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

[IEC]

To be precise and maybe nitpicking, this is not correct.

You can have variable core values, but they introduce an asymmetry in the design which can impact performance depending on where a thread is running.

Per TheStilt, the PSP (Platform Security Processor) disallows asymmetric core CCX and socket configurations, so it is a moot point - there will be no asymmetric configurations even if they are theoretically possible.

 

[maddie]

And even more impressively, all of this is done from one die only, Zeppelin.

How many grew up playing with Lego blocks?

 

[maddie]

Per TheStilt, the PSP (Platform Security Processor) disallows asymmetric core CCX and socket configurations, so it is a moot point - there will be no asymmetric configurations even if they are theoretically possible.

Never claimed there would be any, but if we're trying to truly understand the technicalities of the processor, then we should be clear as to why things are as they are, and not make blanket statements that don't enlighten. That at least, is my position.

 

[Topweasel]

Per TheStilt, the PSP (Platform Security Processor) disallows asymmetric core CCX and socket configurations, so it is a moot point - there will be no asymmetric configurations even if they are theoretically possible.

This is the part I don't understand. I get them deciding to disallow it and right up their papers saying its not going to happen. Specially if its for performance reasons. But they control what the PSP does or doesn't allow. They have had samples that have had asymmetric configurations and those samples would still have had the PSP. What is preventing them from allowing the PSP to allow asymmetric configurations where they want them?

 

[mattiasnyc]

To be precise and maybe nitpicking, this is not correct.

You can have variable core values, but they introduce an asymmetry in the design which can impact performance depending on where a thread is running.

Sorry for asking this, but, is there a place you can point me to where I can read about this? I was on the fence about getting a Ryzen and now this seems like a far more appropriate investment for me, but I'd like to know more about this in case we're "surprised" by a 10 core model or whatever....

 

[mattiasnyc]

Well, that'd be good for AMD's bottom line, assuming there are buyers.
I'd pass at that price :shrug:
Might as well get a better clocking Skylake (unless Intel prices the 12-core at $2k or above).

I guess it'll depend on the workload. If a 16c Ripper can outdo an Intel chip priced the same then I don't see a reason to buy Intel, especially not if both offer quad channel DDR4 and 40+ PCIe lanes.

 

[IEC]

This is the part I don't understand. I get them deciding to disallow it and right up their papers saying its not going to happen. Specially if its for performance reasons. But they control what the PSP does or doesn't allow. They have had samples that have had asymmetric configurations and those samples would still have had the PSP. What is preventing them from allowing the PSP to allow asymmetric configurations where they want them?

Speculation here, but it may simply have to do with yields. Rumor is that yields are very good on 8-core Zeppelin dies. If true, that bodes well for a very competitively priced 16c/32t processor... as well as making it less economically feasible to go through all the testing, validation, etc. for additional variants outside of 12c/24t and 16c/32t SKUs - ones that you may not have many raw materials for (i.e. "defective" dies) to begin with.

 

[mattiasnyc]

What's the cost/revenue on a CPU though? Wouldn't they at some point start making a profit by "artificially" just disabling cores to get a cheaper part? Artificial segmentation of the market for sure, but still... Or is it verified that lower core-counts = defective cores that were shut off?

(I'm not trying to be argumentative btw, I'm trying to learn)

 

[IEC]

What's the cost/revenue on a CPU though? Wouldn't they at some point start making a profit by "artificially" just disabling cores to get a cheaper part? Artificial segmentation of the market for sure, but still... Or is it verified that lower core-counts = defective cores that were shut off?

(I'm not trying to be argumentative btw, I'm trying to learn)

Well that would lead to reduced margins versus selling the full fat chip. Assuming there is sufficient demand for the full fat chip. That would in turn lead to reduced profits. Given that AMD has traditionally seemed allergic to profit, it could happen. See: "4GB" reference RX 480s which were actually 8GB physically and electrically, and could be flashed with 8GB BIOSes as a recent example.

As a counterpoint, the "defective" cores can simply be dies that ran too hot or required too much voltage to hit the desired 8-core binnings - so are sold as dies with fewer cores enabled at lower clockspeeds. In which case they wouldn't be reducing profit, but instead salvaging it from "defective" dies.

 

[mattiasnyc]

I get the 'defective die' approach, it makes sense and it was also part of what I've understood so far. I did however see some people say that it's really not that common relatively speaking.

