many people I know personally have ended up burned by it.

 

Well keep pushing AMD, and I'm sure we will meet up again soon! (We have already since building my personal pc, just not for me again)

[Curufinwe_wins]

3 hours ago, AngryBeaver said:

The only real area where AMD is not currently leading is gaming. Which as I said previously is an issue with draw calls being single threaded. 

 

In that scenario IPC does play a big part. So if I had to guess we will see a decent IPC improvement and probably another 200mhz on all core clocks. If they can do both of those things then the gaming side of things should start being in AMDs favor.

 

Intel has basically backed themselves into a frequency corner where 5ghz is keeping them competitive in gaming. They need to get 10nm or even a smaller process running on their mainstream platforms quick or they are going to be hard pressed to hold market share.

 

Even then I'm not sure where they will be on their own IPC inprovements until 10nm+. This isn't the first time AMD caught them with their pants down and if this is even remotely similar we are going to have a good time as consumers.

I guess we will see. Intel tried P4 for a while against AMD and it was a flying cluster barn fire, and then boom core just obliterated any competition. I'm not saying I expect intel to do it again (advances are way harder now than back then so...), but I do find it significantly more likely than I found AMD actually delivering on both Zen and Zen2 and they did that with spades.

[RejZoR]

9 hours ago, AngryBeaver said:

The only real area where AMD is not currently leading is gaming. Which as I said previously is an issue with draw calls being single threaded. 

 

In that scenario IPC does play a big part. So if I had to guess we will see a decent IPC improvement and probably another 200mhz on all core clocks. If they can do both of those things then the gaming side of things should start being in AMDs favor.

 

Intel has basically backed themselves into a frequency corner where 5ghz is keeping them competitive in gaming. They need to get 10nm or even a smaller process running on their mainstream platforms quick or they are going to be hard pressed to hold market share.

 

Even then I'm not sure where they will be on their own IPC inprovements until 10nm+. This isn't the first time AMD caught them with their pants down and if this is even remotely similar we are going to have a good time as consumers.

Problem is that even if they make 7nm or 10nm chips, they need to redesign most of the CPU pipeline. They can't just ramp clocks up as 5GHz is already a very limit before you start seeing electrical problems inside the CPU due to frequency itself.

[AngryBeaver]

33 minutes ago, RejZoR said:

Problem is that even if they make 7nm or 10nm chips, they need to redesign most of the CPU pipeline. They can't just ramp clocks up as 5GHz is already a very limit before you start seeing electrical problems inside the CPU due to frequency itself.

I don't expect them to grow much past 5ghz. I am saying they need either a new architecture or refine the current one enough to squeeze out another 15-20 ipc over the next few iterations.

 

Cores and IPC will be where the growth needs to happen. Which means that DX needs to find a way around the draw call limitations. That is pretty much all holding gaming back on these high core count CPUs 

[RejZoR]

2 hours ago, AngryBeaver said:

I don't expect them to grow much past 5ghz. I am saying they need either a new architecture or refine the current one enough to squeeze out another 15-20 ipc over the next few iterations.

 

Cores and IPC will be where the growth needs to happen. Which means that DX needs to find a way around the draw call limitations. That is pretty much all holding gaming back on these high core count CPUs 

Generally, clock and IPC are a heavily connected things, mostly mandated by compute pipeline deisign. Generally, if you want huge IPC, you're at least in general limited to lower clock speeds. If you want insane clock speeds, you're sacrificing IPC. Finding perfect balance between both is crucial for well performing CPU.

 

Remember how Athlon XP had unusually low clocks and still competed with Pentium 4's of the time that were clocked much much higher? That was IPC magic. Or how first generation of Core CPU's had low clocks as well. Same thing. Over time they mature and fine tune them and gain some clock headroom too, which is how Core processors slowly creeped up to current 5GHz. Yeah, the current Whatever Lake processors are still heavily based on initial Core architecture with some revisions and additions.

[Curufinwe_wins]

12 hours ago, RejZoR said:

Generally, clock and IPC are a heavily connected things, mostly mandated by compute pipeline deisign. Generally, if you want huge IPC, you're at least in general limited to lower clock speeds. If you want insane clock speeds, you're sacrificing IPC. Finding perfect balance between both is crucial for well performing CPU.

 

Remember how Athlon XP had unusually low clocks and still competed with Pentium 4's of the time that were clocked much much higher? That was IPC magic. Or how first generation of Core CPU's had low clocks as well. Same thing. Over time they mature and fine tune them and gain some clock headroom too, which is how Core processors slowly creeped up to current 5GHz. Yeah, the current Whatever Lake processors are still heavily based on initial Core architecture with some revisions and additions.

Also in many respects we are reaching silicon points where smaller processes might mean lower peak speeds not higher because of leakage issues, though it is possible (albeit not necessarily guaranteed) that with the inevitable switch to the next medium (be that GaAs or C nanotubes or whatever) that a large speed boost will become available as well. 

[AngryBeaver]

On 1/22/2020 at 11:11 PM, Curufinwe_wins said:

Also in many respects we are reaching silicon points where smaller processes might mean lower peak speeds not higher because of leakage issues, though it is possible (albeit not necessarily guaranteed) that with the inevitable switch to the next medium (be that GaAs or C nanotubes or whatever) that a large speed boost will become available as well. 

