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

[Mockingbird]

Has it been announced what is the market name for Threadripper?

It is also going be be "Ryzen"?

 

[.vodka]

Not yet. There's only this slide with both Ryzen and Threadripper on it.

Maybe we get some more on this along with Vega on Computex. 10 days to go!

 

[mattiasnyc]

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),

The 6900k is slower than the 6850k, and so if it beats the latter it has to be because of the added cores, which in turn means that more cores are better. I looked at Puget's review of the 1800X compared to Intel CPUs, and that's what recall seeing.

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.

Yeah, but why 3.1GHz, and why compare to 4.3GHz? It looks like you're picking the 'worst' case scenario for TR and a best one for Intel.

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: ]

I think people who only care about getting 3x the performance of an 1800x if the 16c TR is 3x the price is missing the point. AMD didn't have to create TR at all, period, if it wasn't for the quad-channel DDR4, more lanes etc. That's the only reason for its existence. If it wasn't that, then why not just update Ryzen with the same socket? Or vice versa; if it didn't matter why not release Ryzen 7 on x399?

And yes, I understand it's a partially artificially segmented market, but it is what it is. And I actually think it could be detrimental for AMD to 'under'-price TR. People have perceptions about quality partially based on price. The notion of buying a cheap CPU that seems "cheap" isn't appealing.

 

[dnavas]

Yeah, but why 3.1GHz, and why compare to 4.3GHz? It looks like you're picking the 'worst' case scenario for TR and a best one for Intel.

3.1 was Glo's argument:

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.

I think it missed the mark, that's all.

I actually think it could be detrimental for AMD to 'under'-price TR. People have perceptions about quality partially based on price. The notion of buying a cheap CPU that seems "cheap" isn't appealing.

I don't doubt that there will be a certain amount of, shall we say, "halo pricing" around the performance king. The glow will come off the 1800X and shine on whatever the high end TR version will be. That's as it should be. Until we know where Intel's pricing and performance lands, it's a guessing game as to how much pricing room there'll be. AMD had a lot of room with their 8-core, but with SkylakeX shipping earlier, they may have less ... depending. Hopefully we'll know in a couple of weeks.

 

[Mockingbird]

Not yet. There's only this slide with both Ryzen and Threadripper on it.

I am definitely excite about Zen. It looks like AMD is on the way to a home run.

Hopefully AMD say something about Raven Ridge (for laptops).

Maybe we get some more on this along with Vega on Computex. 10 days to go!

Remember, over a month ago, when I said a lot of about Vega?

Namely that:

A) Vega has been delayed because B) HBM2 is too expensive and that C) Vega is liquid cooled and D) is 10% faster the Geforce GTX 1080.

A lot of people called my a liar and a troll.

If you haven't noticed, they have all gone silent.

 

[jpiniero]

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.

The hit may not be significant but not irrelevant either. Enough that if it loses to the 1800X in most cases I'd have to think that would kill the appeal.

 

[krumme]

The hit may not be significant but not irrelevant either. Enough that if it loses to the 1800X in most cases I'd have to think that would kill the appeal.

Either way its a matter of price. The cost benefit of small cores vs big is enourmous. They can sell 16c at 800 usd and still have fantastic margins. They really have the upper hand here by far.

Ryzen is extremely cheap. R9 will be the same or it will not sell. Read my lips. 799 for cheapest 16c.

 

[Tup3x]

If they can make 8 core cpu and sell it for ~350 € (inc. 24% VAT) and still make money, I don't see how 1300 € pricing would make sense at all for 16 core version. Heck, that should be cheaper to produce than 2x Ryzen 1700 CPUs.

999 € for 16 core threadripper and 799 € for 12 threadripper. High clock and low clock variants may be cheaper or a bit pricier. Though even 999 € price has truckloads of air.

 

[T1beriu]

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:

You post an Ryzen 9 image made up by that junk site that makes up images all the time and you call this proof? Can't you see how bad it is? I can find you 20+ made up images like that from the last year alone.

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).

1. That's a dual socket EPYC schematic. I don't know what you're trying to prove with it. I counted more than 44 PCIe lanes per socket.

2. Who said Threadripper is EPYC?

Has it been announced what is the market name for Threadripper?

It is also going be be "Ryzen"?

AMD confirmed that is TR is the brand name. So no Ryzen 9.

w3rd, here you go. Another proof that you should stop visiting that junk site that treats its readers as drug junkies and gives them anything to keep them going (back to their site).

 

[Topweasel]

You post an Ryzen 9 image made up by that junk site that makes up images all the time and you call this proof? Can't you see how bad it is? I can find you 20+ made up images like that from the last year alone.



1. That's a dual socket EPYC schematic. I don't know what you're trying to prove with it. I counted more than 44 PCIe lanes per socket.

2. Who said Threadripper is EPYC?



AMD confirmed that is TR is the brand name. So no Ryzen 9.

w3rd, here you go. Another proof that you should stop visiting that junk site that treats its readers as drug junkies and gives them anything to keep them going (back to their site).

