AMD's Upcoming Ryzen R9 "Threadripper" 16-core, 32-thread CPU

Message boards : Number crunching : AMD's Upcoming Ryzen R9 "Threadripper" 16-core, 32-thread CPU
Message board moderation

To post messages, you must log in.

Previous · 1 · 2

AuthorMessage
Keith White
Avatar

Send message
Joined: 29 May 99
Posts: 392
Credit: 13,035,233
RAC: 22
United States
Message 1869490 - Posted: 26 May 2017, 7:16:57 UTC

Oh AMD Marketing, how often I face palm at your idiocy. First confusing everyone over the core count of the Bulldozer family. And now the wacky naming conventions , "Infinity Fabric", "Threadripper", "Epyc". Someone please stop them.
"Life is just nature's way of keeping meat fresh." - The Doctor
ID: 1869490 · Report as offensive
Profile HAL9000
Volunteer tester
Avatar

Send message
Joined: 11 Sep 99
Posts: 6534
Credit: 196,805,888
RAC: 57
United States
Message 1869504 - Posted: 26 May 2017, 10:53:29 UTC - in response to Message 1869490.  

Oh AMD Marketing, how often I face palm at your idiocy. First confusing everyone over the core count of the Bulldozer family. And now the wacky naming conventions , "Infinity Fabric", "Threadripper", "Epyc". Someone please stop them.

I can only guess the reason they started using RX on the 400 series Radeon GPUs was their plan to use R3, R5, & R7 on their CPUs. It looks like they will be using RX for all of the 500 series GPUs this go around.
Perhaps the Ryzen Threadripper label is meant to be analogous to the Extreme Edition label Intel uses?
SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[
ID: 1869504 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13720
Credit: 208,696,464
RAC: 304
Australia
Message 1870278 - Posted: 31 May 2017, 9:13:56 UTC

Looks like Intel isn't going to let AMD go unchallenged.
               Cores/Threads
Core i7-7800X     6/12
Core i7-7820X     8/16
Core i9-7900X    10/20
Core i9-7920X    12/24
Core i9-7940X    14/28
Core i9-7960X    16/32
Core i9-7980XE   18/36

The pricing for the 79xx series is expected to start at painful and go to beyond excruciating for the 18/36 core/thread chip.
Annandtech's look at new Intel CPUs.

Supply for the new Intel (and AMD Threadripper) chips is expected to be tight initially, but hopefully after a few months when demand & production levels balance out a bit better, there's a very good chance of seeing some price pressure on Intel, and them lowering prices would put pressure on AMD to do the same.
The end of this year, beginning of next, is looking very interesting when it comes to PC hardware.
Grant
Darwin NT
ID: 1870278 · Report as offensive
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 5517
Credit: 528,817,460
RAC: 242
United States
Message 1870302 - Posted: 31 May 2017, 12:22:51 UTC - in response to Message 1870278.  

While Intel might release a chip with similar cores, it's the number of PCIe lanes that really make this chip something to look at.....

As Jayztwocents states, intel is cutting back on the number of PCIe lanes on lower models and from what I have seen increased it by 4 on the higher end ones..

From the reviews on the Threadripper, it has 64 PCIe lanes...That means true x16 for 4 GPUs....(as long as you don't use anything else in those lanes)..Not a big deal where but at Einstein and other projects where the apps haven't been heavily modified, that would go a long ways to improving throughput..
ID: 1870302 · Report as offensive
Profile HAL9000
Volunteer tester
Avatar

Send message
Joined: 11 Sep 99
Posts: 6534
Credit: 196,805,888
RAC: 57
United States
Message 1870439 - Posted: 1 Jun 2017, 0:46:39 UTC - in response to Message 1870302.  

While Intel might release a chip with similar cores, it's the number of PCIe lanes that really make this chip something to look at.....

As Jayztwocents states, intel is cutting back on the number of PCIe lanes on lower models and from what I have seen increased it by 4 on the higher end ones..

