AMD ThreadRipper/Ryzen-9 info

H

Hetz

n00b
Joined
Jul 2, 2017
Messages
2
Hi,


I've been following AMD for quite a while and from the info that was released and based on their previous product announcements (Ryzen-5,7) I have created 2 things:

  • Model numbers (the ones which are published on the net are the temporary interim numbers, what will be official will be different)
  • Pricing based on the Ryzen-5,Ryzen-7 line
So I created 2 videos:

  • video 1 - Ryzen-9 with what I think will be the prices from 10C/20T up to 16C/32T - it's here:
  • video 2 - Ryzen-9 model numbers compared to what was previously rumored, it's here:
 
Pieter3dnow said:
Why Would AMD need a lot of money and a lot of money right now ?
Click to expand...
Well, they have investors, shareholders who gets the quarterly financial report, so they need to show a bigger profit from the Zen architecture investment.
 
highly unlikely there will be a 10c part, as TR will be a pair of Ryzen dice, each with an equal number of cores per CCX.
 
bwang said:
highly unlikely there will be a 10c part, as TR will be a pair of Ryzen dice, each with an equal number of cores per CCX.
Click to expand...
That's all based on the now rather old fantasy "leak". The only core counts are going to be 12 and 16. 8 core variant is possible but probably unlikely.

The thing is that both Zeppelin dies must have the same amount of cores enabled. Kinda like you can't just enable three cores from one CCX and four from another CCX. According to the Stilt and others who really know their shit.
 
Amd aready stated that it will be even cores across CCXs, so I too doubt there would be a 10 core also. Some sort of balance across dies would offer stability in my thinking..
 
Tup3x said:
That's all based on the now rather old fantasy "leak". The only core counts are going to be 12 and 16. 8 core variant is possible but probably unlikely.

The thing is that both Zeppelin dies must have the same amount of cores enabled. Kinda like you can't just enable three cores from one CCX and four from another CCX. According to the Stilt and others who really know their shit.
Click to expand...

They may still bin the CPUs with defective cores and release them as a lower core count CPU. Remember the triple core CPUs http://www.zdnet.com/article/why-amds-triple-core-phenom-is-a-bigger-deal-than-you-think/ ?
 
Now that the Intel's Skylake-X's are here, I actually cannot wait to see what TR is capable of, especially since some reviews out there right now uses Ryzen's INITIAL release performances as comparison, rather than Ryzen's performance now.
 
chenw said:
Now that the Intel's Skylake-X's are here, I actually cannot wait to see what TR is capable of, especially since some reviews out there right now uses Ryzen's INITIAL release performances as comparison, rather than Ryzen's performance now.
Click to expand...

I really hope they ironed out the lev2 and level3 cache latencies between the CCX modules etc.... might bump IPC up there with Intels best.... or not. But 32 threads is like meth to me. Cant wait to get high on it. Maybe the performance will be so high (wild dream) that ir forces the 18 core I9 down to a thousand and then I will get that.
 
I love the speed of my 1700X and 1700 systems. That said, Threadripper is going to be even better. :)
 
ManofGod said:
I love the speed of my 1700X and 1700 systems. That said, Threadripper is going to be even better. :)
Click to expand...

Whilst it may clock lower I have a strong feeling the IPC will be improved as well as the mASSIVE amount of threads... I am absolutely hyped at the prospect of having one in a month or two we hope. (I have no idea when they release)

One question, and its an unimportant tasking of my systems, can these 32 thread chips encode/transcode faster than my 980ti CUDA cores? Right now Cuda is still faster on my 980ti than my CPU when doing transcoding. Yes I know unimportant but still a valid question.
 
Last edited:
selflove said:
They may still bin the CPUs with defective cores and release them as a lower core count CPU. Remember the triple core CPUs http://www.zdnet.com/article/why-amds-triple-core-phenom-is-a-bigger-deal-than-you-think/ ?
Click to expand...
Triple core Zeppelin is not a possible configuration. Neither are 5 or 7. Since both Zeppelin dies must have the same amount of cores, 10 and 14 Threadrippers are not going to happen.

(EDIT:
Triple core would be possible but that would mean disabling other CCX completely. It's highly unlikely that AMD would do that since so far all CPU's have had both CCXs at least partly enabled. So in practice is might not really work.)
 
