The Ryzen "ThreadRipper"... 16 cores of awesome

[TheGiant]

I am missing the high clocked high tdp 12C. (4GHz max clock)

Whatever the moar coarz hype ends, IMO the HEDT ends up with one purpose- gaming while doing something else in the background. But primary the gaming perf must be there. AMD has the advantage here because intel seriously missed that spot with SKL-X and I believe 4GHz zen performs better in gaming low percentile and min fps than SKL-X.

I still can't imagine what is the desktop workload for 16 cores that doesn't get the penalty of lower clocks ? what do you guys do with 16 cores? (please don't post cinebench that's not desktop workload IMO)

 

[TheGiant]

Correct. Once you have enough heat transfer area, you can dissipate the heat with a low delta T. Of course an air cooler would need a LOT of fin area exposed to an air flow for 300W. The heat pipes are a trivial problem to solve due to the large heatspeader area and the fact that the two ryzen die are not concentrated but well spaced underneath.

I hate the whole bling thing for PCs, but that might be just my age speaking.

I think you need more than that. SKL-X proved in extreme the old engineering truth that the weakest spot in the system makes the bottleneck and that is the heat transfer between the core and the HS, not actual HS to cooler base.

 

[estarkey7]

I am missing the high clocked high tdp 12C. (4GHz max clock)

Whatever the moar coarz hype ends, IMO the HEDT ends up with one purpose- gaming while doing something else in the background. But primary the gaming perf must be there. AMD has the advantage here because intel seriously missed that spot with SKL-X and I believe 4GHz zen performs better in gaming low percentile and min fps than SKL-X.

I still can't imagine what is the desktop workload for 16 cores that doesn't get the penalty of lower clocks ? what do you guys do with 16 cores? (please don't post cinebench that's not desktop workload IMO)

Adobe Premiere Pro to video rendering! After doing color grading, Neat Video noise reduction and a few special effects, I once did a render out of a 4K H.264 project down scaled to 1080p for an hour and a half video output that took >47 hours! All 8 cores of my 8150-FX O.C. to 3.8GHz were at 98% utilization. And that is WITH GPU ASSIST from an RX-580.

I'm thinking Threadripper will get that down to less than six hours, maybe three hours if I use separate 960 Evo M.2 drives for source files and output files.


Sent from my SM-G935T using Tapatalk

 

[mattiasnyc]

I still can't imagine what is the desktop workload for 16 cores that doesn't get the penalty of lower clocks ? what do you guys do with 16 cores? (please don't post cinebench that's not desktop workload IMO)

I really don't understand why people have a hard time understanding this. If TR isn't the chip to use for cinebench-related workloads then which AMD chip is? An EPYC chip? EPYC is server. That leaves Ryzen. Ryzen performs less well.

TR competes with Intel HEDT chips, and as estarkey7 pointed out there are workloads in which you definitely benefit from more cores, and those types of jobs have been done on the Intel HEDT platform.

Why people are so hung up on gaming I will never understand.

 

[scannall]

I really don't understand why people have a hard time understanding this. If TR isn't the chip to use for cinebench-related workloads then which AMD chip is? An EPYC chip? EPYC is server. That leaves Ryzen. Ryzen performs less well.

TR competes with Intel HEDT chips, and as estarkey7 pointed out there are workloads in which you definitely benefit from more cores, and those types of jobs have been done on the Intel HEDT platform.

Why people are so hung up on gaming I will never understand.

Especially 1080 gaming. It just is meaningless in this context and use case.

 

[TheGiant]

Because you can buy a better CPU for that MT workload with lower price (EPYC, Xeon) than their HEDT counterparts.
And I consider EPYC and Xeon as workstation/ server CPU, not server CPU.
So if that workload is THE one, HEDT 16C makes no sense to use price/performance and pure performance wise.

That is they way I am looking at it. No offence and hard feelings OFC.

 

[beginner99]

If you need the pure crunching power, a 24c/48t Epyc 7401P costs $1075 suggested. Ignoring platform costs (both TR and Epyc have unknowns here), this is significantly more total CPU power than a Threadripper 1950X for $76 more, and it is more efficient due to sweet spot clockspeeds.

True but sometimes it's a trade-off if you have a mix of single/low threaded workloads and heavily threaded workloads. Here the Higher clocked less cores part usually wins clearly.

 

[moonbogg]

Especially 1080 gaming. It just is meaningless in this context and use case.

Correct. For these chips, gaming is not a particular focus. Its all about heavily threaded workloads and that's why AMD will find themselves in a hugely favorable position with this amazing release. The good news is that these chips will game just fine, of course. So if someone wants a beast rendering/production rig they can have that but also have good game performance. These aren't bulldozer chips. They have good IPC, its just not the best. With Threadripper, AMD is seriously taking advantage of Ryzen's strengths. Releasing 12 and 16 core chips was such a brilliant move it actually blows my mind.
None of the drawbacks or hesitations of the original Ryzen release with regard to gaming will apply to these new products. Like I said, If I was in the market for something like this I would be jumping out of my skin with excitement right now. I am VERY thrilled for anyone looking for an incredible rendering/production rig. This is nothing short of a new golden era for PC enthusiasts; full blown and full-fledged.

