Ryzen 4000 and X670 scheduled for late 2020

[porina]

28 minutes ago, Taf the Ghost said:

Density matters far, far more for the Caches than the cores themselves. In AMD's CCD, the actual x86 cores are like 20% of the die space, with the L3 at like 50%. For reference.

I love those caches. They in large part mitigate the lack of bandwidth internal to Zen 2. I guess we have 7nm to thank for the more generous amount.

 

28 minutes ago, Taf the Ghost said:

Some of Intel's clocking problems is those AVX512 units. They're power hogs and produce a lot of heat, so any time they're active (even if not in full use) it kills their power budgets and, as a result, clocks.

I've not noticed any adverse effects from having and not using AVX-512. They can use a lot of power in use, but at the same time, they are doing a lot of work. We even see the same with Zen 2 and AVX2 workloads. With the improved FPU compared to earlier Zen, Zen 2 does more work, but it very noticeably uses more power to do so. As far as I can tell AMD have not come up with anything magic in their implementation, other than having a better limiter than Intel to keep it under control. Run some AVX2 heavy code on Zen 2 and you'll see it hit power limit and clocks drop compared to lesser workloads. It is a better limiter than offset ratios of Intel, although Intel also do have a power limiter, just that it is rarely used on enthusiast systems.

 

28 minutes ago, Taf the Ghost said:

That being said, 5 Ghz is probably too high of clocks to expect ever against on silicon-based nodes.

Outside of specific niches, it's only really useful for marketing, and in real world efficiency in some form is more important. As long as we don't get to P4 era again when it comes to chasing clocks...

[PianoPlayer88Key]

3 hours ago, Froody129 said:

You're going to be waiting a long time for minimum 80 times the performance of current CPUs... Nevermind at a third of the cost. 

Well I did give an example of when that kind of a jump actually happened.  Also, Tech Deals said during one of the RogueTech shows, if I heard him right, that he thought the next several years of CPU upgrades could be more like the 1990s instead of the last several years.  Some of us have unfortunately gotten used to relative stagnation in the industry, but it wasn't always that way.  I was a bit young in the 90s (was born in 1981), but I do remember at least some of it.

 

 

I looked up prices of Intel CPUs in some magazine ads (from a local computer magazine, no longer being published) I still have from back in the 1990s to 2000s.  Some vendors had prices all over the place relative to each other, though (for example I saw the same CPU going for like $500 at one place and $350 at another in the same week - the magazine was published weekly).  Also some vendors, I'm guessing because of the price volatility, didn't even publish their prices in the magazine, instead just saying call for current prices.

I mostly looked up CPUs that launched around the approximate price that most mainstream gamers have been buying Intel CPUs the last several years (like i5 or i7 price range - $250 to $350 or so).  Then, for subsequent years, I looked at how far the price had dropped, or if that CPU wasn't listed, picked a higher-tier model that had a lower price, then looked up that year's $250-350 or so range CPUs, generally picking 1 or 2.

 

• In January 1994,
  • (no previous year price reference here)
    • The 486DX2-50 was $279
• In January 1995,
  • The 486DX2-50 had dropped to $105
    • The 486DX4-100 was $355
    • The Pentium 60 was $376
• In November 1995,
  • The 486DX4-100 had dropped to $80
  • The Pentium 75 was $155 (I didn't see the 60 for Nov 1995)
    • The Pentium 100 was $288
• In February 1997,
  • The Pentium 100 had dropped to $99
    • The Pentium 166 was $295
    • The Pentium 166 MMX was $389
• In January 1998,
  • The Pentium 166 had dropped to $98
  • The Pentium 166 MMX had dropped to $112
    • The Pentium 233 MMX was $280
    • The Pentium II 233 was $265
• In January 1999,
  • The Pentium 233 MMX had dropped to $79
  • The Pentium II 266 was $197 (it had been $495 in 1998, I didn't see the 233 for 1999)
    • The Pentium II 400 was $349
• In January 2000,
  • The Pentium II 400 had dropped to $119.95
    • The Pentium III 533 was $299
• In March 2001,
  • The Pentium III 533 had dropped to $114
    • The Pentium III 1000 was $269
• In March 2002,
  • The Pentium III 1000 had dropped to $125
    • The Pentium 4 1.9 was $239
• In March 2003,
  • The Pentium 4 2.0 was $169 (had been $355 in 2002, I didn't see the 1.9 for 2003)
    • The Pentium 4 2.66 was $239
• In August 2004,
  • The Pentium 4 2.8 was $167 (had been $365 in 2003, I didn't see the 2.8 for 2004)
    • The Pentium 4 HT 3.2 was $272

I also had a few magazines with ads / prices going through about September 2007 (I think the Q6600 is in there possibly) but I had already taken a while to look these prices up.  I used to also have copies of the same magazine dating back to 1987, but I don't remember what happened to them.  The oldest one I have is from March 1990, then March 1993 is the next one I have but that one didn't have any CPUs listed individually, although it did have FPUs - they were called x87 back in the day and were separate chips.