Last edited:
tangoseal said:
I really hope they ironed out the lev2 and level3 cache latencies between the CCX modules etc.... might bump IPC up there with Intels best.... or not. But 32 threads is like meth to me. Cant wait to get high on it. Maybe the performance will be so high (wild dream) that ir forces the 18 core I9 down to a thousand and then I will get that.
Click to expand...
there was talk of the Infinity Fabric running 2X speed, but not sure which release that was supposed to be.
 
chenw said:
Now that the Intel's Skylake-X's are here, I actually cannot wait to see what TR is capable of, especially since some reviews out there right now uses Ryzen's INITIAL release performances as comparison, rather than Ryzen's performance now.
Click to expand...

Leaks circulating in the tech sites

Threadripper 1950X @3.4GHz: 4074 points (GB single); 26768 points (GB multi). Skylake i9-7900X @3.3GHz do 5000--5300 points and 32000--34000 points respectively.

Threadripper 1950X @3.9GHz: 434.32 GOPS and 821.64 Mpix/s on SiSoftware Arithmetic and Multimedia workloads, respectively. The Skylake i9-7900X @3.3GHz scores 336.20 GOPS and 1262.68 Mpix/s.

This is the same situation seen in servers, with 28C Skylake Xeon beating 32C Zen EPYC in SiSoft, SPECint, CB15,...
 
chenw said:
*sigh*... :(

Oh well, best wait until actual release then.
Click to expand...

I will take 10% less perf at 40-50% less price. This is where amd is slaughtering Intel. 10 vs 16 is intel killing Threadripper right now. But we wont know if that is just like bullshit photoshopped propaganda floating arond. After all the site was wccftech afterall.
 
I will take 10% less perf at 40-50% less price
Click to expand...

I can see a 12C TR achieving that at least in highly threaded tasks. Intel is going to easily win in per core performance. I don't see AMD getting within 10% on that.

I am interested in TR mostly because of the ECC support and platform longevity. Although with that said I would without any hesitation pay the price stated for the 18C / 36T Skylake X if it had ECC.
 
juanrga said:
Leaks circulating in the tech sites
Threadripper 1950X @3.4GHz: 4074 points (GB single); 26768 points (GB multi). Skylake i9-7900X @3.3GHz do 5000--5300 points and 32000--34000 points respectively.
Threadripper 1950X @3.9GHz: 434.32 GOPS and 821.64 Mpix/s on SiSoftware Arithmetic and Multimedia workloads, respectively. The Skylake i9-7900X @3.3GHz scores 336.20 GOPS and 1262.68 Mpix/s.
This is the same situation seen in servers, with 28C Skylake Xeon beating 32C Zen EPYC in SiSoft, SPECint, CB15,...
Click to expand...

Well there seem to be a few issues with the platform which are quite bad on Intel side :
http://www.mersenneforum.org/showpost.php?p=462751&postcount=64

When I was testing with AVX512, I noticed that the throttling doesn't lower the clock speed. Instead, it seems to cut the IPC in half. So it's architectural throttling of some sort. The most natural hypothesis is that the dedicated FMA on port 5 gets shutdown. But I currently have no evidence to support this theory nor have I run any targeted benchmarks to test it.
Click to expand...

This sort of "phantom throttling" is new to me and I believe new only to Skylake X. And it lets people "think" they're running at 5.0 GHz on air when in reality, everything is getting throttled down to less than half speed and you won't notice until you actually measure the performance and/or power draw. The average overclocker who only cares about showing off their CPUz will not notice and will be very happy.

Some motherboards throttle more than others. My Gigabyte motherboard is particularly bad as maxing out all the thermal limits still doesn't let me run AVX512 on all cores above 3.8 GHz without the "phantom throttling" even though I have plenty of temperature headroom left. Others have reported that ASUS motherboards don't have this throttling at all. (but they have the VRM and power-connector issues instead)
Click to expand...
 
Pieter3dnow said:
Well there seem to be a few issues with the platform which are quite bad on Intel side :
http://www.mersenneforum.org/showpost.php?p=462751&postcount=64
Click to expand...

What a coincidence! The same link was given yesterday, in another forum, by Fottemberg.

Too bad that none of you provide links to later posts from the same user where he explains it is a BIOS/mobo problem:

The AVX512 throtting issues is probably just an early BIOS problem. The throttling that I'm seeing at stock is because it tries to keep all 10 cores turboed at 4.0 GHz while running AVX512 without trying to apply any AVX512 offset. I believe the mobo/BIOS is supposed to apply the AVX512 offset there.
Click to expand...

