32C/64T AMD Ryzen Threadripper Cinebench scores surface

[3dcandy]

Nope but if you want an all core clock of 4ghz I'm pretty sure you're not going down the air cooled route. It's a premium product and will need a premium cooler. But Intel decided to aim for 5 ghz and it took a huge cooler - I'm not sure AMD have the same plans. They are more than happy to get a 16c 32t chip out at a decent speed rather than a hand picked special for 5ghz

[ik9000]

Nope but if you want an all core clock of 4ghz I'm pretty sure you're not going down the air cooled route. It's a premium product and will need a premium cooler. But Intel decided to aim for 5 ghz and it took a huge cooler - I'm not sure AMD have the same plans. They are more than happy to get a 16c 32t chip out at a decent speed rather than a hand picked special for 5ghz

i might have misread but didn't the chart say just that - a stable 4.12GHz on air cooling on all cores? Will need to check when have more time

[3dcandy]

Yup I think it did...but you don't know the cooler Could be mahoosive

[Xlucine]

4.12GHz on all cores should look more like ~7500 points (scaling from here)

[CAT-THE-FIFTH]

4.12GHz on all cores should look more like ~7500 points (scaling from here)

Half the physical chips are not using their memory controllers,but relying on the other two chips instead,so I would imagine some performance losses as a result.

[afiretruck]

4.12GHz on all cores should look more like ~7500 points (scaling from here)

Don't forget that the way the chips in this Threadripper communicate with their memory will drag the per-thread performance down a notch. Especially those threads in the CCX's without memory directly attached.

[mtyson]

This heavy metal 2nd Gen Threadripper teaser video was shared by AMD a few hours ago

[kalniel]

Ermm, so the bracket needs extra support?

[CAT-THE-FIFTH]

Ermm, so the bracket needs extra support?

Maybe they will get Julie to help!

[ETR316]

The battle is heating up. If Intel counters next year with something impressive also, then we all will benefit, because competition will push AMD to engineer better also.

bottom line, dont fall for the hype and take sides like years ago. keep both companies going strong, so they can both engineer greatness, if not, AMD will fall into the same old same old like Intel did for a few years. So, although its great to see AMD doing well, i also want Intel to come back with somthing impressive as well next year.

[DanceswithUnix]

The battle is heating up. If Intel counters next year with something impressive also, then we all will benefit, because competition will push AMD to engineer better also.

bottom line, dont fall for the hype and take sides like years ago. keep both companies going strong, so they can both engineer greatness, if not, AMD will fall into the same old same old like Intel did for a few years. So, although its great to see AMD doing well, i also want Intel to come back with somthing impressive as well next year.

Trying hard doesn't always work. Netburst and Itanium are Intel's recent attempts at revolutionary designs, before they went back to the Pentium Pro derived Core series. Similarly AMD's Bulldozer was their best attempt at the time, they don't all come off as well as Ryzen or Sledgehammer.

I don't think AMD would fall into a rut if they became dominant, they just don't seem to have the same culture of milking customers that Intel does.

[kompukare]

The battle is heating up. If Intel counters next year with something impressive also, then we all will benefit, because competition will push AMD to engineer better also.

bottom line, dont fall for the hype and take sides like years ago. keep both companies going strong, so they can both engineer greatness, if not, AMD will fall into the same old same old like Intel did for a few years. So, although its great to see AMD doing well, i also want Intel to come back with somthing impressive as well next year.

That's too soon surely?
If you think about it, if AMD currently have 12+% of the CPU market (think those were the Q4 2017 figures). and that's marketshare not profit share.

For them to actually be able to properly pay for research and development, they'd want to have 20-30% marketshare and a similar profit share for a few years. Otherwise, Intel are easily able to weather this with their 80% marketshare and immense profits and only delivery the bare minimum improvement (although mainstream going from being quad core for nearly a decade to suddenly hex core and soon octo core is a nice improvements - which Intel were anyway going to do as it absolutely 100% has nothing to with Ryzen!)

Even if Intel are forced to dig into their margins a bit, remember this was the company which wasted for $4 billion on dumping (aka contra revenue) Atom CPUs, plus $billions on things like McAfee, Larrabee and similar things while still making record profits.

[Corky34]

....which Intel were anyway going to do as it absolutely 100% has nothing to with Ryzen!)

Were they? I suspect Intel would have continued marketing four cores as mainstream CPUs for decades if it wasn't for AMD pushing them on core count.

The way Intel connect multiple cores and how they've favored high clock speeds, and the resulting thermal constraints of that, sort of means the more cores they use the less efficient, in terms of cost and performance, they become.

[kompukare]

Were they? I suspect Intel would have continued marketing four cores as mainstream CPUs for decades if it wasn't for AMD pushing them on core count.

The way Intel connect multiple cores and how they've favored high clock speeds, and the resulting thermal constraints of that, sort of means the more cores they use the less efficient, in terms of cost and performance, they become.

Ah, my earlier post might have a need a *sarc* tag.

Although it is true that ring bus used on Intel's mainstream suffers with cores. It's not very noticeable on hex, so we'll see how to octo will fare. But AFAIR they went mesh bus for far higher core counts than eight.

[Corky34]

Sorry i missed the sarcasm.

It's sort of true with Intel going mesh with higher core counts (idk where the cut off point is) but it's a sort of frankenstein mesh as it's a mesh made up from lots of interconnected half-rings so data has to hop from node to node meaning as core count increase so do latencies.

[eugenius]

I wonder if the eventual dual tower Noctua cooler will be capable of sustaining 4ghz all core aka 400W load.