AMD Announces 2nd Generation Ryzen Threadripper 2000, up to 32 Cores/64 Threads!

[holyprof]

But does it run Crysis?

I guess it can run 4 instances simultaneously or maybe 8 for the 32 core CPU.

 

[coonbro]

your right but in my fridge and my cooler ..

 

[GorbazTheDragon]

I see no reason a 4 or 6 core configuration without HT couldn't do just that. But that in the eyes of people buying core counts these days, SKUs like that might make AMD look bad.

It's more a question of what the architecture and process are capable of. If we are going to see 4.5GHz on these high core count chips we should expect to see them pretty soon as well on the 8 core and lower ones.

At the moment (or at least when the reviews hit) the 2700x doesn't seem to like going beyond 4.2GHz, combined with the currently dismal state of mid range AM4 motherboard VRMs, AMDs offerings are somewhat lacking in single thread performance. If they can squeeze even another 5% or maybe I can be hopeful and expect closer to 10%, they will be neck and neck with even intel's 8 core chips.

 

[INSTG8R]

Suppressed they went for Proshop as their Norwegian preorder online retailer, was Komplett on the original Ryzen launch.

Komplett is the Norwegian equivalent to Newegg, Proshop is not that big.




Given good cooling, it should be able to boost beyond 4.4, maybe? since it will have 180W to play with.

I dunno I thought I got a pretty good deal on my Vega from Proshop, but yeah I’m a little surprised they picked them over Komplett.

 

[thebluebumblebee]

i9=ino?

Anyone else find it incredible that the 32 core CPU is simply double the price of the 16 core CPU? Not like the exponential scale that Intel has used over the years, which, in Intel's defense, is because of yield issues with their monolith CPU's.

I don't understand why @btarunr is comparing the ("glued together" ) $1,800 2990WX to the $1,900 Core i9-7980XE, instead of the $4,100 Xeon E5-2699-v4 (22C/44T), especially since he mentioned that the "WX" CPU's are aimed at workstation builds, and the Xeon is a "workstation/server" CPU? AMD bringing out the 2990WX at half the price of Intel's best is jaw dropping. Even the leaked Cinebench score shows the 2990WX about equal to the Xeon E5-2699-v4.

How AMD is wiring the dies out on the 24-core and 32-core MCMs remains a mystery. We know from how EPYC is wired out, that the SP3r2/TR4 socket has four distinct "zones," each corresponding to a die. Wiring out 1 memory channel per die may neither be possible nor desirable. It may not be possible because current socket TR4 motherboards are built for MCMs with just two active dies, the diagonally-opposite ones, and as such the memory wiring of the four channels terminate at the pins of those two active dies; it may not be desirable either, because one memory channel per die would severely cripple memory bandwidth for single-threaded or less-parallelized tasks that AMD likes to localize to one die with dual-channel memory access.

I'm pretty sure they have this worked out: AMD's Jim Anderson: "...We were always planning more than 16 cores." (at 2:45)

 

[theonedub]

i9=ino?

Anyone else find it incredible that the 32 core CPU is simply double the price of the 16 core CPU? Not like the exponential scale that Intel has used over the years, which, in Intel's defense, is because of yield issues with their monolith CPU's.

I don't understand why @btarunr is comparing the ("glued together" ) $1,800 2990WX to the $1,900 Core i9-7980XE, instead of the $4,100 Xeon E5-2699-v4 (22C/44T), especially since he mentioned that the "WX" CPU's are aimed at workstation builds, and the Xeon is a "workstation/server" CPU? AMD bringing out the 2990WX at half the price of Intel's best is jaw dropping. Even the leaked Cinebench score shows the 2990WX about equal to the Xeon E5-2699-v4.


I'm pretty sure they have this worked out: AMD's Jim Anderson: "...We were always planning more than 16 cores." (at 2:45)

Can you link where the Xeon E5-2699-v4 is compared to the 2990WX? I thought that the 1950X went about toe to toe with that Xeon, I would hope the 32c TR2 would outright demolish it?