The same user wrote today another post on how "throttling problem on Gigabyte motherboards has been solved."

Just an update for anyone into overclocking.

Some better overclockers than me have figured out the problem with Gigabyte motherboards that is causing the phantom throttling.

The cause is a voltage droop on the CPU input voltage (VCCIN). When the CPU pulls a large current due to an intensive load, the VCCIN drops from 1.8v to 1.6v. This drop to 1.6v puts the CPU into some sort of "panic" state of unknown nature that causes the IPC drop and subsequent "phantom throttling".

The solution is to make sure the VCCIN stays above 1.7v. This can be done either by increasing the base voltage to higher than the default 1.8v. (1.85 worked for me) Alternatively, enabling Load-Line Calibration will compensate for the vdroop by dynamically increasing the voltage under load to offset the vdroop. (This also worked for me and is what I'm currently running.)

Once the vdroop has been solved, it won't work yet. The moment you try to pull 4 GHz of AVX512, it will hard shutdown the system instead of phantom throttle. This is caused by the VCore current protection. Maxing out the VCore current limit finally let me run AVX512 on all cores @ 4.0 GHz without shutting down or throttling. However, the temperatures did reach 95C. So it's not something I recommend anyone doing unless they've delided the processor and are running custom water. But at the very least, the throttling problem on Gigabyte motherboards has been solved.

The exact nature of the phantom throttling is still unknown. And it is receiving enough attention that people are probably gonna start looking at it.
Click to expand...
 
juanrga said:
What a coincidence! The same link was given yesterday, in another forum, by Fottemberg.

Too bad that none of you provide links to later posts from the same user where he explains it is a BIOS/mobo problem:



The same user wrote today another post on how "throttling problem on Gigabyte motherboards has been solved."
Click to expand...
Thanks for the update!
Right now I am still running my 1366 based X5650 Xeon at 4.5Ghz which has been working great for me but all this new processors are finally making me take notice. I had my eyes on Threadripper as my return to AMD since I left after Phenom 965 I think it was. Intel has been great for me but I am always willing to change if the performance increase merits it. Finally it seems, it is going to happen. Now the decision is 7900xx or Threadripper. I mostly do video editing/encoding and lots of photo editing with plugins etc and little gaming so that is why this 10 or 16 core processors are so interesting to me.
Cant wait for TR to get released to finally have real world results.
 
I'm very, very curious about Threadripper. I really hope that it competes favorably in terms of features/performance with Intel's X299 platform Skylake-X options, for everything from single threaded gaming use to multithreaded loads, on Linux etc. I've long been a fan of the Intel HEDP when it was done right - much like Shaolin95 here, I've had a fantastic experience with X58 , back when Intel put forth some exceptional CPUs and chipsets that hold up even today, but before they started resting on their laurels and jacking prices sky high. I recently upgraded to X99 with a Haswell-E 5960X 8-core on one system which will likely be my main rig. However, I need to rebuild another PC soon and I've been watching Threadripper closely. It looks like X299 and the Skylake-X CPUs are going to be astoundingly expensive, truncated on features without perhaps spending even more, and there are already potential problems, which makes me feel a bit better about my X99 build still holding its own .

Though I prefer AMD from an ethical standpoint versus both Intel and Nvidia in their respective fields, I need the offerings from AMD to keep up with performance. I'm very glad AMD has finally decided on a HEDP of their own with X399 and Threadripper, but it needs to be significantly cheaper and very close in performance to Intel to be worthwhile. That is really where AMD seems to be at their best, the "10% less powerful, 60% cheaper" value proposition. I'm a bit worried about the numbers we see here because if they hold true, combined with the potential for various types of software - benchmark and realworld alike - to be "optimized" for Intel, it may not meet the performance levels necessary. When Ryzen arrived it certainly looked good, but seeing what appeared to be midding gaming performance, single threaded issues, and problems on Linux etc.. was frustrating. If I'm correct most of this has been fixed (though I'd like to check to see if people still think Ryzen is "not a good gaming/single thread CPU" as opposed to simply not topping the benchmarks at single threaded gaming tasks, but still a solid performer when it comes to real world use), but I am curious of what Threadripper will bring. If it is a highh end platform, they can't fail to bring the power or only make it very situational (ie its only good at multithreaded niche loads etc), nor can they make it as expensive as Intel if they want converts. Hopefully Threadripper will be a beast and cost a ton less than the asinine prices that Intel is starting to charge as I want to see success for AMD and them grab some marketshare. I guess we shall see when it arrives and is tested properly, but I can only hope for the best.
 
juanrga said:
This is the same situation seen in servers, with 28C Skylake Xeon beating 32C Zen EPYC in SiSoft, SPECint, CB15,...
Click to expand...