From the reviews on the Threadripper, it has 64 PCIe lanes...That means true x16 for 4 GPUs....(as long as you don't use anything else in those lanes)..Not a big deal where but at Einstein and other projects where the apps haven't been heavily modified, that would go a long ways to improving throughput..

I was thinking Intel had bumped up the number of PCIe lanes going from the ix-6000 series ix-7000 CPUs, but it was actually an increase in the 100 to 200 series chipset PCIe lanes.

I remember the race to break the 1GHz barrier. Maybe the next race will be something like 100+ x86 CPU cores.
However Intel is already pretty close to that with their 72 core CPUs
Being able to buy a new 16+ core CPU in the $300-400 range would be pretty nice. I'm hoping they have a good fight.
SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[
ID: 1870439 · Report as offensive
Al Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Avatar

Send message
Joined: 3 Apr 99
Posts: 1682
Credit: 477,343,364
RAC: 482
United States
Message 1870708 - Posted: 2 Jun 2017, 18:54:34 UTC - in response to Message 1870302.  

While Intel might release a chip with similar cores, it's the number of PCIe lanes that really make this chip something to look at.....

As Jayztwocents states, intel is cutting back on the number of PCIe lanes on lower models and from what I have seen increased it by 4 on the higher end ones..

From the reviews on the Threadripper, it has 64 PCIe lanes...That means true x16 for 4 GPUs....(as long as you don't use anything else in those lanes)..Not a big deal where but at Einstein and other projects where the apps haven't been heavily modified, that would go a long ways to improving throughput..
I've always wondered why they set that arbitrary limit, and would have expected them to have gone to 64 a long time ago, as it is a multiple of the 16 lanes that was specified when they created the PCIe spec years ago. Never understood their reasoning, anyone have any 'out in the weeds' insight to this? There's a lot of smart people here, I figured it wouldn't hurt to at least ask.

ID: 1870708 · Report as offensive
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 5517
Credit: 528,817,460
RAC: 242
United States
Message 1870712 - Posted: 2 Jun 2017, 19:21:05 UTC - in response to Message 1870708.  

I think we all have that question...I've yet to see an answer. But at least we know it's possible thanks to AMD...
ID: 1870712 · Report as offensive
Grant (SSSF)
Volunteer tester

Send message
Joined: 19 Aug 99
Posts: 13720
Credit: 208,696,464
RAC: 304
Australia
Message 1870725 - Posted: 2 Jun 2017, 21:24:05 UTC - in response to Message 1870708.  

I've always wondered why they set that arbitrary limit, and would have expected them to have gone to 64 a long time ago, as it is a multiple of the 16 lanes that was specified when they created the PCIe spec years ago. Never understood their reasoning, anyone have any 'out in the weeds' insight to this?

Cost, market segmentation.
Adding extra lanes requires extra silicon, so impacts on yields and profit margins.
But more importantly, CPUs with high PCIe lane counts have generally been for workstations/servers- and those products command a huge premium over general usage CPUs. Why would someone buy a high end Xeon system if they could get similar performance & connectivity from a CPU that's 1/4 the cost? So high PCIe lane counts have been kept away from the desktop to help keep up the profit margins.
Grant
Darwin NT
ID: 1870725 · Report as offensive
Profile HAL9000
Volunteer tester
Avatar

Send message
Joined: 11 Sep 99
Posts: 6534
Credit: 196,805,888
RAC: 57
United States
Message 1870776 - Posted: 2 Jun 2017, 23:49:24 UTC - in response to Message 1870708.  

While Intel might release a chip with similar cores, it's the number of PCIe lanes that really make this chip something to look at.....

As Jayztwocents states, intel is cutting back on the number of PCIe lanes on lower models and from what I have seen increased it by 4 on the higher end ones..

From the reviews on the Threadripper, it has 64 PCIe lanes...That means true x16 for 4 GPUs....(as long as you don't use anything else in those lanes)..Not a big deal where but at Einstein and other projects where the apps haven't been heavily modified, that would go a long ways to improving throughput..
I've always wondered why they set that arbitrary limit, and would have expected them to have gone to 64 a long time ago, as it is a multiple of the 16 lanes that was specified when they created the PCIe spec years ago. Never understood their reasoning, anyone have any 'out in the weeds' insight to this? There's a lot of smart people here, I figured it wouldn't hurt to at least ask.

