AMD Announces 32-core / 64-threads Ryzen Threadripper (12nm Zen+)

Published by

Click here to post a comment for AMD Announces 32-core / 64-threads Ryzen Threadripper (12nm Zen+) on our message forum
data/avatar/default/avatar09.webp
I wonder if we will see the ghz and core/IPC race end in delineation where we end up with a dual socket system with one huge socket for a "many core" CPU and 2nd socket for an optional "fast" CPU with only 4 cores. There would need to be OS and compiler changes but how cool would it be to have a 32 core system with an additional 4 cores that could do 5.4 ghz (maybe higher if this is a specialized chip)? The workstation stuff I do sometimes pegs every core but I also frequently see a single core pegged.
data/avatar/default/avatar37.webp
Why are people even commenting on this chip from a gamer's perspective. It is being marketed as a HEDT CPU, which means its for people with an intended workload. It would have the potential to allow you and 7 other people to game on it, if all things computing must resolve to gaming potential.
https://forums.guru3d.com/data/avatars/m/196/196426.jpg
This is simply insane !!! Just a couple of years back the best CPU one could hope for a home workstation computer was a decently fast 6-core Intel or a sluggish 8-core bulldozer, and this year 32-core monsters... !! I simply can't imagine what could be done with so much computing power under the hood, I don't think there's anything right now that would get this to 100%, other than benchmarks and scientific computation tasks. (Well maybe 3D rendering but not that many people do that to be honest)
data/avatar/default/avatar22.webp
nizzen:

There is no problem with 250w+ with the top tier x399 mainboards 🙂 I have Asrock x399 fatality pro , and have no problems overclocking my 1950x to 4.2ghz on all cores 🙂
i do think vrm will be an issue, you possible cap the same power with 32 core at 2.1 ghz compare to 16 core to 4.2, some motherboards with updated generous vrm and bigger cooling could be needed. Well i guess is all speculation untill we see a working prototype.
https://forums.guru3d.com/data/avatars/m/246/246171.jpg
I really think AMD should've had the number go up for every 4 cores added (with the 2800X being the exception): 2800X - 8 cores 2900X - 12 cores 2910X - 16 cores 2930X - 24 cores 2950X - 32 cores #970X - 48 cores #990X - 64 cores I don't think we're ever going to see more than 64 cores on socket TR4.
SaLaDiN666:

The same pr stunt as Intel did, cpus with higher core counts /16+/ gonna be bottlenecked thanks to the quad channel and the new added parts will not have the direct access to the memory, that is going to add huge latencies and performance penalty.
So when Intel has 4-channel memory in socket 2066, it's no problem. But when AMD does for sTR4, it's insufficient? 28 cores isn't that big of a difference from 32. If you want the 8 memory channels, just get an Epyc. It's probably going to cost less than Intel's 28-core i9 while also giving you full RAID and ECC support.
data/avatar/default/avatar18.webp
But its still "only" quad channel memory? I mean it would have to to incorporate the new CPU on the same socket? Epyc CPUs are already 8-channel RAM. This is basically a full fledged Epyc CPU but with messed up RAM channels so we only get 4-channel RAM? I need 200GB/sec RAM bandwidth.
https://forums.guru3d.com/data/avatars/m/246/246171.jpg
Ssateneth:

This is basically a full fledged Epyc CPU but with messed up RAM channels so we only get 4-channel RAM? I need 200GB/sec RAM bandwidth.
Then get an Epyc... it's not like they're all that expensive for what you're getting. There are already Epycs with 32 cores and 8 memory channels, so what are you waiting for? Also, if 200GB/s is what you need, you might want to go for Intel on this one. AMD's memory bandwidth on TR4/SR3 can take a real performance loss depending on the application, whereas Intel doesn't.
https://forums.guru3d.com/data/avatars/m/268/268759.jpg
AMD still trolling Intel xD Intel just announced 28c 56t and AMD goes for 32c 64t lol, maybe Intel monothreaded clocks gona be higher but let´s see multithreaded clocks greetings
https://forums.guru3d.com/data/avatars/m/219/219428.jpg
nizzen:

Explane, and make the forums better 🙂
From a GN article “For our standardized Blender test, we were measuring power consumption of 218W with the i9-7960X 16C/32T CPU at stock settings, or 492W when overclocked to 4.6GHz with a 1.22VID. That’s 40A down the EPS12V cables, at that point. We measured about 214W for the stock 7980XE – within usual variance of the 7960X – and also measured nearly 500W with the 4.5GHz overclock. Now, of course, how worthwhile this tradeoff is depends heavily on the actual performance improvement – we’ll get to that next. Comparatively, the AMD Threadripper 1950X CPU consumed about 144W in the same test. We saw the 1920X plot around the same, with a 145W throughput. Overclocking the 1950X landed it at 274W, though we only ever achieved a 4.0GHz overclock in initial testing, so it wasn’t really worthwhile” And this is while running a 16 core chip not the mention 28 cores.
https://forums.guru3d.com/data/avatars/m/156/156133.jpg
Moderator
Ssateneth:

But its still "only" quad channel memory? I mean it would have to to incorporate the new CPU on the same socket? Epyc CPUs are already 8-channel RAM. This is basically a full fledged Epyc CPU but with messed up RAM channels so we only get 4-channel RAM? I need 200GB/sec RAM bandwidth.
That's kind of out of desktop realm, if you want workstation grade performance then you need to buy workstation grade parts.
data/avatar/default/avatar27.webp
schmidtbag:

I really think AMD should've had the number go up for every 4 cores added (with the 2800X being the exception): 2800X - 8 cores 2900X - 12 cores 2910X - 16 cores 2930X - 24 cores 2950X - 32 cores #970X - 48 cores #990X - 64 cores I don't think we're ever going to see more than 64 cores on socket TR4. So when Intel has 4-channel memory in socket 2066, it's no problem. But when AMD does for sTR4, it's insufficient? 28 cores isn't that big of a difference from 32. If you want the 8 memory channels, just get an Epyc. It's probably going to cost less than Intel's 28-core i9 while also giving you full RAID and ECC support.
Threadripper is 2-way NUMA dual-channel memory (2x2 channels). Since Threadripper 2 has more than 2-way NUMA nodes for 4 dies and still have quad channel memory from out side, you got single channel memory for each die (4x1channels) or dual channel for 2 die and far memory latency for other 2(2x2 + 2x0 channels). Both configurations are terrible. Intel is using unified mesh for ram access so quad/hex channel works for them. PS: Thant 28 cores Intel chip is hex channel.
https://forums.guru3d.com/data/avatars/m/56/56686.jpg
intresting times coming, wonder what is gona be sacrifced for that many cores, First thing i see being saraficed speed ( GHZ ) cause heat is gona become factor. and latency? renew the cpu wars, on who is better faster and cheaper!
https://forums.guru3d.com/data/avatars/m/246/246171.jpg
tsunami231:

intresting times coming, wonder what is gona be sacrifced for that many cores, First thing i see being saraficed speed ( GHZ ) cause heat is gona become factor. and latency? renew the cpu wars, on who is better faster and cheaper!
Ryzens don't OC that high to begin with and the huge surface of TRs are good for dissipating heat, so I don't think heat is going to be a major concern (remember, 32-core Epycs already exist). I'm sure latency will be absolutely terrible, but systems like these aren't built with responsiveness in mind; they're built for crunching huge amounts of data quickly. If you want responsive, get socket AM4 or 1151 (and even then, 1151 is the better option).
https://forums.guru3d.com/data/avatars/m/243/243702.jpg
Mikeztm:

Threadripper is 2-way NUMA dual-channel memory (2x2 channels). Since Threadripper 2 has more than 2-way NUMA nodes for 4 dies and still have quad channel memory from out side, you got single channel memory for each die (4x1channels) or dual channel for 2 die and far memory latency for other 2(2x2 + 2x0 channels). Both configurations are terrible. Intel is using unified mesh for ram access so quad/hex channel works for them. PS: Thant 28 cores Intel chip is hex channel.
Come back once you have real numbers for latencies and bandwidths. Till then there is IF between CCXes and IMC, same way as intel has UM.
https://forums.guru3d.com/data/avatars/m/270/270233.jpg
Looks like investors are impressed with AMD's showing. Stock is up 5% today. 🙂
https://forums.guru3d.com/data/avatars/m/269/269560.jpg
I just wonder what all these cores are going to talk about: Core 1: hey Core 32, how is It up there ? Core 32: pretty cool ! Just warning up actually. And you ? Core 1: running hot down here. 100% busy for the time being. Would like a multi thread break. I am going to complain to the workload manager. Core 32: good luck with that. I heard he is coming from Microsoft.
https://forums.guru3d.com/data/avatars/m/268/268848.jpg
According to Anandtech the 32 core is like 4* Ryzen 7 2700 with 32*3.4GHZ@250W! and there is chance AMD will raise this turbo speeds before official launch!
Both the 24-core and 32-core sample CPUs are clocked at 3.0GHz base and 3.4GHz all-core turbo, with the latter being a work-in-progress according to the company. The 32-core system was equipped with DDR4-3200 memory. This is notable because the Ryzen processors based on the same 12nm Zeppelin dies officially max out at DDR4-2933. The codename for the processor family is listed as "Colfax". This is the first we've heard this codename from AMD. Despite the high TDP, both CPUs used in AMD's demos were air-cooled, using AMD's Wraith Ripper Air Cooler
https://forums.guru3d.com/data/avatars/m/258/258688.jpg
Amx85:

AMD still trolling Intel xD Intel just announced 28c 56t and AMD goes for 32c 64t lol, maybe Intel monothreaded clocks gona be higher but let´s see multithreaded clocks greetings
Generally, when core count ratchets way up like this, MHz drops like a rock. So that's the challenge...bringing up the core count and maintaining the MHz clocks and power consumption...;) It's really tough to do, imo. IMO--and this is purely a guess--I'd think the 32 core clock for TR is going to come down significantly from the 16 core TR clock, but, and this is the whole point, in the same multithreaded software the 32-core cpu will blow it away, nonetheless. These will not be all that great for gaming, possibly. But at 2560x1440 and up, what difference does the cpu make? What I would love to see from AMD is a new Ryzen chipset for the gaming high end, that would offer 2-channel/4channel support (determined by how many of the slots the user elects to fill, etc.) that would also offer more PCIex16/8x as well. Games just don't benefit from more than ~6 cores, max--and most do well to use 4 cores. And as I mention above--if you prefer gaming @ 2560x1440 (I think that QHD + FSAA is the sweet spot for gaming today) then it almost doesn't matter what cpu you are running as frame-rate is bottlenecked by the GPU.
https://forums.guru3d.com/data/avatars/m/258/258688.jpg
D3M1G0D:

Looks like investors are impressed with AMD's showing. Stock is up 5% today. 🙂
It's been far undervalued, imo, for a long time.
BigMaMaInHouse:

According to Anandtech the 32 core is like 4* Ryzen 7 2700 with 32*3.4GHZ@250W! and there is chance AMD will raise this turbo speeds before official launch!
Impressive, indeed! Better than I would have thought they could do--that 12nm they're using is a really nice process.
https://forums.guru3d.com/data/avatars/m/267/267641.jpg
Well, image world without stupid x86 licencing.. Why well have to wait on AMD, it could happen long time ago, image if Nvidia and other would have x86 licence too.