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

[w3rd]

Hey, explain again why there won't be 10c, 12c, 14c ThreadRippers...?

As far as we know there will be 8 or 9 skus for threadripper HEDT. It seems kind of ignorant to suggest there will not be any 10core, or 14 core skus. There are also rumors of the 16C starting at $999.

There is your HEDT platforms for late 2017.

 

[T1beriu]

Anything posted on the internet is real these days? It screams fake from a mile away.

Ignoring this is a table that anyone could have created, ignoring the stupid naming scheme, ignoring the incredibly large number of SKUs, ignoring the large number of TDP tiers, that official speed memory support tops at quad channel 3200Mhz (when Ryzen starts from dual-2666 & EPYC 8x2400, really? ), but the number of PCI lanes should signal clearly fake. TR is half of EPYC, that has 128 PCIe lanes, so TR should have 64.

LE: These will not be named Ryzen 9 because these CPUs are going to be called ThreadRipper.

Look for Ryzen Strictly Technical thread to find out why we won't see 10/14C.

There is your HEDT platforms for late 2017.

AMD said TR is coming summer 2017.

 

[Topweasel]

Hey, explain again why there won't be 10c, 12c, 14c ThreadRippers...?

As far as we know there will be 8 or 9 skus for threadripper HEDT. It seems kind of ignorant to suggest there will not be any 10core, or 14 core skus. There are also rumors of the 16C starting at $999.

There is your HEDT platforms for late 2017.

Supposedly their is something in the white papers that says that they are requiring CCX's to match across dies. On top of that something about the PSP not allowing mismatched configuration.

I looked at the document someone suggested and didn't find the portion covering the topic they suggested we look at. Nor any reference to limitations on core configuration at all. But I also didn't have hours to read it page by page.

So basically there are two or three people that say it over and over again and that's all I have seen. Even though we know it's possible and that AMD has created ES's with mismatched configurations while deciding what to ship for the R5 range.

Sent from my Pixel XL using Tapatalk

 

[Malogeek]

Anything posted on the internet is real these days? It screams fake from a mile away

Especially given the source of the image....

 

[Topweasel]

I don't think that's going to happen because:

1. $499 is not an ultra-premium segment that AMD said is going for.

2. There are not going to be 10 core TR. Or 14. Because that's impossible. There are 4 or 5 threads with multiple pages where this has debunked. I think even this one has 2 pages about this.



Hmmm.... No they did not? They said this is for ultra-premium HEDT. No mention of gaming. You can watch the launch again here. TR starts ar 28:20.



I'm pretty sure most HEDT CPUs get bought by digital content creators (games, VR, video, animators, image visualisation etc) than gamers, at least that's what I'm seeing.

I sure on the prosumer end people are using those systems for work and games. But their is better support and better cost on the Xeon workstation end for these types of jobs. HEDT has always been aimed at the multi-use, multi-tasked, extreme gaming type.

 

[Glo.]

Hey, explain again why there won't be 10c, 12c, 14c ThreadRippers...?

As far as we know there will be 8 or 9 skus for threadripper HEDT. It seems kind of ignorant to suggest there will not be any 10core, or 14 core skus. There are also rumors of the 16C starting at $999.

There is your HEDT platforms for late 2017.

It is because there is not possible to have 5 or 7 core Ryzen CPUs.

Only configs possible for each die, are 4, 6 or 8 setups. ThreadRipper brand is made from dual die setup. 8 Core ThreadRipper is pointless, apart from higher amount of PCIe lanes, and quad channel memory.
12 and 16 core, are the ones which we will see.

No, do not expect 16 core for less than 1000$. I personally expect 1499$ for top end ThreadRipper CPU.
750-800$ for base 12 core CPU.

 

[jpiniero]

No, do not expect 16 core for less than 1000$. I personally expect 1499$ for top end ThreadRipper CPU.
750-800$ for base 12 core CPU.

I don't think $1499 is going to fly. $999 is good for the top model but we'll have to see what Intel does.

TBH AMD really should only do two models - the 12 and the 16. Maybe they could do a third 12C model with cut lanes or something. In case you haven't noticed, Ryzen sales have pretty much consolidated around the 1600 and 1700, understandably given that there's not much of a benefit to the other 6/8C models plus you also get a cooler. Same thing would happen if AMD did the same thing with Threadripper.

