It is official. AMD announced and demonstrated Heavy Metal 32c/64t Threadripper 2. 7nm on the way.

Page 2 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

mattiasnyc

Senior member
Mar 30, 2017
356
337
136
I can understand. AMD is saying "you don't need chilled water to run our new CPU". I am guessing clocks might even dip below 3 ghz like EPYC. You can't take 2 180 watt TR's and have it all running at 3.4 ghz and me 250 watt. But I do think we can get overclocked 3.7 or 3.8 on a good AIO.

But I will have one, and with a new motherboard as soon as I can get my hands on it. I only have 4 threadrippers now.

hmm... For 32 cores... that might be correct. I'm guessing though that with a better boost algorithm and better efficiency in general + being the top notch dies produced it might actually not lag that far behind the older Threadrippers, if at all.
 
  • Like
Reactions: Gikaseixas

IEC

Elite Member
Super Moderator
Jun 10, 2004
14,323
4,904
136
You can expect baseline clocks of (at minimum) a R7 2700. Which would be 3.2GHz base and 3.8GHz+ all core boost @ 65W depending on chip quality. Multiply x4 dies (barely handpicked) and you would easily meet the 250W TDP target.

But given that these are the highest few % binned dies of Ryzen 2000 series chips, I would expect it to be closer to 3.7GHz base and 4.0GHz all core boost at 250W TDP. The best Ryzen 2700X chips manage <1V at 3.7GHz and <1.1V at 4GHz, which would easily fall within TDP. If they've truly been holding back the best dies, it could easily be better than this...
 

The Stilt

Golden Member
Dec 5, 2015
1,709
3,057
106
You can expect baseline clocks of (at minimum) a R7 2700. Which would be 3.2GHz base and 3.8GHz+ all core boost @ 65W depending on chip quality. Multiply x4 dies (barely handpicked) and you would easily meet the 250W TDP target.

But given that these are the highest few % binned dies of Ryzen 2000 series chips, I would expect it to be closer to 3.7GHz base and 4.0GHz all core boost at 250W TDP. The best Ryzen 2700X chips manage <1V at 3.7GHz and <1.1V at 4GHz, which would easily fall within TDP. If they've truly been holding back the best dies, it could easily be better than this...

AM4 Pinnacle Ridge CPUs have >= 35% higher power limit than the advertised TDP.
We'll see if that will be the case with 2nd gen. TR as well.

1950X already drops it frequency below it's base in some cases, usually when the memory has been pushed above the maximum official speeds.
And that's at 180W power limit and 16 cores.

At 3.7GHz / 1.150V I've measured 220W+ consumption for the CPU itself.

12nm LP does provide higher frequency at ISO power than the 14nm LPP used on gen. 1 TR, however the difference is marginal (< 4% on average).

If the 250W power limit is accurate, the frequencies will be EPYC'ish.
 
  • Like
Reactions: Drazick and ub4ty

piesquared

Golden Member
Oct 16, 2006
1,651
473
136
hmm... For 32 cores... that might be correct. I'm guessing though that with a better boost algorithm and better efficiency in general + being the top notch dies produced it might actually not lag that far behind the older Threadrippers, if at all.

In the video with PCWorld, Jim Anderson also mentioned that there will be a couple new features in Ryzen Master for Threadripper 2. There has been speculation about that PB Overdrive feature that surfaced a while ago and will likely be a feature of these new Ripper beasts. Together with the best of the best binned dies, it will be interesting to see what boost frequency these hit.
 

IEC

Elite Member
Super Moderator
Jun 10, 2004
14,323
4,904
136
AM4 Pinnacle Ridge CPUs have >= 35% higher power limit than the advertised TDP.
We'll see if that will be the case with 2nd gen. TR as well.

1950X already drops it frequency below it's base in some cases, usually when the memory has been pushed above the maximum official speeds.
And that's at 180W power limit and 16 cores.

At 3.7GHz / 1.150V I've measured 220W+ consumption for the CPU itself.