 

 

[cj09beira]

57 minutes ago, yian88 said:

Hopefully late 2020 actually means August 2020... and hopefully AMD doesnt hold on new releases too long if intel has no new products competitive. AMD must continue being on a wave, if they stop for a second to withhold performance to make more money they will fail hard. By the time intel gets its shit togheter in 2021 AMD must be on par or better and make lots of money to ensure at least 2-3 arhitecture generations investment (around ~5-6 years of CPU releases).

But even so i still believe AMD will fail, dont know about intel they have money and fabs to design new shit. Why? the future is mostly mobile and power optimized, current x86 design is slow big,power hungry and outdated alongside the windows 10 shit OS that has no future. I can clearly see mobile OS'es and ARM cpu's taking more of the market and making high end workstation and gaming PC's a niche, AMD cant survive on a niche market of CPU. Intel on the other hand with their own fabs, talent and $ they have they can invent and adapt to new markets easily, one revolution could be FOVEROS that adapts big.Little design with stacked RAM or HBM-like memory for mobile laptops and mid range desktop they can easily overtake all mobile segment, the only problem is we dont have a desktop mobile OS, windows is the worst OS in existance, google refuses to make a push for Android desktop, its such a huge potential they just cant see it, by improving Google Play store for Pro Work Apps, and the interface for a mobile laptop experience android with ARM/Foveros on desktop would easily takeover.

I feel like any x86 Windows laptop is severely handicapped, neither the arhitecture or OS is made for mobile, its slow, heavy, big files, not power optimized and buggy, looking at my Pocophone SD845 and i cringe at my Ryzen 3 laptop or any laptop its just too slow/inefficient.

i dont think amd would be caught of guard because of either arm or risk-v, they have already done various arm processors, and probably only stopped due to them being in a bad situation financially, with their know how they would also not have much of a problem moving to arm/risk-v if need be, they are certainly playing close attention to that part of the market.

21 minutes ago, PianoPlayer88Key said:

Well I did give an example of when that kind of a jump actually happened.  Also, Tech Deals said during one of the RogueTech shows, if I heard him right, that he thought the next several years of CPU upgrades could be more like the 1990s instead of the last several years.  Some of us have unfortunately gotten used to relative stagnation in the industry, but it wasn't always that way.  I was a bit young in the 90s (was born in 1981), but I do remember at least some of it.

 

 

I looked up prices of Intel CPUs in some magazine ads (from a local computer magazine, no longer being published) I still have from back in the 1990s to 2000s.  Some vendors had prices all over the place relative to each other, though (for example I saw the same CPU going for like $500 at one place and $350 at another in the same week - the magazine was published weekly).  Also some vendors, I'm guessing because of the price volatility, didn't even publish their prices in the magazine, instead just saying call for current prices.

I mostly looked up CPUs that launched around the approximate price that most mainstream gamers have been buying Intel CPUs the last several years (like i5 or i7 price range - $250 to $350 or so).  Then, for subsequent years, I looked at how far the price had dropped, or if that CPU wasn't listed, picked a higher-tier model that had a lower price, then looked up that year's $250-350 or so range CPUs, generally picking 1 or 2.

 