As far as cannibalizing profit from a higher margin item I get that too, but I'm just wondering if they really do saturate the market though. It would basically only make sense as long as there aren't more CPUs than customers (talking about 'all-cores' chips).

 

[Ajay]

I just wonder if there is some issue when running unpaired CCXs. We have 4,6 & 8 cores for Ryzen and a similar situation situation with Snowy Owl. Either that, or pricing would be so close that it just didn't make marketing sense.

 

[maddie]

Sorry for asking this, but, is there a place you can point me to where I can read about this? I was on the fence about getting a Ryzen and now this seems like a far more appropriate investment for me, but I'd like to know more about this in case we're "surprised" by a 10 core model or whatever....

If I'm remembering correctly, it was discussed in the strictly technical ryzen thread. A lot of posts there and to be honest I can't remember which ones. Have a look through if you want.

 

[Ajay]

If I'm remembering correctly, it was discussed in the strictly technical ryzen thread. A lot of posts there and to be honest I can't remember which ones. Have a look through if you want.

Guess I missed it - don't think I'll go back over the whole huge thread

 

[AMDisTheBEST]

Unfair considering that those 4 thread chips were often able to defeat an 8 thread AMD chip...

That was the bulldozer days. If amd hasnt made that mistake, we wouldve have 4 cores as mainstream for a while now and 4 core plus would no longer be the premium $500-$1700 demanded by intel.

 

[w3rd]

I am so burnt-out on reading.
But I am glad some people are coming around to the realization of what AMD is bringing to the table. The secret sauce is their fabric.

 

[ericlp]

It's funny that no one is really talking about how to COOL such a monster or its power needs. I guess Intel's 10 Core CPU is going to run at 140 watts plus. I'm thinking AMD's 16 core would be rocking around the 200 Watt Mark maybe even more... Are you sure the average gamer geek is going to say, well, now I am really looking at the 1000 watt power supplies as a new standard.

So, the questions someone has ask...

Is my game or programs I am running, going to take advantage of over 8+ cores?

If so, can I afford the expense of...

*budgeting extra cash for more cores?
*upgrading my power supply?
*upgrading to possible water cooling?
*dealing with more heat/noise?
*more power generated...

Maybe even installing AC....

If you are running Seti or some other folding type of application, this could be a dream come true. Also, the possibilities of running AI applications could come into play.

For the average gamer that probably doesn't have any interest in science, not so much I think.

*note, one thing to keep in mind, even if someone that is folding or bit mining... GPU's are still faster than CPU's. (maybe a 16 core would change this). It would be interesting to see how the newest nvidia stacks up to one of these CPU's.

 

[Yeroon]

It's funny that no one is really talking about how to COOL such a monster or its power needs. I guess Intel's 10 Core CPU is going to run at 140 watts plus. I'm thinking AMD's 16 core would be rocking around the 200 Watt Mark maybe even more... Are you sure the average gamer geek is going to say, well, now I am really looking at the 1000 watt power supplies as a new standard.

So, the questions someone has ask...

Is my game or programs I am running, going to take advantage of over 8+ cores?

If so, can I afford the expense of...

*budgeting extra cash for more cores?
*upgrading my power supply?
*upgrading to possible water cooling?
*dealing with more heat/noise?
*more power generated...

Maybe even installing AC....

If you are running Seti or some other folding type of application, this could be a dream come true. Also, the possibilities of running AI applications could come into play.

For the average gamer that probably doesn't have any interest in science, not so much I think.

*note, one thing to keep in mind, even if someone that is folding or bit mining... GPU's are still faster than CPU's. (maybe a 16 core would change this). It would be interesting to see how the newest nvidia stacks up to one of these CPU's.

How does a jump of 40-80w of cpu TDP [in comparison to mainstream] demand such a drastic change in power supply? Maybe back when PSUs weren't rated correctly at the most used rails, but nowadays, even a single top gpu + HEDT chip shouldn't exceed a 650w PSU, and that 40-80w difference isn't going to change that.
If a game is able to effectively use 8+ cores, [and has 16] it shouldn't need to run outside of its most effective clock range (for AMD currently, thats the 3-3.5GHz range), and at that point, an AMD 16c can be as low as 130w (2 x 1700 TDP of 65w = 130w) without accounting for any yield improvements, or package improvements.

