Jump to content

Threadripper coming: split for low and high end, 8 channel memory

Fasauceome
1 hour ago, Den-Fi said:

My body is ready.

it is PGA.

I live in misery USA. my timezone is central daylight time which is either UTC -5 or -4 because the government hates everyone.

into trains? here's the model railroad thread!

Link to comment
Share on other sites

Link to post
Share on other sites

2 minutes ago, will4623 said:

it is PGA.

No, this is Patrick.

Link to comment
Share on other sites

Link to post
Share on other sites

4 minutes ago, will4623 said:

it is PGA.

PGA threadripper be like

image.jpeg.1dc14f6168ca43a392d32c2f90879666.jpeg

I WILL find your ITX build thread, and I WILL recommend the SIlverstone Sugo SG13B

 

Primary PC:

i7 8086k - EVGA Z370 Classified K - G.Skill Trident Z RGB - WD SN750 - Jedi Order Titan Xp - Hyper 212 Black (with RGB Riing flair) - EVGA G3 650W - dual booting Windows 10 and Linux - Black and green theme, Razer brainwashed me.

Draws 400 watts under max load, for reference.

 

How many watts do I needATX 3.0 & PCIe 5.0 spec, PSU misconceptions, protections explainedgroup reg is bad

Link to comment
Share on other sites

Link to post
Share on other sites

3 minutes ago, Fasauceome said:

PGA threadripper be like

image.jpeg.1dc14f6168ca43a392d32c2f90879666.jpeg

*snores*

I live in misery USA. my timezone is central daylight time which is either UTC -5 or -4 because the government hates everyone.

into trains? here's the model railroad thread!

Link to comment
Share on other sites

Link to post
Share on other sites

1 hour ago, S w a t s o n said:

As excited as I am for the 16 core 4 channel TR, where the fuck is the 3950x??? It's September Lisa, the clock is ticking...

It's it supposed to be Sept 23rd?

 

Edit: looks like it's just Twitter rumor mill not official yet.

Link to comment
Share on other sites

Link to post
Share on other sites

On 9/9/2019 at 3:01 AM, BiG StroOnZ said:

Wondering if the split has anything to do with this (for AMD to be even more competitive):

 

Intel-Cascade-Lake-Xeadripper-CPUs.thumb.png.b476d6699e2f53dafe4cdfe07636ab81.png

Arguments aside. If Intel double their price-performance metric in a single half generation, I will fall off my chair.

Link to comment
Share on other sites

Link to post
Share on other sites

On 9/9/2019 at 5:43 AM, VegetableStu said:

one last thing before i leave this thread temporarily for my own good: i set aside a thread to anticipate any TR announcement event, and in it i had a list of current CCD counts in EPYC 7002 parts.

 

the higher the CCD count the more Lx cache it has:

  • 8c (you read that right)
    • 1x8c 7232P (you read that right)
    • 2x4c 7252
    • 4x2c 7262 (you read THAT right)
  • 12c - 2x6c 7272
  • 16c
    • 4x4c - 7320P etc
    • 2x8c - 7282
  • 24c - 4x6c - 7420P etc
  • 32c - 4x8c - 7502P etc
  • 48c - 6x8c - 7642 etc (oddly does not have a single-CPU part)
  • 64c - 8x8c - 7702P, etc

 

oddly there wasn't a 6x6c, but I think it might be a viable alternative, like for example 4x8c on sTRX4 and 6x6c on sWRX8

I'm interested in how a home PC (gaming etc) performs with 128mb cache... my first PC had that in RAM!!! ?

Link to comment
Share on other sites

Link to post
Share on other sites

4 minutes ago, TechyBen said:

Arguments aside. If Intel double their price-performance metric in a single half generation, I will fall off my chair.

It can be achieved in a combination of two ways: increase performance, lower selling price. The only question is what balance of those factors are used. AMD have historically chosen to price low compared to Intel (and nvidia). If the costs to produce are low, that helps enable that without impacting profits. Since pricing is a choice, personally I like to analyse the hardware independently of price to understand their inherent strengths/weaknesses, before factoring in value later on when buying.

 

2 minutes ago, TechyBen said:

I'm interested in how a home PC (gaming etc) performs with 128mb cache... my first PC had that in RAM!!! ?