• In January 1994,
  • (no previous year price reference here)
    • The 486DX2-50 was $279
• In January 1995,
  • The 486DX2-50 had dropped to $105
    • The 486DX4-100 was $355
    • The Pentium 60 was $376
• In November 1995,
  • The 486DX4-100 had dropped to $80
  • The Pentium 75 was $155 (I didn't see the 60 for Nov 1995)
    • The Pentium 100 was $288
• In February 1997,
  • The Pentium 100 had dropped to $99
    • The Pentium 166 was $295
    • The Pentium 166 MMX was $389
• In January 1998,
  • The Pentium 166 had dropped to $98
  • The Pentium 166 MMX had dropped to $112
    • The Pentium 233 MMX was $280
    • The Pentium II 233 was $265
• In January 1999,
  • The Pentium 233 MMX had dropped to $79
  • The Pentium II 266 was $197 (it had been $495 in 1998, I didn't see the 233 for 1999)
    • The Pentium II 400 was $349
• In January 2000,
  • The Pentium II 400 had dropped to $119.95
    • The Pentium III 533 was $299
• In March 2001,
  • The Pentium III 533 had dropped to $114
    • The Pentium III 1000 was $269
• In March 2002,
  • The Pentium III 1000 had dropped to $125
    • The Pentium 4 1.9 was $239
• In March 2003,
  • The Pentium 4 2.0 was $169 (had been $355 in 2002, I didn't see the 1.9 for 2003)
    • The Pentium 4 2.66 was $239
• In August 2004,
  • The Pentium 4 2.8 was $167 (had been $365 in 2003, I didn't see the 2.8 for 2004)
    • The Pentium 4 HT 3.2 was $272

I also had a few magazines with ads / prices going through about September 2007 (I think the Q6600 is in there possibly) but I had already taken a while to look these prices up.  I used to also have copies of the same magazine dating back to 1987, but I don't remember what happened to them.  The oldest one I have is from March 1990, then March 1993 is the next one I have but that one didn't have any CPUs listed individually, although it did have FPUs - they were called x87 back in the day and were separate chips.

 

 

1980s like improvements is a bit much, but i can see a big step like improvement coming from moving to have hbm like memory on package, and another from moving the cores themselves to a better semi conductor, 

the latter will take time but should completely change the market, that is probably when amd will in the most danger of dying.

the word i would use for the next decade would probably be turbulent 

[JoshB2084]

Ryzen 4000 and X670 already?  I just got Ryzen 5 2400G and B450 today.... 

[williamcll]

How does VIA make good chipsets when their own processor lineup is so subpar?

[Dabombinable]

Just now, williamcll said:

How does VIA make good chipsets when their own processor lineup is so subpar?

Ample practice. Apollo Pro 133A was damn near perfect from my experience. And they dominated for a while when Intel tried to persist with chipsets that only worked with RAMBUS.

[leadeater]

1 hour ago, porina said:

I've not noticed any adverse effects from having and not using AVX-512. They can use a lot of power in use, but at the same time, they are doing a lot of work. We even see the same with Zen 2 and AVX2 workloads. With the improved FPU compared to earlier Zen, Zen 2 does more work, but it very noticeably uses more power to do so. As far as I can tell AMD have not come up with anything magic in their implementation, other than having a better limiter than Intel to keep it under control. Run some AVX2 heavy code on Zen 2 and you'll see it hit power limit and clocks drop compared to lesser workloads. It is a better limiter than offset ratios of Intel, although Intel also do have a power limiter, just that it is rarely used on enthusiast systems.

Clocks are a lot like Who's Line Is It Anyway, "where the clocks points don't matter". Work being done is what matters and work requires energy, that's also why even for how efficient CPUs have gotten a lot of the time peak power has actually gone up. Until the 14nm 4 core stagnation product performance increase never came with lower power draw, trend line had until then always been up and is over the last few years returned to that same trend again. That trend won't last long though as we've already hit thermal transfer threshold of CPUs to coolers and there is no way water cooling in any form will become mandatory across any wider set of products.

 

https://en.wikipedia.org/wiki/List_of_CPU_power_dissipation_figures#Intel_Core_i7

 

Power and thermal limits are going to be a rude slap in the face to people expecting great improvements to come in the next few years from either AMD or Intel.

[pas008]

33 minutes ago, leadeater said:

Clocks are a lot like Who's Line Is It Anyway, "where the clocks points don't matter". Work being done is what matters and work requires energy, that's also why even for how efficient CPUs have gotten a lot of the time peak power has actually gone up. Until the 14nm 4 core stagnation product performance increase never came with lower power draw, trend line had until then always been up and is over the last few years returned to that same trend again. That trend won't last long though as we've already hit thermal transfer threshold of CPUs to coolers and there is no way water cooling in any form will become mandatory across any wider set of products.

 

https://en.wikipedia.org/wiki/List_of_CPU_power_dissipation_figures#Intel_Core_i7

 

Power and thermal limits are going to be a rude slap in the face to people expecting great improvements to come in the next few years from either AMD or Intel.

think this is why intel is trying for 30% of all silicon chips, not solely on cpus

bob and even brain have made this clear for intel

[porina]

1 hour ago, leadeater said:

Until the 14nm 4 core stagnation product performance increase never came with lower power draw, trend line had until then always been up and is over the last few years returned to that same trend again.