 

[T1beriu]

I sure on the prosumer end people are using those systems for work and games. But their is better support and better cost on the Xeon workstation end for these types of jobs. HEDT has always been aimed at the multi-use, multi-tasked, extreme gaming type.

You're just arguing with yourself, you know that?

they specifically stated that ThreadRipper was for ultra high-end gaming systems (not true - edit by T1beriu)

in the end ThreadRipper is a gaming CPU

Especially given the source of the image....

Oh, yeah, good point. The source is that junk site that is known for making things up as facts.

 

[Glo.]

I don't think $1499 is going to fly. $999 is good for the top model but we'll have to see what Intel does.

TBH AMD really should only do two models - the 12 and the 16. Maybe they could do a third 12C model with cut lanes or something. In case you haven't noticed, Ryzen sales have pretty much consolidated around the 1600 and 1700, understandably given that there's not much of a benefit to the other 6/8C models plus you also get a cooler. Same thing would happen if AMD did the same thing with Threadripper.

Erm, that is the point people have trouble with understanding. ThreadRipper is not cheapo platform. It is high-end, high margin, high price, enthusiast platform. 1499$ is appropriate amount of money for 16 core/32T, 3.1 GHz, 180W CPU.

Lets get a little perspective. Intel will top out their HEDT platform at 12 C. That CPU will cost 1723$, most likely.

AMD can price their 12C CPU at 800$. How do you think, what people will pick in this scheme of things? AMD has the advantage of higher core count. They can sell it at 1499$.

Think about this. Package is made from two dies, that cost AMD to make around 30$ per die. Total package, with assembly will lock up in what cost, 150$? That is huge margin AMD can earn.

 

[dnavas]

Erm, that is the point people have trouble with understanding. ThreadRipper is not cheapo platform. It is high-end, high margin, high price, enthusiast platform. 1499$ is appropriate amount of money for 16 core/32T, 3.1 GHz, 180W CPU.

There are two kinds of people who are in the market for these. The first already have BD-E and paid those kinds of prices. The second don't because they refuse to pay/can't afford those prices. There's roughly a 10% difference in performance between a 16 core 3.1G Ryzen and an OC'd 10 core 6950X. $1500 is a ridiculous amount of money to pay for a 10% bump (mho). It's even sillier if you assume that the 12-core Skylake with better clocks, and IPC will sell at $1700. Presumably there will be OC headroom of some kind for Ryzen, and maybe $1500 becomes worth it for some subset of the market once you start seeing performance advantage between the likely 4.5G 12 core and the Ryzen processor. If TR could OC to 4.2+, $1500 would seem reasonable to me, but different apps will respond differently, so mileage will vary.

 

[postmortemIA]

One core 4C/8T i7 is about $330, multiply that by 4 you get $1300.. plus you'd need something that can take 4 physical CPUs... AMD would be nuts to charge less than $1400 for TR

 

[Paul98]

I am interested in what the performance is like when using multi-die communication with the infinity fabric. Any information on that yet?

 

[Puffnstuff]

I want to see some legit benchmarks using real world conditions by regular people rather than cherry picked instances that might paint a skewed picture of performance and is applicable to both I9 and R9.

 

[Glo.]

There are two kinds of people who are in the market for these. The first already have BD-E and paid those kinds of prices. The second don't because they refuse to pay/can't afford those prices. There's roughly a 10% difference in performance between a 16 core 3.1G Ryzen and an OC'd 10 core 6950X. $1500 is a ridiculous amount of money to pay for a 10% bump (mho). It's even sillier if you assume that the 12-core Skylake with better clocks, and IPC will sell at $1700. Presumably there will be OC headroom of some kind for Ryzen, and maybe $1500 becomes worth it for some subset of the market once you start seeing performance advantage between the likely 4.5G 12 core and the Ryzen processor. If TR could OC to 4.2+, $1500 would seem reasonable to me, but different apps will respond differently, so mileage will vary.

Current pricing structure of Intel CPUs is obvious. Everybody expected with BDW-E that Intel will bring 10 cores to 1000$ price tag. It did not happen. 8 cores went up by 50$, and HEDT got another price tag: 1723$ for 10 cores.

People who could not afford to pay, for HEDT will never pay those prices. Thankfully for them, there is Ryzen lineup, which brought HEDT to Mainstream.