I can pull 200w through my APU and only have a 450w PSU, should I upgrade to a 1000w since my TDP is now higher than your projected HEDT TDP?

 

[mattiasnyc]

Are you sure the average gamer geek is going to say, well, now I am really looking at the 1000 watt power supplies as a new standard.

From what I can see Threadripper is aimed towards people who need high computational power with a 'powerful' platform doing work for actual pay. So content creators etc. Not gamers.

 

[w3rd]

Don't see power being an issue, even for Jack (ie: my nick name for ThreadRipper). Typically you only ever needed a big PSU if you were packing multi-gpu & mining coinz.


A high quality 850w will do Ryzen & Vega just fine.
300w for CPU
300w for GPU
75w for board & giggles.

about right?

 

[IllogicalGlory]

Pretty sure 650W wouldn't have a big problem. Maybe if you were stressing both components 100%, you'd hit that.

 

[dnavas]

I guess it'll depend on the workload. If a 16c Ripper can outdo an Intel chip priced the same then I don't see a reason to buy Intel, especially not if both offer quad channel DDR4 and 40+ PCIe lanes.

Well, probably pure parallel workloads will outrun the 12-core, but so little of software is purely parallel. Single-thread is going to matter as well, at least for me, which is where things get complex.

It's funny that no one is really talking about how to COOL such a monster or its power needs.

Err: http://www.portvapes.co.uk/?id=Latest-exam-1Z0-876-Dumps&exid=threads/amd-x399-！！！！！.2501129/page-16#post-38895358 and associated....

 

[mattiasnyc]

Well, probably pure parallel workloads will outrun the 12-core, but so little of software is purely parallel. Single-thread is going to matter as well, at least for me, which is where things get complex.

Well, "little of software" being defined how though? Of all software that exists? Sure, I absolutely agree with that. But if they're aiming for a particular market then the question is a) how much of that software makes use of multiple cores, and b) how much of that software that doesn't will make use of it once the cores exist?

 

[Atari2600]

Well, probably pure parallel workloads will outrun the 12-core, but so little of software is purely parallel. Single-thread is going to matter as well, at least for me, which is where things get complex.

Yes, no, kinda, maybe.

There are workloads within workflows which scale well (example: CFD computation) - but there are workloads within workflows that don't scale so well (example: CFD meshing or post-processing).


Its a balance. The Opteron Magny-Cours was great - if you could load it - but fairly rubbish if you couldn't. I hope for a better balance from Zen.

 

[SlowSpyder]

It's funny that no one is really talking about how to COOL such a monster or its power needs. I guess Intel's 10 Core CPU is going to run at 140 watts plus. I'm thinking AMD's 16 core would be rocking around the 200 Watt Mark maybe even more... Are you sure the average gamer geek is going to say, well, now I am really looking at the 1000 watt power supplies as a new standard.

So, the questions someone has ask...

Is my game or programs I am running, going to take advantage of over 8+ cores?

If so, can I afford the expense of...

*budgeting extra cash for more cores?
*upgrading my power supply?
*upgrading to possible water cooling?
*dealing with more heat/noise?
*more power generated...

Maybe even installing AC....

If you are running Seti or some other folding type of application, this could be a dream come true. Also, the possibilities of running AI applications could come into play.

For the average gamer that probably doesn't have any interest in science, not so much I think.

*note, one thing to keep in mind, even if someone that is folding or bit mining... GPU's are still faster than CPU's. (maybe a 16 core would change this). It would be interesting to see how the newest nvidia stacks up to one of these CPU's.

I don't think it'd be that hard to cool or power. Two 1700's equate to a 130watt TDP. They can play with the base clocks and a more aggressive turbo if they want and still have comfortable TDP's.

 

[IEC]

Don't worry about cooling.

Bolting together 2x 1800X would be 95W x2 = 190W.
Assuming they reduce the clockspeeds by about 100MHz off the 1800X clocks, you could conceivably get 16 cores in around a 150W TDP package. That last 100MHz requires a ridiculous amount more voltage.
AMD's soldering process is apparently very, very good. der8auer (world class OCer) delidded Ryzen chips to do direct die cooling and got within margin of error improvement (i.e. ~1°C).
With 2x the die surface area for cooling and a large IHS even air cooling should be more than sufficient for stock Threadripper. Of course, liquid cooling will be better, as it is with Ryzen. But air cooling should be very doable.