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

[IntelUser2000]

exactly linear scaling???.... I'm not convinced that's true.

It's believable. Scaling for 1600X to 1800X is 98% on extra cores. You'd have ever slightly higher perf/clock on lower clocks, hence whatever the miniscule scaling loss is made up by increased perf/clock with lower clocks.

Now that's if the clocks are true.

 

[mattiasnyc]

It's believable. Scaling for 1600X to 1800X is 98% on extra cores.

Right, that's not exactly the same, it's 98%, and it's going from six cores on one die to eight cores on one die, not from eight cores on one die to sixteen on two dies. If the scaling was 98% it would have resulted in a score of 3021.732, not 3083.

You'd have ever slightly higher perf/clock on lower clocks, hence whatever the miniscule scaling loss is made up by increased perf/clock with lower clocks.

I don't understand how that works. How does IPC not scale linearly with frequency?

 

[aigomorla]

when are we suposed to get ECC support with ryzen on boards?

I really think they would make perfect servers if they ran ECC REG ram.

 

[scannall]

when are we suposed to get ECC support with ryzen on boards?

I really think they would make perfect servers if they ran ECC REG ram.

While Ryzen supports ECC Ram, the motherboard manufacturers haven't set it up or validated it YET. They may or may not. It's a wait and see kinda thing.

 

[IEC]

when are we suposed to get ECC support with ryzen on boards?

I really think they would make perfect servers if they ran ECC REG ram.

There is no "official" support for ECC on X370, though Wendell @ Level1Techs has found that it appears to work/mostly work with certain boards such as the ASRock Taichi and Gigabyte K7. Read the HardwareCanucks' article for more details:
http://www.hardwarecanucks.com/foru...ws/75030-ecc-memory-amds-ryzen-deep-dive.html

Official ECC support will likely come on the "Threadripper" HEDT and Epyc server platforms.

 

[formulav8]

12 and 16 core configs are possible

I could see an entry level 8 core cpu. Basically 8/12/16 core options for TR. An 8 core cpu with more pci-express lanes, 4 channel, etc. would be the sweet spot for many depending on price, especially the increased IO potential. But with the over 80% yield success rates I read for Ryzen wouldn't leave a large number for 8 core cpu's for the 4k pin setup.

 

[formulav8]

work with certain boards

Yeah AMD said it's up to the MB manufacturer's to support the feature.

 

[mattiasnyc]

I could see an entry level 8 core cpu. Basically 8/12/16 core options for TR. An 8 core cpu with more pci-express lanes, 4 channel, etc. would be the sweet spot for many depending on price, especially the increased IO potential. But with the over 80% yield success rates I read for Ryzen wouldn't leave a large number for 8 core cpu's for the 4k pin setup.

Also what I was thinking. With an 8core TR part one would have an upgrade path where one can just swap the CPU.

 

[cbn]

What about people that need a lot of IO and memory, but not as much processing power? An 8 core, quad channel, 44 PCIe lane system.

We get 8, 12 and 16 core options as feasible. They should still make more than selling 4 core 1400 and 1500X die.

I would like to think they would but I should note that it can't be a Ripper of Threads if it's only got 8 cores.

An eight core should still benefit from the quad channel though (re: AM4 8C/16T benefits from overclocked dual channel DDR4). This especially if slow DDR4 memory (commonly found in consumer pre-builts and workstations) is used at the starting point.

P.S. I think a processor made via two Raven Ridge (iGPU disabled) dies MCM together would be interesting. Less PCIe lanes and cache than using 8C/16T dies MCM as the starting point though.

 

[Topweasel]

An eight core should still benefit from the quad channel though (re: AM4 8C/16T benefits from overclocked dual channel DDR4). This especially if slow DDR4 memory (commonly found in consumer pre-builts and workstations) is used at the starting point.

P.S. I think a processor made via two Raven Ridge (iGPU disabled) dies MCM together would be interesting. Less PCIe lanes and cache than using 8C/16T dies MCM as the starting point though.

It was more if a joke than anything. Like how could it be a ThreadRipper if it's only got the cores of a regular Ryzen.

 

[cbn]

I could see an entry level 8 core cpu. Basically 8/12/16 core options for TR. An 8 core cpu with more pci-express lanes, 4 channel, etc. would be the sweet spot for many depending on price, especially the increased IO potential. But with the over 80% yield success rates I read for Ryzen wouldn't leave a large number for 8 core cpu's for the 4k pin setup.