People right now expect AMD to price their HEDT CPU line as a bargain. Fine. But that is actually pretty unreal considering how HEDT CPUs are priced, and what they offer.

 

[mattiasnyc]

I don't think $1499 is going to fly. $999 is good for the top model but we'll have to see what Intel does.

TBH AMD really should only do two models - the 12 and the 16. Maybe they could do a third 12C model with cut lanes or something. In case you haven't noticed, Ryzen sales have pretty much consolidated around the 1600 and 1700, understandably given that there's not much of a benefit to the other 6/8C models plus you also get a cooler. Same thing would happen if AMD did the same thing with Threadripper.

I still don't see how they can sell a 16c Ripper for only $k. It just seems far too low compared to Intel.

I also think the smarter thing is to do 8, 12 and 16 cores (if those are the only possible permutations). The 8 core is still desired by many (me) because the core count being 'lower' keeps cost down yet still gives good performance, but more importantly Ripper adds the rest of the platform, i.e. 44 lanes and quad-channel DDR4.

So to me it would probably make more sense to have an 8 core at the low end (Ripper) for about $600 or so, a 12 core at around $900, and then top out at 16 cores for around $1400. That to me would make sense more or less.

 

[IEC]

Comparing to Intel would be the wrong move. Intel still has the lead in IPC and fMax on their chips, so it would make sense that AMD would at minimum offer more cores at each price point a la Ryzen. This could mean that $1000 may not be too far off for at least one 16c/32t SKU. The top binning may cost more, however.

 

[mattiasnyc]

There are two kinds of people who are in the market for these. The first already have BD-E and paid those kinds of prices. The second don't because they refuse to pay/can't afford those prices. There's roughly a 10% difference in performance between a 16 core 3.1G Ryzen and an OC'd 10 core 6950X. $1500 is a ridiculous amount of money to pay for a 10% bump (mho).

Ok, but do we know that that's going to be the clock speed and performance gap though? I mean, I've seen some benchmarks for content creation (Premiere Pro) where the 1800x runs extremely close to the 6900k, and beats the 6850k, so I have a very hard time seeing a 16 core chip only managing to beat the 6950k by 10%.

 

[jpiniero]

The 8 core is still desired by many (me) because the core count being 'lower' keeps cost down yet still gives good performance, but more importantly Ripper adds the rest of the platform, i.e. 44 lanes and quad-channel DDR4.

The problem with the 8 core is that there's going to be some amount of a hit due to cross-die traffic. Tough to say if QCM would be able to make up for it compared to the single die models OCed.

 

[krumme]

Papermaster claimed that on spec int ryzen was scaling perfectly across dies and even sockets.

Obviously it doesnt but imo it hint to the posibility that the drawbacks is probably irrelevant.

I remember the amd marketing of true quad core and the intel solution was in that respect nearly just as good in practical situations.

I take Papermaster word as solid here.

 

[maddie]

Agreed, this is what they claim.

 

[Puffnstuff]

Agreed, this is what they claim.

This is exactly why we need independent testing results to verify and substantiate any performance claims made.

 

[Elixer]

This is exactly why we need independent testing results to verify and substantiate any performance claims made.

That is true of all marketing slides.

Trying to find a fair, impartial person to verify is the tricky part.

 

[w3rd]

Anything posted on the internet is real these days? It screams fake from a mile away.

Ignoring this is a table that anyone could have created, ignoring the stupid naming scheme, ignoring the incredibly large number of SKUs, ignoring the large number of TDP tiers, that official speed memory support tops at quad channel 3200Mhz (when Ryzen starts from dual-2666 & EPYC 8x2400, really? ), but the number of PCI lanes should signal clearly fake. TR is half of EPYC, that has 128 PCIe lanes, so TR should have 64.

LE: These will not be named Ryzen 9 because these CPUs are going to be called ThreadRipper.

Look for Ryzen Strictly Technical thread to find out why we won't see 10/14C.

AMD said TR is coming summer 2017.

So, what you are saying is that you are just choosing to arbitrarily dismiss all of it & facetiously pretend NONE of it is true.?
lol..


Here:

And here:

Study closely^.
Not sure what you are beating your head over and trying to understand. Or even what you are rambling on about. Also, Threadripper is not EPYC, it is Ryzen9 (R9).