 

[DrMrLordX]

its amazing what great yields can do glue baby glue

Now I just want to see TDP numbers. I must admit I was sort of hoping for that $849 16c/32t part, but $999 for a clockspeed of 3.4 GHz is fine too. If they can keep TDP in the 140-160W range it'll be a killer.

4: X299 is dead. People buy these for highly threaded workloads. AMD just bested Intel's best.

I think it's premature to stick the fork in it just yet. We haven't seen how Intel will handle the 12-18c rollout whenver that actually happens. We have seen the prices, and the prices are ugly. You can basically get Epyc for what you would have to pay to get the 18c Skylake-X.

There will be some use cases for x299. It's caught in an awkward spot between Threadripper/Epyc on the MT side and Coffeelake on the ST/MT side.

Why people are so hung up on gaming I will never understand.

People are looking for the next big gaming chip beyond the 7700k. Ryzen and TR aren't clearly that, nor is Skylake-X. We'll see if Coffeelake delivers.

Especially 1080 gaming. It just is meaningless in this context and use case.

It's the "pro" gaming scene (actual pro or wannabe). They want/need those high framerates, and now we have vendors selling 240MHz refresh rate monitors . . .

 

[mattiasnyc]

Because you can buy a better CPU for that MT workload with lower price (EPYC, Xeon) than their HEDT counterparts.
And I consider EPYC and Xeon as workstation/ server CPU, not server CPU.
So if that workload is THE one, HEDT 16C makes no sense to use price/performance and pure performance wise.

That is they way I am looking at it. No offence and hard feelings OFC.

I don't see how that's true. Which server chips perform better at these types of workloads for the same price with the same required connectivity and flexibility on the motherboard?

From what I can tell the sweet spot for serious content creation has always been the HEDT lineup. You get all the connectivity you need on the motherboard, ranging from USB to Firewire to Thunderbolt etc, and you get a decent amount of cores, and the cores run at a nice and crisp speed.

For us in audio there are some workloads that benefit from higher frequencies and some workloads that benefit from more cores. It can even change within one project within a day. I expect video to be the same.

EDIT: I just did a quick googling and found one test giving the 2p EPYC 7601 a Cinebench score of 6879. TR 1950X scored 2431. The 7601 is $4000+, multiplied by two. The Threadripper is $1000.

What am I missing here?

 

[TheGiant]

True but sometimes it's a trade-off if you have a mix of single/low threaded workloads and heavily threaded workloads. Here the Higher clocked less cores part usually wins clearly.

and this is why I said I am missing the high clock 12C TR. 10C at 4,6 GHz is most of the time more than 16C at 3.5GHz

but....zen can't get past 4GHz and SKL-X is a furnace

We will see how TR handles the high MT workloads. Only cinebench is presented but corona, handbrake and c-ray have a good performance on SKL-X. It will be interesting...

I want to try the 12-16C threadripper on my CFD calculations...

 

[mattiasnyc]

beginner99 said: ↑
True but sometimes it's a trade-off if you have a mix of single/low threaded workloads and heavily threaded workloads. Here the Higher clocked less cores part usually wins clearly.

and this is why I said I am missing the high clock 12C TR.

The keyword is "sometimes".

10C at 4,6 GHz is most of the time more than 16C at 3.5GHz

but....zen can't get past 4GHz and SKL-X is a furnace

Exactly. So this is why these HEDT chips are the sweetspot for content creation. They strike a balance between speed and multi-core.

 

[beginner99]

and this is why I said I am missing the high clock 12C TR. 10C at 4,6 GHz is most of the time more than 16C at 3.5GHz

Yeah true. Wanted to write earlier that the 12 core is IMHO $50-$100 too expensive as a the 7820x especially OCed will come very close. And the TR out of ryzen 7 experience will have a hardwall at 4 ghz.

 

[TheGiant]

Yeah true. Wanted to write earlier that the 12 core is IMHO $50-$100 too expensive as a the 7820x especially OCed will come very close. And the TR out of ryzen 7 experience will have a hardwall at 4 ghz.

That

That was my maybe not quite clear question. On HEDT, where the 10C high frequency is less than 16C low frequency. I am saying that if it is, then your workload is suitable for the xeon,epyc CPU.

Maybe with TR AMD can keep up the freq the same- ryzen or TR- 4GHz is achievable, but if Intel somehow solves the SKL-X thermal issue then they will win this round

The next one I think AMD is the clear winner- Intel is slow those times- so until they react properly to zen 2 it will take another 2 years...

 

[StinkyPinky]

Exciting times in the CPU market that's for sure. And we still have that mystical 6 core coffee lake due in the next couple of months. Finally after a decade of tedious incremental updates we are seeing some movement.

 

[MarkPost]

Yeah true. Wanted to write earlier that the 12 core is IMHO $50-$100 too expensive as a the 7820x especially OCed will come very close. And the TR out of ryzen 7 experience will have a hardwall at 4 ghz.