The 28 core Skylake Xeon also costs 3 times as much as the 32 core Zen EPYC.. you could likely build a full 2 socket 64core EPYC system for the price of a single 28c Xeon chip. I don't know if i would consider that a win.
 
Briguyatx said:
The 28 core Skylake Xeon also costs 3 times as much as the 32 core Zen EPYC.. you could likely build a full 2 socket 64core EPYC system for the price of a single 28c Xeon chip. I don't know if i would consider that a win.
Click to expand...

If CPU price was the only factor Opterons would have won long ago.
 
Briguyatx said:
I may be wrong anyway - it looks like the 28core skylake is only around 3200.00, assuming that is a real listing.

https://world.taobao.com/item/42659072321.htm?fromSite=main

Interesting.
Click to expand...
Briguyatx said:
I may be wrong anyway - it looks like the 28core skylake is only around 3200.00, assuming that is a real listing.

https://world.taobao.com/item/42659072321.htm?fromSite=main

Interesting.
Click to expand...

Sounds like a phony listing. I saw nothing previously about a 28 core skylake chip.
 
Briguyatx said:
The 28 core Skylake Xeon also costs 3 times as much as the 32 core Zen EPYC.. you could likely build a full 2 socket 64core EPYC system for the price of a single 28c Xeon chip. I don't know if i would consider that a win.
Click to expand...

Where is the official pricing for both chips? How much faster can be the Xeon? 5989.90Mpix/s vs 974.33Mpix/s looks as 6x faster to me. 1425.82GOPS vs 706.18GOPS looks as 2x faster. Also maintainance cost is more important on servers than acquisition cost.
 
Density is also an important consideration.

If 2 Epycs is equal to 1 Xeon in performance, chances are most will probably still prefer 1 Xeon for their smaller footprint.
 
Shintai said:
https://videocardz.com/newz/amd-ryzen-threadripper-1950x-to-cost-999-usd

1950X 16C/32T 3.4Ghz base, 4Ghz max turbo. 999$
1920X 12C/24T 3.5Ghz base, 4Ghz max turbo. 799$
Click to expand...
Not sure about this, but I think they should price them like this. Remember when the 1800X went down the assumption was to make room for the 12 core TR, which this price list doesn't seem to fall in line with. But given Intels clocking woes on X299 maybe they decided to keep them a little aloft for now with a reduction later like AMD usually does.
 
chenw said:
Density is also an important consideration.

If 2 Epycs is equal to 1 Xeon in performance, chances are most will probably still prefer 1 Xeon for their smaller footprint.
Click to expand...

The benchmarks coming show the epyc beating the xeon in performance and power.
 
Shintai said:
The AT benchmark that didn't match any other reviews, or their own SKL-X? Ye that one was real "trustworthy". Must have been another advertorial :D
Click to expand...
so the fact they spent one week with EPYC and 2 weeks with Xeon doesn't count? They were very transparent and gave pretty decent explanations for each and every test. The biggest one left, even they admit needs further testing is the power usage, I have to agree as if true then that is a HUGE win for AMD.
 
JustReason said:
so the fact they spent one week with EPYC and 2 weeks with Xeon doesn't count? They were very transparent and gave pretty decent explanations for each and every test. The biggest one left, even they admit needs further testing is the power usage, I have to agree as if true then that is a HUGE win for AMD.
Click to expand...

Nothing of that explains the discrepancy in performance and/or power with other reviews or even with their own former reviews.
 
juanrga said:
Nothing of that explains the discrepancy in performance and/or power with other reviews or even with their own former reviews.
Click to expand...
regardless, unless you have proof, that one review is actually spot on, all you have is conjecture. They speak of the power usage as not being given enough time in their review and would revisit later, so any mentions are not deep dive not conclusive, thereby can not be accused of bias or intentionally misleading as both of you are inferring, hell not inferring you two are outright accusing.
 
Which is funny since that site used to be bashed for being anti AMD, now it's anti Intel according to them.
 
You must log in or register to reply here.
Back
Top