12nm LP does provide higher frequency at ISO power than the 14nm LPP used on gen. 1 TR, however the difference is marginal (< 4% on average).

If the 250W power limit is accurate, the frequencies will be EPYC'ish.

In that case stock base/boost will depend entirely on what their definition of "250W" and what it encompasses entails.

If they are strict about power limits I could see closer to Epyc clocks. Which would be disappointing, but not unreasonable.

We know the dies themselves should be capable of 2700X+ clocks and are the best of the best, so I'm still optimistic a 4GHz all-core clock could be achieved at ~1.0V (since the better 2700X can get away with around 1.1V for those clocks).
 
  • Like
Reactions: ub4ty

Ajay

Lifer
Jan 8, 2001
15,332
7,792
136
We know the dies themselves should be capable of 2700X+ clocks and are the best of the best, so I'm still optimistic a 4GHz all-core clock could be achieved at ~1.0V (since the better 2700X can get away with around 1.1V for those clocks).

That seems unlikely at any reasonable power level. Sure, some overclocker will do it, or higher, using a chiller with a newer motherboard. Someone will go even higher with LN2, but these will just be interesting demonstrations. This whole bit with 24/32 cores seems like AMD is just trolling Intel (SKL-X).
 

Fir

Senior member
Jan 15, 2010
484
194
116
The Wraith Ripper air cooler


Please no more tower coolers!
This CPU is going to require a decent AIO for stock or light OC and a real LC system for high end OCs.
Think about it, 1950X pushes the limits at 4GHz with most AIO today. The Enermax seems to work best because its coldplate has an active area designed with TR4 in mind. This trend should follow suit but adding what is essentially twice the thermal output is going to raise water temperatures and ultimately core temperatures. A few benchmarks won't be enough to show the difference but huge rendering projects or distributed computing certainly will. The VRMs on current X399 boards will likely need some form of active cooling, even a slow 120mm fan helps tremendously.
 

ub4ty

Senior member
Jun 21, 2017
749
898
96
They should try running it at 5GHz lol
With the proper cooling, anything is achievable. Just ask intel
623523434125.jpg
 
  • Like
Reactions: IEC and lightmanek

ub4ty

Senior member
Jun 21, 2017
749
898
96
Please no more tower coolers!
This CPU is going to require a decent AIO for stock or light OC and a real LC system for high end OCs.
Think about it, 1950X pushes the limits at 4GHz with most AIO today. The Enermax seems to work best because its coldplate has an active area designed with TR4 in mind. This trend should follow suit but adding what is essentially twice the thermal output is going to raise water temperatures and ultimately core temperatures. A few benchmarks won't be enough to show the difference but huge rendering projects or distributed computing certainly will. The VRMs on current X399 boards will likely need some form of active cooling, even a slow 120mm fan helps tremendously.
Nope. I run my current 1950x with a slight OC and wattage creeps at or above 250Watts when max'd and the rushed Noctua works just fine :
index.php


With 250watts worth of heat spread out even more (4 dies vs 2) this should be just fine for stock. VRMs are fine too. I noticed the back of my mobo around the CPU bracket and VRMs around the CPU area where the v-neck cut out usually is in a computer cases gets kind of toasty so I put an exhaust fan in that area. Everything runs just fine and that's with some reference GPUs doing external exhaust and some after-market GPUs dumping toasty heat into the case. All you need is proper air flow and you're just fine. No way in heck i'm ever putting water anywhere near a workstation with soo many pricey components and data.
 