My first PC had a 250MB HD! Never mind ram... As often the case, it will depend on the application. Ryzen architecture is essentially in chunks of CCX. If it isn't stored on local CCX, it has to go back to ram. This is different from Intel where all L3 cache is available to any core. Some tasks that benefit more from a unified L3 may still favour Intel, but code that is more easily split will scale with the brute force of throwing cores at it.

Gaming system: R7 7800X3D, Asus ROG Strix B650E-F Gaming Wifi, Thermalright Phantom Spirit 120 SE ARGB, Corsair Vengeance 2x 32GB 6000C30, RTX 4070, MSI MPG A850G, Fractal Design North, Samsung 990 Pro 2TB, Acer Predator XB241YU 24" 1440p 144Hz G-Sync + HP LP2475w 24" 1200p 60Hz wide gamut
Productivity system: i9-7980XE, Asus X299 TUF mark 2, Noctua D15, 64GB ram (mixed), RTX 3070, NZXT E850, GameMax Abyss, Samsung 980 Pro 2TB, random 1080p + 720p displays.
Gaming laptop: Lenovo Legion 5, 5800H, RTX 3070, Kingston DDR4 3200C22 2x16GB 2Rx8, Kingston Fury Renegade 1TB + Crucial P1 1TB SSD, 165 Hz IPS 1080p G-Sync Compatible

Link to comment
Share on other sites

Link to post
Share on other sites

9 minutes ago, porina said:

If it isn't stored on local CCX, it has to go back to ram.

I thought between CCX on same CCD could be done? Otherwise back to IOD. Really not sure on that, Zen 2 arch info is very light but maybe a review on Zen/Zen+ between CCX might help. I'm not even sure it matters either since L3 cache in Zen is more a dumping ground for the CCX so cross access might be a thing at all, designed not to be.

Link to comment
Share on other sites

Link to post
Share on other sites

10 minutes ago, leadeater said:

You had an HDD!? Dual floppy only here ?

Think we could deduce an age range for people based on 1st systems, although it could also be skewed by when in their life they got their first PC. I got my first PC after I got my 1st car.

 

Actually, my first "PC" could be even earlier, if you include software emulation. I had PC Ditto on Atari ST. I recall Landmark benchmark rated it around 18% the speed of an XT, which was 8088 at 4.77MHz. So... really not fast!

Gaming system: R7 7800X3D, Asus ROG Strix B650E-F Gaming Wifi, Thermalright Phantom Spirit 120 SE ARGB, Corsair Vengeance 2x 32GB 6000C30, RTX 4070, MSI MPG A850G, Fractal Design North, Samsung 990 Pro 2TB, Acer Predator XB241YU 24" 1440p 144Hz G-Sync + HP LP2475w 24" 1200p 60Hz wide gamut
Productivity system: i9-7980XE, Asus X299 TUF mark 2, Noctua D15, 64GB ram (mixed), RTX 3070, NZXT E850, GameMax Abyss, Samsung 980 Pro 2TB, random 1080p + 720p displays.
Gaming laptop: Lenovo Legion 5, 5800H, RTX 3070, Kingston DDR4 3200C22 2x16GB 2Rx8, Kingston Fury Renegade 1TB + Crucial P1 1TB SSD, 165 Hz IPS 1080p G-Sync Compatible

Link to comment
Share on other sites

Link to post
Share on other sites

9 minutes ago, leadeater said:

I thought between CCX on same CCD could be done? Otherwise back to IOD. Really not sure on that, Zen 2 arch info is very light but maybe a review on Zen/Zen+ between CCX might help. I'm not even sure it matters either since L3 cache in Zen is more a dumping ground for the CCX so cross access might be a thing at all, designed not to be.

On Zen 2, it was stated that all CCX to CCX communication has to go via IOD, even if they were located on the same CCD. Reason given was to allow more predictable latency.

 

Kinda related (click through for thread for context):

 

Gaming system: R7 7800X3D, Asus ROG Strix B650E-F Gaming Wifi, Thermalright Phantom Spirit 120 SE ARGB, Corsair Vengeance 2x 32GB 6000C30, RTX 4070, MSI MPG A850G, Fractal Design North, Samsung 990 Pro 2TB, Acer Predator XB241YU 24" 1440p 144Hz G-Sync + HP LP2475w 24" 1200p 60Hz wide gamut
Productivity system: i9-7980XE, Asus X299 TUF mark 2, Noctua D15, 64GB ram (mixed), RTX 3070, NZXT E850, GameMax Abyss, Samsung 980 Pro 2TB, random 1080p + 720p displays.
Gaming laptop: Lenovo Legion 5, 5800H, RTX 3070, Kingston DDR4 3200C22 2x16GB 2Rx8, Kingston Fury Renegade 1TB + Crucial P1 1TB SSD, 165 Hz IPS 1080p G-Sync Compatible

Link to comment
Share on other sites

Link to post
Share on other sites

5 minutes ago, porina said:

On Zen 2, it was stated that all CCX to CCX communication has to go via IOD, even if they were located on the same CCD. Reason given was to allow more predictable latency.