Yes and no. Smaller manufacturing processes to mean that electrical migration can happen and lower voltages, but the smaller processes also require less energy.

 

So yes the 16 cores can use a lot of energy even on the 7nm process, but then it comes down to how the lanes and gates are setup. Even though they might be using a lot of energy that energy is being spread to multiple dies so it isn't being concentrated.

 

You can also offset the "leakage" by lowering temps. So your average consumer might not see 5ghz anymore, but your enthusiasts with big air or even custom loops will more than likely be able to keep temps in the realm to make it possible.

 

So in the end 7nm can avoid problems with "leakage" by spreading that power ot multiple die on the chip, requiring less overall energy due to efficiency improvements on the smaller process, and keeping temps under control.  The limiting factor is going to the quality of the silicon.

 

So in general high overclocks = higher voltage and higher voltage = higher temps.  Higher temps = more "leakage" which means you need more voltage... which is a vicious cycle. So as long as you have good silicon 5ghz should be doable. I mean the current golden chips on the 3x series can get pretty close on custom loops (on the good chiplets) and pretty much anything sub-zero can. So in this regard AMD us only a refinement away. Intel has already showed the new 10xx stuff can do 5ghz on their laptop cpus. So I think we will see them stay at the 5ghz range for a whole yet. That is unless make architecture that has huge IPC gains, but lower clocks.

[leadeater]

21 hours ago, AngryBeaver said:

So in the end 7nm can avoid problems with "leakage" by spreading that power ot multiple die on the chip, requiring less overall energy due to efficiency improvements on the smaller process, and keeping temps under control.  The limiting factor is going to the quality of the silicon.

Those two things are unrelated though, chiplets are not mutually inclusive of process nodes. EUV will help compared to DUV as it'll lower then resistiveness of the traces in the die through more accurate and defined traces and will allow arrangements that were not possible or problematic before because of the deficiencies of DUV compared to EUV.

 

This is what Intel has been fighting with 10nm, increase in resistance and lack of accuracy which lead to the reduction in clocks. They still made really good gains but because of the inherit issues with DUV cannot offer high clocks due to the power which comes with leakage, EUV will help mitigate those issues.

 

Right now we have people effectively trying to paint precisely accurate 1mm width lines using 10cm brushes and masking tape, it's possible but much easier to use a smaller brush.

[Curufinwe_wins]

8 hours ago, AngryBeaver said:

Yes and no. Smaller manufacturing processes to mean that electrical migration can happen and lower voltages, but the smaller processes also require less energy.

 

So yes the 16 cores can use a lot of energy even on the 7nm process, but then it comes down to how the lanes and gates are setup. Even though they might be using a lot of energy that energy is being spread to multiple dies so it isn't being concentrated.

 

You can also offset the "leakage" by lowering temps. So your average consumer might not see 5ghz anymore, but your enthusiasts with big air or even custom loops will more than likely be able to keep temps in the realm to make it possible.

 

So in the end 7nm can avoid problems with "leakage" by spreading that power ot multiple die on the chip, requiring less overall energy due to efficiency improvements on the smaller process, and keeping temps under control.  The limiting factor is going to the quality of the silicon.

 

So in general high overclocks = higher voltage and higher voltage = higher temps.  Higher temps = more "leakage" which means you need more voltage... which is a vicious cycle. So as long as you have good silicon 5ghz should be doable. I mean the current golden chips on the 3x series can get pretty close on custom loops (on the good chiplets) and pretty much anything sub-zero can. So in this regard AMD us only a refinement away. Intel has already showed the new 10xx stuff can do 5ghz on their laptop cpus. So I think we will see them stay at the 5ghz range for a whole yet. That is unless make architecture that has huge IPC gains, but lower clocks.

Not to be an ass.... but I'm a graduate degree holding nuclear engineer who is quite familiar with the actual electronics and quantum interactions (varying with temperature and potential etc) that are of more and more relevance at these ever decreasing gap lengths.

 

I was over simplifying for the sake of the previous conversational audience. Suffice to say that regardless of temperature (within realistic limits that actual consumers can see) there are not many inevitable and automatic speed increases left on silicon at the same architecture depth with further node shrinks.

[AngryBeaver]

8 hours ago, Curufinwe_wins said:

Not to be an ass.... but I'm a graduate degree holding nuclear engineer who is quite familiar with the actual electronics and quantum interactions (varying with temperature and potential etc) that are of more and more relevance at these ever decreasing gap lengths.

 

I was over simplifying for the sake of the previous conversational audience. Suffice to say that regardless of temperature (within realistic limits that actual consumers can see) there are not many inevitable and automatic speed increases left on silicon at the same architecture depth with further node shrinks.

Again I am not saying there isn't. I think we are pretty much at the realistic cap of frequency.

 

I was just saying that going down to a smaller process doesn't automatically mean you lose frequency. There are a lot of factors at play for it.

[PHYLO]

Honestly depending on price, I'll be upgrading my 3700x to a 4000 series 12 core (or more) CPU. I thought 8 cores would be enough for motion graphics (it is in most cases), but I really would like to boost that preview and render speed a liiiiittle bit more. Although for now, it is a very nice step up from my 4790K.

 

Let's also hope that Zen 3 12 core CPUs will play nicely with my X470 board.