I think TR 12c price is correct. imo it will be clearly faster overall than even an oced 7820x in MT tasks. We only have Cinebench score, but its a start:

-TR 1920X 12/24 (3.5ghz, supposedly running with XFR at 3.6): 2431
-i7 7820X 8/16 (running at 4.0 all cores with default turbo): 1734 (Anand review)

So 1920X is 40% faster than 7820X

A 7820X can be oced (best scenario, with top noch cooling) to 4.5-4.7 (24/7, i'm talking about 100% stable, with sustainable temperatures and consumption). Thats as much 18% more clock (keeping in mind 7820X runs at 4.0ghz all cores with default turbo).

And of course, TR can be oced too.. I think to the Ryzen levels (3.8-4.0). Thats about 11% more clock.

Really, at the end of the day, OC isnt a key factor here.

 

[krumme]

Amd sells a lot of r7 6c dies vs 8c looking at the sale ranks. That leaves a lot of unused 8c.
I think pricing just reflects they want to sell 16c variants.
Shouldn't be that hard at 999...for list price. Retail might easily go as low as 899.
It looks to me they are not only fishing for all the sklx 10 to 14c variants but also a few of the 7820x customers might be lured away by the tr 16c part.

The tim is a damn mess for Intel. A tr 16c will easily be cooled by eg an noctua d15 and keep rpm low. An 7820x cant do the same. You are forced to keep that cooler running like a beast. Especially if you favor sub 75c.

And then comes efficiency.

 

[MarkPost]

The only "bad" thing here is that TR needs new coolers. But for someone planning to buy a $800-1000 CPU, it wont be a great issue after all.

 

[formulav8]

I must admit I was sort of hoping for that $849 16c/32t part

They will more than likely add more sku's later. Probably including an 8 core with more 12/16 core options.

 

[french toast]

Thanks for the explanation @ the stilt et al.

This just confirms my suspicions- Lisa su just pulled Brian krzanich's pants down with TR.

 

[Timmah!]

Amd sells a lot of r7 6c dies vs 8c looking at the sale ranks. That leaves a lot of unused 8c.
I think pricing just reflects they want to sell 16c variants.
Shouldn't be that hard at 999...for list price. Retail might easily go as low as 899.
It looks to me they are not only fishing for all the sklx 10 to 14c variants but also a few of the 7820x customers might be lured away by the tr 16c part.

The tim is a damn mess for Intel. A tr 16c will easily be cooled by eg an noctua d15 and keep rpm low. An 7820x cant do the same. You are forced to keep that cooler running like a beast. Especially if you favor sub 75c.

And then comes efficiency.

This. I still cant comprehend why would they downgrade from solder to TIM the exact moment they finally have some competition. It goes against all logic.

 

[FIVR]

It makes sense if you consider their enterprise customers won't care and it probably saves them a good amount of money. They think they can "weather the PR storm" and come out on the other side with bags of extra cash, no worse for wear.

I wouldn't be surprised if they are right.

 

[Timmah!]

It makes sense if you consider their enterprise customers won't care and it probably saves them a good amount of money. They think they can "weather the PR storm" and come out on the other side with bags of extra cash, no worse for wear.

I wouldn't be surprised if they are right.

Then again arent enterprise customers buying Xeons instead of HEDT CPUs? But i agree, they may be right. I am still considering it myself. But i would be past that and already bought, if not for these thermal issues - they saved few bucks on solder, but lost (at least for now) entire grand on the missed sale cause of it. Past few years, i would have no other choice, and if i wanted/needed it, my money would end in their pocket anyway. This year they may not.

 

[scannall]

Then again arent enterprise customers buying Xeons instead of HEDT CPUs? But i agree, they may be right. I am still considering it myself. But i would be past that and already bought, if not for these thermal issues - they saved few bucks on solder, but lost (at least for now) entire grand on the missed sale cause of it. Past few years, i would have no other choice, and if i wanted/needed it, my money would end in their pocket anyway. This year they may not.

Yes, enterprise customers buy Xeons for a variety of reasons. ECC memory being just one of them. And now they have another great choice there as well to evaluate. The Intel HEDT parts this generation seem pushed pretty hard from the factory with little overhead and a lot of heat. It's a shame they choked this time around.

 

[Aenra]

The only "bad" thing here is that TR needs new coolers. But for someone planning to buy a $800-1000 CPU, it wont be a great issue after all.

Point. Theorising is all good (though boring for me), practice however is paramount.
You want these babies, you're possibly in for a long wait, and one that won't necessarily have the typical outcome (28374625 different coolers available). And i do mean long wait, because i can recall EKWB, Swiftech, even mainstream companies -or perhaps especially mainstream companies- occasionally having issues with their first water AIO iterations; they are not new, but even so, we get the same old "issues" every time a new "platform" comes up.. some of them end up leaky, sporting bad contacts, having overdraws, software glitches, etc etc., even though the tech is by now over a decade old

I personally never risk those.. they launch now, i give it at least a month and copious forum reading later before i order one.

(and since someone is just bound to say it, no. I wouldn't buy a traditional cooler, not for an unlocked 180 TDP chip, not for an HEDT platform i'm already wasting thousands on. Stingy and HEDT just should not mix, sorry to be so blunt, but truth be told)