Hmm then that has changed from Zen/Zen+. Most of the stuff I could find talked about latency consistency to the system memory across the cores, but on his Zen 2 architecture analysis article it says this:

Quote

AMD manages its L3 by sharing a 16MB block per CCX, rather than enabling access to any L3 from any core.

 

I'm thinking since each CCD has a SDF/IF in it and it's actually possible to have that data path it'll come at some later point when there is better market adoption and more reason to optimize. I guess it's just removing a potential problem point.

Link to comment
Share on other sites

Link to post
Share on other sites

19 hours ago, Taf the Ghost said:

It's it supposed to be Sept 23rd?

 

Edit: looks like it's just Twitter rumor mill not official yet.

i just read somewhere that talked about it being on 30th

 

Link to comment
Share on other sites

Link to post
Share on other sites

1 hour ago, leadeater said:

Hmm then that has changed from Zen/Zen+. Most of the stuff I could find talked about latency consistency to the system memory across the cores, but on his Zen 2 architecture analysis article it says this:

Was there a short cut in Zen(+)? I don't recall testing that particular detail at the time.

 

1 hour ago, leadeater said:

I'm thinking since each CCD has a SDF/IF in it and it's actually possible to have that data path it'll come at some later point when there is better market adoption and more reason to optimize. I guess it's just removing a potential problem point.

I hadn't come across SDF as an acronym before, I have to thank google for the following when I searched for it... back on topic, it has never been clear to me how cores/CCX, IF, L3 are logically connected. CCX as smallest logical unit makes sense for scaling with what they have, even if I'd still prefer CCD to be such. The first processor beyond 4 core CCX will be interesting, and IMO a necessity as some point if they want to keep scaling up to ever higher core counts.

 

amd-sdf.png.45cc27e3efda995f70a9fbd67e4c9ea5.png

Gaming system: R7 7800X3D, Asus ROG Strix B650E-F Gaming Wifi, Thermalright Phantom Spirit 120 SE ARGB, Corsair Vengeance 2x 32GB 6000C30, RTX 4070, MSI MPG A850G, Fractal Design North, Samsung 990 Pro 2TB, Acer Predator XB241YU 24" 1440p 144Hz G-Sync + HP LP2475w 24" 1200p 60Hz wide gamut
Productivity system: i9-7980XE, Asus X299 TUF mark 2, Noctua D15, 64GB ram (mixed), RTX 3070, NZXT E850, GameMax Abyss, Samsung 980 Pro 2TB, random 1080p + 720p displays.
Gaming laptop: Lenovo Legion 5, 5800H, RTX 3070, Kingston DDR4 3200C22 2x16GB 2Rx8, Kingston Fury Renegade 1TB + Crucial P1 1TB SSD, 165 Hz IPS 1080p G-Sync Compatible

Link to comment
Share on other sites

Link to post
Share on other sites

30 minutes ago, porina said:

I hadn't come across SDF as an acronym before, I have to thank google for the following when I searched for it... back on topic, it has never been clear to me how cores/CCX, IF, L3 are logically connected. CCX as smallest logical unit makes sense for scaling with what they have, even if I'd still prefer CCD to be such. The first processor beyond 4 core CCX will be interesting, and IMO a necessity as some point if they want to keep scaling up to ever higher core counts.

IF is just the slightly broader term for things inside of it like the SDF. In Zen/Zen+ this was used for cores to access the other L3 slices, across CCXs. This gave slightly odd results in Ryzen, overall not that serious but with TR and EPYC these multiple different latency hops and accesses was a lot of trouble. In CCX, across CCX, cross die to CCX, across die to IMC. All levels of optimization hell lol.