 

[eidairaman1]

With the new feeds naysaying AMD will find a way to launch 8 channel boards while maintaining backwards compatibility with tr 1 adopters. Afaik X399 chipset appears to be a scalable design.

 

[ZoneDymo]

So much want

 

[Durvelle27]

4.4GHz on a 12C is very interesting

I just can’t wait for reviews

 

[Brusfantomet]

With the new feeds naysaying AMD will find a way to launch 8 channel boards while maintaining backwards compatibility with tr 1 adopters. Afaik X399 chipset appears to be a scalable design.

Considering the memory controller is in the CPU, and the Treadripper socket is based on EPYC it should be possible, but the boards will be extremely expensive

 

[eidairaman1]

Considering the memory controller is in the CPU, and the Treadripper socket is based on EPYC it should be possible, but the boards will be extremely expensive

What you expect for a wx chip lol.

 

[RejZoR]

AMD needs to split up CPU's into gaming and productivity parts where gaming one is pushing higher clocks with fewer cores and productivity ones with more cores and lower clocks. I think they ares slowly going in that direction.

 

[repman244]

i9=ino?

Anyone else find it incredible that the 32 core CPU is simply double the price of the 16 core CPU? Not like the exponential scale that Intel has used over the years, which, in Intel's defense, is because of yield issues with their monolith CPU's.

And even those, are Xeon rejects, so you are basically buying garbage from them at higher price.

 

[thebluebumblebee]

Can you link where the Xeon E5-2699-v4 is compared to the 2990WX? I thought that the 1950X went about toe to toe with that Xeon, I would hope the 32c TR2 would outright demolish it?

I'm simply looking at the leaked Cinebench score and compared that to https://us.rebusfarm.net/en/tempbench?view=benchmark I don't know much about either source.

 

[Captain_Tom]

With the new feeds naysaying AMD will find a way to launch 8 channel boards while maintaining backwards compatibility with tr 1 adopters. Afaik X399 chipset appears to be a scalable design.

I would actually be a little surprised if they added 8-channel memory boards next year. I am sure they CAN, but I think this is the perfect way to segment HEDT from full Server chips.

After all you can overclock DDR4 to ~3200-4000MHz with x399, and thus you can almost makes up for having less bandwidth for those who care. Furthermore AMD is supposedly doubling core count per CCX with 7nm Ryzen 3, and so TR3 will likely be a 32/64 design utilizing only two dies once again.

The idea is that if you need just a ton of threads right now (and not much else), the 2990WX is a complete steel for the price. If you need the CCX's to all have proper bandwidth (Or gaming is a large concern), then just get a 2950X or pay big for EPYC!

 

[Brusfantomet]

I would actually be a little surprised if they added 8-channel memory boards next year. I am sure they CAN, but I think this is the perfect way to segment HEDT from full Server chips.

After all you can overclock DDR4 to ~3200-4000MHz with x399, and thus you can almost makes up for having less bandwidth for those who care. Furthermore AMD is supposedly doubling core count per CCX with 7nm Ryzen 3, and so TR3 will likely be a 32/64 design utilizing only two dies once again.

The idea is that if you need just a ton of threads right now (and not much else), the 2990WX is a complete steel for the price. If you need the CCX's to all have proper bandwidth (Or gaming is a large concern), then just get a 2950X or pay big for EPYC!

32 core epyc launched at $2100
24 core epyc launched at $1075

I think its to offload dies with working cores but defect memory controllers.

 

[Captain_Tom]

32 core epyc launched at $2100
24 core epyc launched at $1075

I think its to offload dies with working cores but defect memory controllers.

Perhaps, and thus even more reason to not launch 8-channel kits. This allows them to sell 32-core chips for insane prices.

 

[notb]

Intel paused the core count war in 2010, what did we(consumers) got for that? same cpu every year with %5 performance increasment most of which coming from similar clock spead increasement? so no core count war is welcome.

We got laptops with useful performance levels that work for half a day and take less space than an A4 (paper) notebook.

Can you link where the Xeon E5-2699-v4 is compared to the 2990WX? I thought that the 1950X went about toe to toe with that Xeon, I would hope the 32c TR2 would outright demolish it?