Fir

Senior member
Jan 15, 2010
484
194
116
Your VRMs benefit from having the airflow of a tower cooler, yes.
Blocking slots for ram and graphics cards is another reason.
There is no way a tower cooler is going to handle it. It will throttle or lock up when under heavy load. (2950X or whatever the 32 core part is going to be called)
And if the motherboard is vertical meaning the heatpipes are horizontal you lose even more capacity.
 

ub4ty

Senior member
Jun 21, 2017
749
898
96
Your VRMs benefit from having the airflow of a tower cooler, yes.
Blocking slots for ram and graphics cards is another reason.
There is no way a tower cooler is going to handle it. It will throttle or lock up when under heavy load. (2950X or whatever the 32 core part is going to be called)
And if the motherboard is vertical meaning the heatpipes are horizontal you lose even more capacity.
No ram slots are blocked. All you have to do is slide the fan mount up on the Noctua. I would like a more fleshed out bottom up revision from Noctua but what I have works and was the best Air cooler out at the time.

My graphics cards are definitely not blocked by it either (I chose Asrock that went for a much more sensible layout) nor did I have to use the Noctua adjusters at all to make room. I have a 1080 sitting right in slot #1. All of the PCIE slots are populated. Some with reference. Some with aftermarket GPUs. I have 1 NVME. 3 sata and 1 HD. I can max out every single component in a benchmark with a 1950x OC that hits 260+watts and nothing overheats or throttles. I ran Prime 95 for a full day and I did memtest 4 cycles worth on my OC'd ram. Nothing locks up. It's as stable as concrete.

I'm not speaking of theoreticals here. I run a pretty maxed out Threadripper setup at TDPs above the new 32 core. Everything runs just fine on air. OC'ing to absurd limits was fun when I was young and wanted to feel alive but it no longer appeals to me due to Absurd amounts of heat and power for no sensible performance. Water is too risky and unnecessary. Air is just fine and it helps to actually cool other components in the case besides the CPU/GPU.
 
  • Like
Reactions: Space Tyrant

Markfw

Moderator Emeritus, Elite Member
May 16, 2002
25,478
14,434
136
My take on this:....

First, All Ryzen, TR and EPYC cores are ECC capable and essensially the same CCX, but the motherboards for Ryzen and TR are not setup to support this.
Next, the 2700x is a nice improvement over the 1800x better memory and OC's better.
The EPYC chip (7551P to be exact) already has 32 cores and a 180 watt TDP.
So to take 4 of the 2700x "CCX's", wire them to fit the existing 4 channel memory on TR4 socket would be easy.
Take that 180watt TDP (@2.551 ghz all-core turbo) and bump it up to say 3 ghz or more due to the better cores in the 2700x, and 250 watt TDP is very doable.

The only questions I have, is how bad a hit will it take without direct memory access on 2 CCX's, and what will the final all-core turbo speed be. I say the speed will be 3.0-3.2, lets hope they beat that. And I hope the IF works well enough to make the hit only 10% or so. I looked at the 7551, $2,300 and 2.5 GHZ all-core turbo and the motherboard is $650. I will take even a 20% performance hit if the new chip is say $1500 or so.

My predictions... Lets see how close I come.
 

Thunder 57

Platinum Member
Aug 19, 2007
2,647
3,706
136
My take on this:....

First, All Ryzen, TR and EPYC cores are ECC capable and essensially the same CCX, but the motherboards for Ryzen and TR are not setup to support this.
Next, the 2700x is a nice improvement over the 1800x better memory and OC's better.
The EPYC chip (7551P to be exact) already has 32 cores and a 180 watt TDP.
So to take 4 of the 2700x "CCX's", wire them to fit the existing 4 channel memory on TR4 socket would be easy.
Take that 180watt TDP (@2.551 ghz all-core turbo) and bump it up to say 3 ghz or more due to the better cores in the 2700x, and 250 watt TDP is very doable.

The only questions I have, is how bad a hit will it take without direct memory access on 2 CCX's, and what will the final all-core turbo speed be. I say the speed will be 3.0-3.2, lets hope they beat that. And I hope the IF works well enough to make the hit only 10% or so. I looked at the 7551, $2,300 and 2.5 GHZ all-core turbo and the motherboard is $650. I will take even a 20% performance hit if the new chip is say $1500 or so.

My predictions... Lets see how close I come.