Picking some assorted CPUs through the ages from that list:

38W TDP - Pentium II 266 (my 3rd system)

72W TDP - Athlon 1400 (my fastest single core AMD CPU ever)

65W TDP - Core 2 Duo E6600 (I had many of these)

95/105W TDP - Core 2 Quad Q6600 (I had many of the 105W models as I was too tight to pay extra for the 95W ones)

95/125W - Phenom II X6 1055T (6 cores in 2010!)

95W TDP - i7-2600k (my first DDR3 era system)

91W TDP - i7-6700k (my first DDR4 system)

95W TDP - i9-9900k

127W TDP - i9-9900ks

 

I wonder what TDP the 486 in my 1st system was... I recall I did have a fan on that, not all of them did. The heatsink is tiny by today's standards and it was probably a 40mm fan, 60mm tops but I might be thinking of later Celeron system. If you were to slap a big Noctua on it, you probably wouldn't even need a fan.

 

Top end consumer CPUs seem to be hovering around 100W, outside of some niche models. This ties in with your comment that cooling has to be practical for mainstream systems. Before anyone jumps in with the "Intel takes more than TDP", 1, so do AMD, and 2, big name OEM systems do stick to TDP. Enthusiast systems tend to run unlimited. Both are considered in spec (not overclocking) by Intel.

 

Quote

Power and thermal limits are going to be a rude slap in the face to people expecting great improvements to come in the next few years from either AMD or Intel.

And we're back to perf per watt, as that will drive increases if sustained power were not to continue going up.

 

Zen 2 is doing great in that area. By default I think every stock system respects the PPT limit so you don't get unlimited power draw. Within that power limit, you can still exercise higher single core clocks, or balanced multi-core performance. More cores, better efficiency per core, more performance overall. The big requirement is that software can scale well with ever increasing core count, and not all software can ever do that.

 

A caution is, moving data around is eating more into the power budget. Zen 2 is maybe the 1st consumer CPU where it starting to be noticeable. Minimising that will be key to delivering more performance per watt in the near future.

[Founders]

5 hours ago, Teddy07 said:

Do mainboard manufacturers so far upgrade their bios to support new Ryzen processors?

It depends on the manufacturer, most will upgrade the BIOS to the newest revision before shipping boards out. Depending on how long they sit in a warehouse somewhere or on a retail shelf will determine whether the BIOS revision is still current. Best practice is to keep your BIOS up to date on your own.
 

5 hours ago, Mihle said:

Not sure if I want to wait for that or if I should just get Ryzen 3000 now...

 

I have i5 3570k now.

This depends entirely on you. If you feel your computer is still performing adequately for your needs then there’s really no need to upgrade. If you decide it’s time for an upgrade the end of the year is always the best time with sales. If you keep waiting for the next iteration to be released you’ll be waiting awhile.

 

2 hours ago, JoshB2084 said:

Ryzen 4000 and X670 already?  I just got Ryzen 5 2400G and B450 today.... 

Don’t feel bad, that’s just the way it is with tech. Just enjoy what you have now.

[MageTank]

5 hours ago, porina said:

Skylake might do it with extreme cooling.

Kaby Lake could do it overclocked on a good sample.

Coffee Lake v1 was the first offered at 5 GHz single core turbo, and fair chance to hitting all core with manual OC.

Coffee Lake v2 was the first offered at 5 GHz all core turbo, if you exclude the special Skylake-X part which was also rated all cores at 5 GHz.

Broadwell: "Am I a joke to you?"

[LeSheen]

3 hours ago, yian88 said:

I can clearly see mobile OS'es and ARM cpu's taking more of the market and making high end workstation and gaming PC's a niche, AMD cant survive on a niche market of CPU.

I don't see ARM taking over. There is not enough compatible actually used software. The biggest market is still business, not general consumers. As long as there is no absolute need for it they won't make the jump. Change is expensive and mostly slow so I wouldn't bet on it.

 

[Doobeedoo]

Nice, each year a very good improvement. Curious to see clock improvements too. 

[porina]

17 minutes ago, MageTank said:

Broadwell: "Am I a joke to you?"