How about using two Raven Ridge dies (with defect in the iGPU) MCM for a 8C/16T quad channel processor?

If that is workable more 8C/16T dual channel dies could be allocated to the X399 platform (as well as be made into 12C/24T and 16C/32T processors for X390)

 

[formulav8]

How about using two Raven Ridge dies

The RR core would be more so for an AM4 only APU setup, not the 4k pin package. So I couldn't see an 8 core 4k pin cpu that way. A bad iGpu RR cpu would most likely be a discreet AM4 cpu marketed as an R3 based sku.

 

[cbn]

The RR core would be more so for an AM4 only APU setup, not the 4k pin package. So I couldn't see an 8 core 4k pin cpu that way. A bad iGpu RR cpu would most likely be a discreet AM4 cpu marketed as an R3 based sku.

Yep, traditionally the iGPU defect APU dies become Athlon x 4 processors (eg, Athlon x 4 860K).

But I wonder how useful that would really be? (re: refurb hardware with good Workstation CPUs are getting cheaper all the time. Furthermore, new pre-built desktops with Core i5 7400 can be really cheap. See this example at $399.99. It has gone on and off sale for that price and below at numerous retailers. See this thread (post #6, post #17, post #18, post #19, post #22, post #23, post #24 (just added today)) for a partial history of that)

 

[maddie]

It was more if a joke than anything. Like how could it be a ThreadRipper if it's only got the cores of a regular Ryzen.

To think that several months ago, 16 threads would have been seen as a threadripper CPU. It's amazing how the landscape has changed with the arrival of Ryzen, that we now consider 16 threads as quite normal.

 

[T1beriu]

So, what you are saying above is that everyone is right... you just want to argue with them?

#1: It is true, because there is a Ryzen 9.
#2: You clearly stated there will be no Ryzen 9.
#3: NO, AMD Ryzen 9 (1900x ~ 1950x, or whatever).
#4: Your bias, because you do not give AMD a chance, or objectiveness..
#5: Again, ThreadRipper is the name of the HEDT platform, your link is only speculation.
#6 You have media bias, unable to accept information, unless it comes from approved site. (laugh)
#7: Again, another "junk site" comment...?
#8: In what posts, have you thoroughly dismembered the Ryzen 9 sku..? And conversely, do understand I am not arguing with you... I stated things that you went ballistic over and want to argue about. I am only showing/telling/responding to your bias questions. Because the argument is real simple: Ryzen 9


Bro, look back at your posts. The only thing going on here is you learning about the AMD brand names.
i3 = R3
i5 = R5
i7 = R7
i9 = R9 HEDT

If you do not understand that little bit of marketing, there is no hope for you. Seems like you are being purposely obtuse.

I'm sorry. I find no point in continuing a discussion that lacks basic characteristics of logical reasoning and it's based mostly on Fake News Network™.

LE: I guess there's no Ryzen 9 after all. (Source: James Prior - AMD's Product Manager)

LE2: 2 AMD employees + James Prior confirmed on video, in a Q&A session, that there's no Ryzen 9 name. (Source, min 31)

 

[CatMerc]

The 1700 doesn't run @ 3.6 GHz

I did say with headroom to spare

That's also ignoring any potential binning or new revision that might shift the power game further.

 

[pantsaregood]

If the clock speeds rumored are correct, does anyone else find it surprising how consistent clocks are on Zen cores?

It seems that almost every Zen core can reach between 3.9 GHz and 4.1 GHz. That's not particularly high, but it is extremely consistent. Haswell and Skylake were landing everywhere from 4.3 GHz to 5.1 GHz, so it's pretty crazy to see practically every sample fall within such a small range.

 

[Topweasel]

To think that several months ago, 16 threads would have been seen as a threadripper CPU. It's amazing how the landscape has changed with the arrival of Ryzen, that we now consider 16 threads as quite normal.

The segmentation of the i7 did this. 1000 for 16 threads 1700 for 20 threads. When you only have 4c8t at a reasonable cost getting above that barrier looks great. I remember from SB to IVB to Broadwell the amount of effort Intel went into advertising the efforts they put into the iGPU. But really they did this so they didn't have increase the core count to get a die big enough to support the packaging. They went out of the way to keep consumers on the low end limited to prop up pricing on the high end. It's not surprising that it was this easy for AMD to make people re-evaluate our stance on thread count, they just needed a competitive arch.