The last time I read it the PCIe spec also included a x32 slot. Probably meant for riser cards. It seems like for GPGPU use, probably in server situations, hardware for such a lot could be developed.

One thing to keep in mind is that not of those PCIe will be available for GPUs. From one article I read:
The Ryzen processors will allow for a mammoth amount of PCIe lanes which amount to 64. There’s 48 PCIe lanes entirely dedicated to the graphics cards whereas Intel’s X299 features a total of 44 PCIe lanes and less than that would be dedicated to the graphics cards.
From that I would expect the Ryzen Threadripper to probably be setup for 16/16/16 or 16/16/8/8.
SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[
ID: 1870776 · Report as offensive
Al Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Avatar

Send message
Joined: 3 Apr 99
Posts: 1682
Credit: 477,343,364
RAC: 482
United States
Message 1871317 - Posted: 5 Jun 2017, 20:25:46 UTC - in response to Message 1870776.  
Last modified: 5 Jun 2017, 20:27:02 UTC

The last time I read it the PCIe spec also included a x32 slot. Probably meant for riser cards. It seems like for GPGPU use, probably in server situations, hardware for such a lot could be developed.

One thing to keep in mind is that not of those PCIe will be available for GPUs. From one article I read:
The Ryzen processors will allow for a mammoth amount of PCIe lanes which amount to 64. There’s 48 PCIe lanes entirely dedicated to the graphics cards whereas Intel’s X299 features a total of 44 PCIe lanes and less than that would be dedicated to the graphics cards.
From that I would expect the Ryzen Threadripper to probably be setup for 16/16/16 or 16/16/8/8.
Thanks for the insight Hal, that raises one more question for me then. Do the current crop of 'latest and greatest' cards (thinking maybe the GTX1080Ti for example) actually _use_ that full bandwidth that an x16 slot provides? If so, would it mostly be in a gaming situation, or would our crunching also put that kind of strain on the system bus as well? I thought I remembered asking something similar a couple years ago when maybe it was the 900 series cards being released and the consensus was not really, but the bus hasn't gotten any faster, yet I'd think that the demands put on it by ever higher performance cards would be getting greater. Though, I would also have to think that if we were really honestly getting close to the edge, they would be announcing maybe PCIeX or something, because I'd believe they would be very sensitive to having their expensive hardware throttled by the system bus and would be pushing to keep it well ahead of current technologies capabilities?

ID: 1871317 · Report as offensive
rob smith Crowdfunding Project Donor*Special Project $75 donorSpecial Project $250 donor
Volunteer moderator
Volunteer tester

Send message
Joined: 7 Mar 03
Posts: 22160
Credit: 416,307,556
RAC: 380
United Kingdom
Message 1871339 - Posted: 5 Jun 2017, 21:36:56 UTC

Not for SETI@Home, other projects may, and gaming probably does.
Bob Smith
Member of Seti PIPPS (Pluto is a Planet Protest Society)
Somewhere in the (un)known Universe?
ID: 1871339 · Report as offensive
Profile HAL9000
Volunteer tester
Avatar

Send message
Joined: 11 Sep 99
Posts: 6534
Credit: 196,805,888
RAC: 57
United States
Message 1871358 - Posted: 5 Jun 2017, 22:49:28 UTC - in response to Message 1871317.  

The last time I read it the PCIe spec also included a x32 slot. Probably meant for riser cards. It seems like for GPGPU use, probably in server situations, hardware for such a lot could be developed.