Broadwell (desktop) was very late and was only out for a short time before Skylake came along anyway. I can't remember exactly, but my 5675C overclocked to around 4.0 or 4.1 without extreme cooling. Either side of it we have Haswell and Skylake, both of which were easily pushed to high 4.x GHz without below ambient cooling.

[Drak3]

1 hour ago, MageTank said:

Broadwell: "Am I a joke to you?"

Yes.

[MageTank]

1 hour ago, porina said:

Broadwell (desktop) was very late and was only out for a short time before Skylake came along anyway. I can't remember exactly, but my 5675C overclocked to around 4.0 or 4.1 without extreme cooling. Either side of it we have Haswell and Skylake, both of which were easily pushed to high 4.x GHz without below ambient cooling.

True, but it does serve to illustrate that the first launch of 14nm wasn't all that great, and it took some to progressively mature into something faster. I kinda like the L4 cache structure of those Broadwell S chips, and would like to see Intel return to that design in the future.

[porina]

8 minutes ago, MageTank said:

I kinda like the L4 cache structure of those Broadwell S chips, and would like to see Intel return to that design in the future.

While I also like that fat L4 cache, the bigger L3 approach of Zen 2 is also interesting. Would Intel take either or neither route going forwards? While Sunny Cove gives a Zen 2 like improvement in IPC, it sticks to the L3 cache sizes traditionally used. Then again, in Ice Lake form anyway, it is mobile first so maybe power is more important than performance, and we can see something bigger on future desktop. 

 

4 hours ago, williamcll said:

How does VIA make good chipsets when their own processor lineup is so subpar?

Just for you, today they announced a new x86 CPU that seems, on paper at least, competitive with AMD and Intel architectures. Will have to wait and see how actual products perform of course.

https://fuse.wikichip.org/news/3099/centaur-unveils-its-new-server-class-x86-core-cns-adds-avx-512/

 

I had to look it up, but Centaur is owned by Via.

[Dabombinable]

6 hours ago, porina said:

While I also like that fat L4 cache, the bigger L3 approach of Zen 2 is also interesting. Would Intel take either or neither route going forwards? While Sunny Cove gives a Zen 2 like improvement in IPC, it sticks to the L3 cache sizes traditionally used. Then again, in Ice Lake form anyway, it is mobile first so maybe power is more important than performance, and we can see something bigger on future desktop. 

 

Just for you, today they announced a new x86 CPU that seems, on paper at least, competitive with AMD and Intel architectures. Will have to wait and see how actual products perform of course.

https://fuse.wikichip.org/news/3099/centaur-unveils-its-new-server-class-x86-core-cns-adds-avx-512/

 

I had to look it up, but Centaur is owned by Via.

My old SS7 board actually handles old VIA and Centaur (before being bought) CPU...

[Unixsystem]

Wow, fake news. I just read in that other thread that Ryzen 4000 CPUs were coming out in Q1! 

 

Spoiler

/s

 

[aaradorn]

Big PP RN, ngl. Should I wait for DDR5 to upgrade?

[cj09beira]

20 hours ago, porina said:

While I also like that fat L4 cache, the bigger L3 approach of Zen 2 is also interesting. Would Intel take either or neither route going forwards? While Sunny Cove gives a Zen 2 like improvement in IPC, it sticks to the L3 cache sizes traditionally used. Then again, in Ice Lake form anyway, it is mobile first so maybe power is more important than performance, and we can see something bigger on future desktop. 

 

Just for you, today they announced a new x86 CPU that seems, on paper at least, competitive with AMD and Intel architectures. Will have to wait and see how actual products perform of course.

https://fuse.wikichip.org/news/3099/centaur-unveils-its-new-server-class-x86-core-cns-adds-avx-512/

 

I had to look it up, but Centaur is owned by Via.

there were some rumors amd is also looking to increase L1/L2 bandwidth a lot, since they are behind on that one, not sure if the rumor was for zen 3 or 4 though

[JediFragger]

4 hours ago, aaradorn said:

Big PP RN, ngl.

Queen's English?

[aaradorn]

3 hours ago, JediFragger said:

Queen's English?

Good god no, not for the queen. I had an erection at the moment I was reading the article due to the announcements made by AMD.

[MageTank]

23 hours ago, JediFragger said:

Queen's English?

I believe what @aaradorn was saying amounted to "The endowment of my nether region hath increased at this moment in time, for that I doth not deceive."

 

That's about as close as I can get when translating his text to the Queen's English. 

[JediFragger]

Thanks @MageTank that helps!