 

[T1beriu]

First actual proof that AMD's troll is real - X399 - the platform that TR drops in.

Source: Videocardz.

 

[w3rd]

Almost everyone accepted the x399 was real when AMD announced ThreadRipper and their HEDT platform. We have seen x399 block diagrams, as you well know. So this is good news.

What we need is a SP3r2 socket sku slide. Like this:

Makes one think if AMD team was trolling us with some smirks, when they (perhaps) facetiously announced:
AMD Ryzen Threadripper

Look at that^ stand-alone name, for a sec.
I keep asking myself threadripper what? 1900x ..? Will the naming scheme change because even though Threadripper is marketed under the Ryzen name, because it is on a new HEDT socket/platform..?

And before you ask, I want to know, because I want dibs on a Decacore Ryzen Threadripper T-Shirt...
Deca Zen Ripper

"X Jack" < T-shirt

 

[T1beriu]

Almost everyone accepted the x399 was real when AMD announced ThreadRipper and their HEDT platform. We have seen x399 block diagrams, as you well know.

So I guess you were not phased at all when the X390 block diagram had its socket labeled as AM44 that is pixel perfect to an AM4? Something that's impossible for a quad-channel CPU to have the same number of pins as a dual one, thus size?

Or that there's no AM44 in the leaked cpuids from january?

Or that it contradicted Cannard PC naming of the socket as SP3r2 who is very reliably leaker that broke the news on ThreadRipper?

Or that they named it X399 but the diagram labels it as X390?

Or seen this debunking article?

Update: Those diagrams were most likely fan-made by Baidu user. ASUS X370-PRO and PRIME diagrams were used as a mockup.

Another fake brought to you by the Fake News Network™.

 

[beginner99]

But with the over 80% yield success rates I read for Ryzen wouldn't leave a large number for 8 core cpu's for the 4k pin setup.

So you can unlock them! One can dream...

 

[beginner99]

An eight core should still benefit from the quad channel though (re: AM4 8C/16T benefits from overclocked dual channel DDR4). This especially if slow DDR4 memory (commonly found in consumer pre-builts and workstations) is used at the starting point.

Yeah but it profits almost certainly because increasing memory speed increases fabric speed and hence inter-ccx communication lowering that bottleneck. It's not actually a memory bandwidth issue. In fact having more sticks makes higher clocks more difficult and could be a negative for ST performance.

 

[CatMerc]

If the clock speeds rumored are correct, does anyone else find it surprising how consistent clocks are on Zen cores?

It seems that almost every Zen core can reach between 3.9 GHz and 4.1 GHz. That's not particularly high, but it is extremely consistent. Haswell and Skylake were landing everywhere from 4.3 GHz to 5.1 GHz, so it's pretty crazy to see practically every sample fall within such a small range.

I have a feeling that anything that can reach above 4.1GHz is being binned for future models. Maybe 1900X or even Threadripper.

 

[w3rd]

So I guess you were not phased at all when the X390 block diagram had its socket labeled as AM44 that is pixel perfect to an AM4? Something that's impossible for a quad-channel CPU to have the same number of pins as a dual one, thus size?

Or that there's no AM44 in the leaked cpuids from january?

Or that it contradicted Cannard PC naming of the socket as SP3r2 who is very reliably leaker that broke the news on ThreadRipper?

Or that they named it X399 but the diagram labels it as X390?

Or seen this debunking article?


Another fake brought to you by the Fake News Network™.

I am not sure if you are serious or not^.

Dude, the X390 is on the AM4 platform & it was looked upon as just a more robust chip(set) over the X370. That was obvious, the misleading part was AMD most likely tracing what channels were leaking what. And the crafty public is left putting the pieces together, of these many different sites, different rumors and slides. That is why you have to read everything and take in ALL information. You also get to see what sites are pushing what bias.


Secondly, Your link sites are well known, what are you trying to say?
Not even sure what you are upset about here, or what your questions are suppose to answer? It is not like you are the only one who reads those sites, or those exact articles. Myself, I have 7 or 8 sites I read articles from & usually catch most of them within 8 hours after being released. But the frequency of new information about hardware (intel, amd, nvidia, etc) right now, is streaming in at a phenomenal rate.

All news is good news. But some links all just point back to the same spot. So frapping off over a news story is pointless, but that doesn't mean you dismiss the first bit of news you hear.



What kind of T-Shirt are you getting?