 

[IEC]

I looked closely, and WCCFTech is as usual throwing unverified rumor out there as fact. The fact that they are pushing the rumor of 10-core and 14-core Threadrippers alone should tell you they're off mark.

They also change their (wrong) speculation articles after details are released to make it look like they were right all along.

Don't give them clicks and don't post their garbage here.

 

[dnavas]

Ok, but do we know that that's going to be the clock speed and performance gap though?

Nope, we have no idea. We don't know what the clockspeed is going to be, we don't know what Intel's pricing will be, we don't know how well the 12-core will scale. Nor AMD's MCM setup. We don't know what extras Intel will ship with, and we don't know what extras AMD may have in store.

I mean, I've seen some benchmarks for content creation (Premiere Pro) where the 1800x runs extremely close to the 6900k, and beats the 6850k, so I have a very hard time seeing a 16 core chip only managing to beat the 6950k by 10%.

Maybe, but the Premiere tests I've seen show the 6900k with a clear lead given many actions don't scale particularly well (the multithread-friendly benches are close, but don't comprise a clear majority of the work), and to be clear, the comparison was to a 3.1Ghz processor, not the 1800X. If we assume all other things equal, and the IPC is the same, (3.1 * 16 - 4.3 * 10) / 3.1 * 16 = 15%. I knocked 5% off for IPC diff. :shrug: That's not to say that's a fair assessment of where the product will actually land - it's a well-OC'd 6950k compared to an assumed stock 3.1Ghz. I was merely pointing out that 3.1Ghz wouldn't really be worth much *by itself*. If it ships @3.1, but can OC to 4, that's a whole other story.

But, let's ignore Intel for a paragraph. If the 16 core is double an 1800X, it's basically giving $1000 perf (2 * 1800X) + 12pcie lanes. At that point, it's obviously worth more than $1000, but not so obviously worth $1500 (it isn't giving me 3x an 1800X even in just bandwidth, nevermind compute throughput). [I realize that Intel hasn't priced its HEDT in any kind of rational perf/$ manner, but, y'know, that's kind of why I don't own one of those chips, and if AMD follows suit, well, I won't own theirs either :shrug: ]

What makes this story even less compelling for AMD is that anyone who already is happy forking over $1500, and is sitting there with a 6950, has to think not only about whether the 16 core is worth the upgrade, but how it compares to the competing Skylake 12 core. Now, Intel could double-down on their high-price strategy, leave their 8-core at $1k, leave their 10-core at $1700, and put their 12-core at $2500. That, I think, leaves a lot of room for AMD to price at $1500. But I have a hard time believing that would happen, for exactly the reason you mention above. At the end of the day, for multithreaded apps, the 6900k is ~= a $500 Ryzen processor. Now, it almost certainly will overclock better, and the Skylake version will have better IPC, but I have a hard time envisioning it as selling for more than $600. That logically pulls down the 10 core to the $1000 range, and then maybe they can price the 12-core at $1700. If Skylake has 15% higher IPC than Ryzen, and can stable OC to 4.5, the 16-core Ryzen would have to OC to 3.9 to be equivalent for multi-threaded, and would still lose for any poorly-threaded code. It's an uphill battle being fought here....

It can come down to other features of the processor, though. There's been a number of rumors regarding 10GBE, and if the processor comes with a few 10GBE ports, that itself might be worth a couple of hundred. I mean, the first is almost certainly worth a G, maybe even two, but it's unclear if most of us would be able to take advantage of more than that (server, yes, HEDT?). Are there other unique features? Stick a few graphics pipelines on the thing and stuff them behind something that supports software already using Intel's Quicksync and I'd be more than ready to throw two grand at them. Low-precision matrix multiplies for machine learning? 128bit floating point math for high-end physics? It's niche stuff, but valuable for segments of the market. AMD keeps mentioning their custom hardware capabilities, maybe it's time to heat up the market and start using it? The really awesome thing about using the Naples socket for HEDT is that two 1800Xs take up only half the total available area. At some point, someone is going to think to stick other stuff in there....

I *don't* think we're getting Ripper-16 for $800, but, if we believe that they want to repeat their "same perf for half the money" story, it seems reasonable to expect the 12-core to slot right about there. I still think the most logical pricing is 6/9/12 for 8/12/16 TR cores, with the 1800X getting a $50 haircut, but, I've been wrong before, and lord knows I'll be wrong again soon enough, so, we'll just have to wait and see!