E5-2699-v4 works in dual CPU setups.

Intel are due for some serious ass kicking in the HPC HEDT segment,serves them right.

Possibly just as much "ass kicking in HEDT" as they did with the first TR.

 

[phill]

So much love for AMD right now.. So much want about having one of the 2990WX CPUs!! Roll on the reviews

 

[mohammed2006]

Will not upgrade well not upgrade wait ddr5. Talking to myself.

 

[Nkd]

2920X @ 649 looks like good bang for buck being 12 cores and a fair price that an average consumer might pay; they do for gfx cards so why not cpus? Beyond that 12 or even 16 cores, our software is so far behind these core counts; we need actual progress in frequency increase rather than core spam. Both Intel and AMD need to stop the rush to glue cores together instead of actual innovation. We need 8 cores / 16 T at 5or 6ghz before 24 cores at 3.

I am not sure what you are talking about. World doesn't revolve around gaming only. There are plenty of people who can utilize 32 cores. Its not designed for gaming alone in mind. By software being far behind core count you must mean gaming. Because plenty of other software supports multicore processors beyond eight.

AMD needs to split up CPU's into gaming and productivity parts where gaming one is pushing higher clocks with fewer cores and productivity ones with more cores and lower clocks. I think they ares slowly going in that direction.

Clocks will come with 7nm. Just wait for it. Across the board likely. Thats the beauty of their design. Even with more cores they can run around 4ghz. I can bet with good vrm on a board and good cooling the 32 core can probably run 4ghz across all cores.

 

[Captain_Tom]

Will not upgrade well not upgrade wait ddr5. Talking to myself.

I have lightning fast DDR4 (3600MHz @ 14-14-28) that I got for $76 in 2016 lol, so I am ready for Ryzen 3. 16-cores @5GHz won't be that expensive of an upgrade from my 6700K.

 

[Frick]

Clocks will come with 7nm. Just wait for it. Across the board likely. Thats the beauty of their design. Even with more cores they can run around 4ghz. I can bet with good vrm on a board and good cooling the 32 core can probably run 4ghz across all cores.

I'm slightly worried over this prevalent attitude. It sets the stage for dissapointment as it's incredibly vague. It's like the people saying No Mans Sky will be incredible with the new dlc.

 

[Prima.Vera]

I have lightning fast DDR4 (3600MHz @ 14-14-28) that I got for $76 in 2016 lol, so I am ready for Ryzen 3. 16-cores @5GHz won't be that expensive of an upgrade from my 6700K.

I know you are bullshitting since currently there are no CL14 RAM sticks at 3600MHz , and not even in 2016 you couldn't buy that cheap, common!
Try again.

 

[londiste]

AMD's 16-core $999 1950X proved competitive with even Intel's 12-core and 14-core SKUs priced well above the $1200-mark;

Even competitive with cpus that have 2 (12,5%) or 4 (25%) less cores? Yeah, that sounds very competitive indeed

The 4-die 2970WX and 2990WX should logically have 8 memory channels and up to 128 PCIe lanes, because you are adding up the connectivity of all four dies on the MCM; however, there are no socket TR4 motherboards in the market or on the horizon, which have 8-channel memory slots, or PCIe slots that utilize all those PCIe lanes.

Are you trying to say AMD is scumbags who are artificially limiting the awesome 8-channel memory capabilities of the package?

Regarding memory channel wiring, that is a performance optimization decision. There is no right or wrong decision here, just depends on what they are going for. I would say there is no problem for them to wire one channel to each die as that is done (or can be done) on the processor package and does not depend on socket/motherboard knowing about it. While you are right that having one channel per die would limit the uses cases where threads are confined to one die and would not have access to dual-channel memory, the flip side of that is when there are 2 channels connected to two dies, the 2 other dies will not have direct access to memory at all and will have to go through the (already bottlenecking) IF links for memory accesses. That would have implications to many-thread performance which at least theoretically would be worse for workstation use cases. I am really curious if we will see how they did it once reviews are out.