It probably is just ThreadRipper, but it could be like Vega. Could be called Ryzen R9 ThreadRipper. Being called ThreadRipper doesn't preclude it from using the R modifier. Remember the slide they used had Ryzen above ThreadRipper so unlike EPYC, ThreadRipper is part of the Ryzen family.

 

[T1beriu]

Being called ThreadRipper doesn't preclude it from using the R modifier.

What is this "R" you're talking about? You must be confusing the Radeon "R" 200/300 series branding with the CPUs.

 

[Topweasel]

What is this "R" you're talking about? You must be confusing the Radeon "R" 200/300 series branding with the CPUs.

Ryzen R3, Ryzen R5, Ryzen R7. I mean I am not an engineer for AMD but I am like 98% sure I own a Ryzen R7 1700.

Sent from my Nexus 6 using Tapatalk

 

[T1beriu]

Sure you do.

I dare you to find R3/R5/R7 naming on any official AMD channel.

LE: Your answer reminds me of this.

 

[Puffnstuff]

I see a flaw in the performance comparison between R7 and R9 in that they are different architectures. R9 has over 4k pins which should increase throughput over the AM4 R7 but we won't know for certain until it becomes available for testing.

 

[wildhorse2k]

16C and 4 channel memory controller are nothing exceptional.

Look at this beauty:

Naples 32C + 8 channel memory

That will be many people's dream.

 

[Topweasel]

16C and 4 channel memory controller are nothing exceptional.

Look at this beauty:

Naples 32C + 8 channel memory

That will be many people's dream.

You mean wet dream right? Because that's what it does for me.

64 Threads and 256GB of memory for a workstation does something dirty to me.

 

[mattiasnyc]

3.1 was Glo's argument:

I think it missed the mark, that's all.

Oh, sorry, I missed that somehow. I understand and agree then.

Until we know where Intel's pricing and performance lands, it's a guessing game as to how much pricing room there'll be. AMD had a lot of room with their 8-core, but with SkylakeX shipping earlier, they may have less ... depending. Hopefully we'll know in a couple of weeks.

Yep.

 

[Kenmitch]

Looks like an interesting product line too me. Will be interesting to read the reviews once they drop.

It's amazing that after being comatose for a decade AMD seems to be waking up without severe brain damage....Guess the last part doesn't really apply to the marketing department. Maybe they're just still on the wrong meds?

 

[Gideon]

First Benches apparently:

[source: reddit]

 

[pcslookout]

The question is will it help Firefox performance any ?

 

[krumme]

First Benches apparently:

[source: reddit]

Ok if this is the cb score Intel hedt line is ripped apart.

 

[Gideon]

Ok if this is the cb score Intel hedt line is ripped apart.

Threadripped™ apart

... Ok, I'll show myself out

 

[mikk]

I bet it's another fake, as usual. It's from reddit, and even worse a leak for AMD, the risk for a fake is very high therefore.

 

[dnavas]

Either way its a matter of price. The cost benefit of small cores vs big is enourmous. They can sell 16c at 800 usd and still have fantastic margins.

If AMD follows R7 launch and has both a "cheap" and "halo" product which are essentially identical once OC'd (1700 vs 1800X) I could see both pricing extremes come true. 2.5x 1700 is ~$800, while 2.5x 1800x is $1250. I expect there will be additional features that will increase the pricing on both, but that's just an assumption. At anyrate, the pricing spread would nicely cover the value segment and not abandon the benefits of a higher-priced halo sku.

Different question -- what would be the likely speeds?
If we believe that the TR stepping(s) provide no tdp improvements, 3G and something like 3.4G seem about right. I'm assuming a loss of ~10W in the MCM (which seems high to me, but I have zero experience, so it's just a round-number to start with). That leaves 170W for the cores, or about 85W each. If we generally buy the premise that the 1800X runs about 10W higher than the tdp numbers might indicate, then we could see our way to pretending that the 1700X is an 85W part, thus the 3.4G. Twice 1700, meanwhile, is 130W plus 10W for the MCM, and now it looks like ~SkylakeX tdps. Thus 3.0 and 3.4.

The interesting question, in my mind, is what improvements might have been made. Given the curve in TheStilt's tech posting, it seems unlikely that much headway could be made in top-end frequency, but fixing the two (or I think three, as it looks like there's another bump at 3.7G) critical volt/freq response changes might reduce required voltage by 10% or so. That would probably allow a dual-1800X to fit the 180W tdp ceiling. It would also allow linear OC up to 3.9, which would be quite nice (long way from 4.5, but an impressive first step at anyrate). Has anyone heard whether we might be getting an improved stepping in TR, and what those improvements might entail?

 

[mikeymikec]

16C and 4 channel memory controller are nothing exceptional.

Look at this beauty:

Naples 32C + 8 channel memory

That will be many people's dream.

That's bigger than full-size ATX isn't it? IIRC most "ATX" boards don't go full size ATX just like most mATX boards I see aren't the maximum size allowed by mATX specs.