Sounds about right. FWIW, this piece says "Both will have a base frequency of 3GHz with a boost of 3.4GHz, though this last is a work-in-progress figure and could change before launch.". Not sure of the source, but Joel is reliable IMO.
 

formulav8

Diamond Member
Sep 18, 2000
7,004
522
126
link says it's compatible but some boards might struggle with the power requirements....

IIRC by Ian, the mobo vendors said the 250w TDP was conservative. So many existing boards could very well be quite fine with the new chips.
 

tamz_msc

Diamond Member
Jan 5, 2017
3,708
3,554
136
Nope. I run my current 1950x with a slight OC and wattage creeps at or above 250Watts when max'd and the rushed Noctua works just fine :


With 250watts worth of heat spread out even more (4 dies vs 2) this should be just fine for stock. VRMs are fine too. I noticed the back of my mobo around the CPU bracket and VRMs around the CPU area where the v-neck cut out usually is in a computer cases gets kind of toasty so I put an exhaust fan in that area. Everything runs just fine and that's with some reference GPUs doing external exhaust and some after-market GPUs dumping toasty heat into the case. All you need is proper air flow and you're just fine. No way in heck i'm ever putting water anywhere near a workstation with soo many pricey components and data.
It is to be noted that Be Quiet showed a TR4 version of their dual-tower Dark Rock Pro 4, with full IHS coverage and rated for 250W. Noctua also plans to update their main line-up(D15, U14s, U12s), giving the heatsinks more surface area, though given their long product development cycles I wonder if they'll come out this year, or if the corresponding TR4 models would be updated as well.

The showed the new prototype U12 with dual 12x25 fans(the fancy new ones) achieving the same performance as the D15.
 

piesquared

Golden Member
Oct 16, 2006
1,651
473
136
Looking at the 2700, wich is not as well binned as the 2700X, and wich hit 3.45GHz@65W on Prime 95 at stock settings, they should get something like 3.4-3.5GHz with all 32 cores within 250W.

I also expect the clocks will increase and be the fastest in it's class, but way lower price. Should be interesting to see how badly affected intel's chips are by their mesh issues.
 

ultimatebob

Lifer
Jul 1, 2001
25,135
2,445
126
So, what's the pricing going to be on one of these? I'm looking for an excuse to upgrade my older generation Core i7, but only if I can afford it.
 

piesquared

Golden Member
Oct 16, 2006
1,651
473
136
Trying to guess AMD's pricing strategy is futile and a crap shoot. Its practically guaranteed that they will be far less expensive than the best intel can offer though.
AMD is emerging as the defacto leader in the HEDT space. The comparison between the 2 demonstrations was lol. One was a hack and one was a real product. I dont think AMD would have much trouble getting 32/64 Threadripper 2 up to 5 GHz also under extreme conditions considering 12nm node and improvements to Zen+. Thats quite amazing that AMD has beaten intel for the fastest CPUs made in the span of one generation at probably a quarter of the cost and 4x the yield. And then a 7nm Zen 2 architecture that has been improved in multidimensional is coming in less than a year with 7nm Radeon Instinct GPUs coming this year (there no way another optimization of 14nm is going to beat 7nm). I bet that stings. Intel probably would have been better off not even showing anything.
 
Last edited:

Mopetar

Diamond Member
Jan 31, 2011
7,797
5,899
136
I dont think AMD would have much trouble getting 32/64 Threadripper 2 up to 5 GHz also under extreme conditions considering 12nm node and improvements to Zen+.

I don't think we'll see 5 GHz outside of a golden CPU with an overbuilt cooling setup (similar to the case with Intel's demo). Most people with a 2700X are hitting a wall around 4.3 GHz. You'd need LN2 in order to get to 5 GHz and that's pointless from a consumer point of view.

A 4 GHz all-core turbo is probably about as far as they can reasonably push it, and keep in mind that this will be compatible with older boards so keep in mind that they're somewhat constrained by that. I suspect that's why they'll have a base clock for this that's a bit conservative so that older boards designed to handle the original Threadripper's 180W TDP will work without hassle.