Thoughts on the AMD Threadripper?

[ir_cow]

I don't ever remember a time where AMD chips had a reasonable TDP. They list one thing and than the actual numbers are very different. I think they Threadripper is going to be a very hot chip that requires AIO coolers or a completely redesigns of air coolers for more surface area.

[Guest_Jim_*]

I don't ever remember a time where AMD chips had a reasonable TDP. They list one thing and than the actual numbers are very different. I think they Threadripper is going to be a very hot chip that requires AIO coolers or a completely redesigns of air coolers for more surface area.

I'd have to look for a source, but one thing I have seem mentioned by some before is that Intel and AMD have different processes for measuring TDP. I think AMD's is something more of an average TDP, so lower than particularly heavy workloads produce, while Intel's represents more of a worst-case scenario, so higher values than most people will see.

 

Took a while to find anything with much specifics, but I did find a forum post elsewhere (so not official, and not really detailed either) that says AMD measures TDP at the core, while Intel measures TDP across the entire chip.

[Waco]

Not at all worried about latency or clock speeds tbh. Ryzen and it's Infinity Fabric were created from the ground up to be incredibly scale-able with little to no loss in efficiency. It's the main point of the architecture.

 

Just look at how there's no difference in clock speeds in the range of R5 vs R7 processors. It means that the platform just gets better with more cores. But the low core count models might suffer some.

 

In terms of temps even the largest model (16C/32T) is rumored to be 155W, which is great for such a high core count

 

Not something I'll be looking into right now, but I'll be keeping a close eye on it. I'll maybe look into the future equivalents in Zen 2 or Zen 3

Much of what you posted is marketing fluff, not reality.

[Burasutsu]

 

Not at all worried about latency or clock speeds tbh. Ryzen and it's Infinity Fabric were created from the ground up to be incredibly scale-able with little to no loss in efficiency. It's the main point of the architecture.

 

Just look at how there's no difference in clock speeds in the range of R5 vs R7 processors. It means that the platform just gets better with more cores. But the low core count models might suffer some.

 

In terms of temps even the largest model (16C/32T) is rumored to be 155W, which is great for such a high core count

 

Not something I'll be looking into right now, but I'll be keeping a close eye on it. I'll maybe look into the future equivalents in Zen 2 or Zen 3

Much of what you posted is marketing fluff, not reality.

 

 

In what way is this not reality? All the cores that I've seen reviewed have reached similar OC of somewhere between 3.9 and 4.2, with no difference that can't be put down to the silicon lottery. 

 

The naming for infinity fabric and the simple marketing explanation are fluff for sure, but the concept is still there and it has so far been proven to have an affect. I can only assume that trend will continue into the higher core counts. 

[Braegnok]

If prices are close to rumored,.. which sound too good to be true. I would love to build 1S workstation with ThreadRipper (aka IntelRipper) and a 2S server with EPYC,.. and of coarse some multinode parallel software. 

 

The fact they can sustain the clock frequency thanks to modularity is great. The only detail is the NUMA design but should not be too much of a problem. Instead of a 1 level crossbar, I can see 2 level crossbar, but that does not equal to 2x latency. If I am correct, they are using the I/O Hub of each Ryzen die to communicate between them, as they use PCIe lanes to run InifinityFabric protocol over them. It is still a P2P protocol, like CrossFireX.

Edited June 4, 2017 by Braegnok

[Waco]

 

 

Not at all worried about latency or clock speeds tbh. Ryzen and it's Infinity Fabric were created from the ground up to be incredibly scale-able with little to no loss in efficiency. It's the main point of the architecture.

 

Just look at how there's no difference in clock speeds in the range of R5 vs R7 processors. It means that the platform just gets better with more cores. But the low core count models might suffer some.

 

In terms of temps even the largest model (16C/32T) is rumored to be 155W, which is great for such a high core count

 

Not something I'll be looking into right now, but I'll be keeping a close eye on it. I'll maybe look into the future equivalents in Zen 2 or Zen 3

Much of what you posted is marketing fluff, not reality.

In what way is this not reality? All the cores that I've seen reviewed have reached similar OC of somewhere between 3.9 and 4.2, with no difference that can't be put down to the silicon lottery.

 

The naming for infinity fabric and the simple marketing explanation are fluff for sure, but the concept is still there and it has so far been proven to have an affect. I can only assume that trend will continue into the higher core counts.

Infinity Fabric adds more latency than the way Intel designs their rings for large chips.

 

Clocks are related purely to power usage - Threadripper and Epyc won't be clocked at 4 GHz for power reasons alone (unless you're okay with 200-400 watts under load).

[Braegnok]

I have to say, AMD is doing an amazing job with it's Ryzen lineup, and the X399 Whitehaven platform.

 

Let's just hope AMD's platform stability and compatibility are fully ironed-out until the expected June release.

Edited June 5, 2017 by Braegnok

[Waco]

Agreed 100%. Ryzen is a surprisingly good showing so far, minus the launch issues caused by pushing up the date before motherboards were ready.

 

Threadripper and Epyc look to shake up the HEDT and server markets a bit as well - I know my NAS would welcome 16+ cores to chew up 4K transcodes.

[Braegnok]

Looks like Intel got out of the bag, and said bring it. 

 

    Toppc's impressive 5,500MHz overclock,.. on Intels X299 Platform. https://www.pcper.com/news/Memory/GSkill-Memory-Used-World-Record-Breaking-DDR4-5500-MHz-Overclock

 

AMD is up for the challenge with recently announced AGESA update 1.0.0.6,.. adding 26 configurable memory options (including subtimings) 

 

The biggest change is the clockspeeds where AMD will now allow stable memory clocks up to 4,000 MHz without needing to adjust the CPU base clock,.. in increments of 133 MT/s rather than the currently supported 266 MT/s,.. also AMD will  allow 2T command rates with the new update for dialing in the highest possible stable clockspeeds. 

 

It's going to be great next month to see Ryzen paired with stable 4GHz memory and how that will help gaming and everyday performance,.. improve things in the Infinity Fabric and CCX to CCX latency department.    

Edited June 8, 2017 by Braegnok