One thing to keep in mind is that not of those PCIe will be available for GPUs. From one article I read:
The Ryzen processors will allow for a mammoth amount of PCIe lanes which amount to 64. There’s 48 PCIe lanes entirely dedicated to the graphics cards whereas Intel’s X299 features a total of 44 PCIe lanes and less than that would be dedicated to the graphics cards.
From that I would expect the Ryzen Threadripper to probably be setup for 16/16/16 or 16/16/8/8.
Thanks for the insight Hal, that raises one more question for me then. Do the current crop of 'latest and greatest' cards (thinking maybe the GTX1080Ti for example) actually _use_ that full bandwidth that an x16 slot provides? If so, would it mostly be in a gaming situation, or would our crunching also put that kind of strain on the system bus as well? I thought I remembered asking something similar a couple years ago when maybe it was the 900 series cards being released and the consensus was not really, but the bus hasn't gotten any faster, yet I'd think that the demands put on it by ever higher performance cards would be getting greater. Though, I would also have to think that if we were really honestly getting close to the edge, they would be announcing maybe PCIeX or something, because I'd believe they would be very sensitive to having their expensive hardware throttled by the system bus and would be pushing to keep it well ahead of current technologies capabilities?

They keep increasing the speed of PCI Express. So increasing the number of lanes isn't really required. The final draft for PCIe 4.0 with 16GT/s ( double PCIe 3.0's 8GT/s) is expected to be released this year.
IIRC the SETI@home SoG GPU apps should use less PCIe bandwidth vs the non-SoG apps.
I seem to recall someone doing test with the CUDA app with GPUs in different PCIe slots. I want to say they they saw a minor increase in run times with a x4 slot and significant slowdowns with a x1 slot.
With my GTX 750 ti changing from a 2.0 x16 to a 2.0 x1 slot didn't make any difference. Currently the 750 ti is averaging ~20% for bus usage with two instance of the SoG app and running in a x1 slot.
SETI@home classic workunits: 93,865 CPU time: 863,447 hours
Join the [url=http://tinyurl.com/8y46zvu]BP6/VP6 User Group[
ID: 1871358 · Report as offensive
Profile Zalster Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 27 May 99
Posts: 5517
Credit: 528,817,460
RAC: 242
United States
Message 1871361 - Posted: 5 Jun 2017, 22:57:56 UTC - in response to Message 1871358.  
Last modified: 5 Jun 2017, 22:58:28 UTC

Since Seti uses highly optimized applications, the bandwidth of the GPU and PCIe aren't as important here.

But in other projects, say Einstein, which don't have our great developers of apps, it does play a significant difference.

Back when Einstein was running the cuda versions, the PCIe speeds affected how quickly work got done. The recent changes in GPUs memory speed has also show a difference for that project, not so much for this project.

Jayztwocents put up a new video where he talks about what this means for consumers as well as professionals. Worth the watch.
https://youtu.be/QJNRtGo5IMc
ID: 1871361 · Report as offensive
Profile Keith Myers Special Project $250 donor
Volunteer tester
Avatar

Send message
Joined: 29 Apr 01
Posts: 13161
Credit: 1,160,866,277
RAC: 1,873
United States
Message 1871373 - Posted: 5 Jun 2017, 23:42:28 UTC - in response to Message 1871361.  

Thanks for the link. One of my favorite YT channels. Follows along with the commentary from Linus in his walk of the streets of Computex I watched yesterday. Seems the consensus of most of the tech reviewers is Intel...... what are you doing???? Makes no sense to most people on the upcoming product placements. And I do believe like most of the comments I've read that Intel decision about KL-X SL-X and X299 is simply a quick reaction to AMD with the surge in Ryzen and Threadripper interest. Interesting comment about the rumor about Coffee Lake getting pulled 6 months forward.

About time we got a real contest going in CPU development of which there hasn't been one for ten years.
Seti@Home classic workunits:20,676 CPU time:74,226 hours

A proud member of the OFA (Old Farts Association)
ID: 1871373 · Report as offensive
Previous · 1 · 2

Message boards : Number crunching : AMD's Upcoming Ryzen R9 "Threadripper" 16-core, 32-thread CPU


 
©2024 University of California
 
SETI@home and Astropulse are funded by grants from the National Science Foundation, NASA, and donations from SETI@home volunteers. AstroPulse is funded in part by the NSF through grant AST-0307956.