 

Quote

AMD organized Zen in groups of cores called a CPU Complex (CCX). Each CCX consists of four cores connected to an L3 cache. The L3 cache is an 8 MiB 16-way set associative victim cache and is mostly exclusive of the L2. The L3 cache is made of four slices (providing 2 MiB L3 slice/core) organized by low-order address interleaved. Every core can access every L3 cache slice with the same average latency. When a certain core starts working on a chunk of memory it will fill up the L2 and as it continue to execute and fetch new data any spillover will find its way in the L3.

https://en.wikichip.org/wiki/amd/microarchitectures/zen#Scalability

 

Wikichip is pretty awesome, but not much on there for Zen 2 yet.

Link to comment
Share on other sites

Link to post
Share on other sites

On 9/8/2019 at 10:02 PM, williamcll said:

What even is the reason to buy low-end EPYC at this point then?

Threadripper is designed-as its name implies-to destroy tasks like video editing, machine learning, causing small blackouts in your neighborhood, and whatever other big tasks you set in front of it. It is not designed to be kept on all day though.

 

EPYC is the Enterprise version of Threadripper. Still a monster of a CPU, but with more stringent Quality Control standards and the consideration of being turned on and left on for 3+ years as well as being put in to rack mounted cases.

 

乇乂丅尺卂 丅卄工匚匚

Link to comment
Share on other sites

Link to post
Share on other sites

On 9/8/2019 at 11:40 PM, Fasauceome said:

On the other hand, maybe they'll manage an ITX TR4 motherboard, somehow

 

Maybe quantum superposition? The ram and CPU socket occupy the same region of space 

What cooler that's properly designed for threadripper is going to fit into an ITX form factor case?

Link to comment
Share on other sites

Link to post
Share on other sites

59 minutes ago, Crowbar said:

What cooler that's properly designed for threadripper is going to fit into an ITX form factor case?

We'll cross that bridge when we get to it

I WILL find your ITX build thread, and I WILL recommend the SIlverstone Sugo SG13B

 

Primary PC:

i7 8086k - EVGA Z370 Classified K - G.Skill Trident Z RGB - WD SN750 - Jedi Order Titan Xp - Hyper 212 Black (with RGB Riing flair) - EVGA G3 650W - dual booting Windows 10 and Linux - Black and green theme, Razer brainwashed me.

Draws 400 watts under max load, for reference.

 

How many watts do I needATX 3.0 & PCIe 5.0 spec, PSU misconceptions, protections explainedgroup reg is bad

Link to comment
Share on other sites

Link to post
Share on other sites

3 minutes ago, RotoCoreOne said:

Define low end... :(

In this context, the lower performance parts within the product stack.

Come Bloody Angel

Break off your chains

And look what I've found in the dirt.

 

Pale battered body

Seems she was struggling

Something is wrong with this world.

 

Fierce Bloody Angel

The blood is on your hands

Why did you come to this world?

 

Everybody turns to dust.

 

Everybody turns to dust.

 

The blood is on your hands.

 

The blood is on your hands!

 

Pyo.

Link to comment
Share on other sites

Link to post
Share on other sites

On 9/9/2019 at 9:22 AM, Den-Fi said:

My body is ready.

But is your wallet ready?

5 hours ago, TechyBen said:

I'm interested in how a home PC (gaming etc) performs with 128mb cache... my first PC had that in RAM!!! ?

5 hours ago, porina said:

My first PC had a 250MB HD! Never mind ram.

5 hours ago, leadeater said:

You had an HDD!? Dual floppy only here ?

1MB RAM, no HDD and dual floppies in a Commodore Amiga 2000, with an additional 5.25" floppy drive hooked up to a PC emulator (which I never got working, unfortunately).

 

Ah, those were the days.  Ironically, my second computer (an AMD 486 DX/2 66MHz) had a 500MB HDD, but only 640KB of RAM.  Yes, a single stick of 640KB EDO SIMM RAM.

Link to comment
Share on other sites

Link to post
Share on other sites

Just now, Jito463 said:

But is your wallet ready?

Yes. Yes it s.

Link to comment
Share on other sites

Link to post
Share on other sites

12 minutes ago, Den-Fi said:

Yes. Yes it s.

WHERE DO YOU WORK?!

I live in misery USA. my timezone is central daylight time which is either UTC -5 or -4 because the government hates everyone.

into trains? here's the model railroad thread!

Link to comment
Share on other sites

Link to post
Share on other sites

Create an account or sign in to comment

You need to be a member in order to leave a comment

Create an account

Sign up for a new account in our community. It's easy!

Register a new account

Sign in

Already have an account? Sign in here.

Sign In Now

×