*** AMD ThreadRipper ***

[theforce]

Does anyone think we will see an 8 or 16 core Zen CPU with a Vega GPU on the same die?

TR4 seems to have enough land for it.

With a 16 core Zen + a 2000ish shader Vega to occupy the place of one of the dummy cores along with 8-16GB of HBM memory for the on board Vega graphics and as an L4 cache for the CPU should be an interesting product.

Would a Vega GPU on a different die occupy the place of one of the dummy cores and be connected via IF? Or is the bandwidth provided by IF in it's current form too slow to accommodate such a configuration?

 

[4K8KW10]

G.Skill Announces AMD X399-optimized DDR4-3466 32GB (4x8GB) Trident-Z Kit

 

[BongoHunter]

Now that 32GB DDR4 modules are available from Samsung & co. do we expect TR boards to support them with a BIOS update? Would love to move to 256GB of RAM in my next build.

 

[sandys]

G.Skill Announces AMD X399-optimized DDR4-3466 32GB (4x8GB) Trident-Z Kit

Latencies are quite high, will it even be worth it.

 

[4K8KW10]

Now that 32GB DDR4 modules are available from Samsung & co. do we expect TR boards to support them with a BIOS update? Would love to move to 256GB of RAM in my next build.

The max is 128 GB, and if you install 2 kits of 32 GB (4 X 8 GB), you will have 64 GB.

From your signature, I see you already have so.

 

[4K8KW10]

Latencies are quite high, will it even be worth it.

Can be. Someone needs to test them and see if higher frequencies with the same latencies will be stable.

 

[Freddie1980]

Surely any memory/RAM optimisation needs be done at a BIOS level and/or CPU microcode? Sounds like marketing gimmick to me G-Skill.

 

[sandys]

Surely any memory/RAM optimisation needs be done at a BIOS level and/or CPU microcode? Sounds like marketing gimmick to me G-Skill.

They have mentioned board partner which would suggest appropriate timings will be in that board BIOS for stability and performance.

 

[Deleted member 66701]

Can be. Someone needs to test them and see if higher frequencies with the same latencies will be stable.

I'm running 16-18-18-18 @ 3466 on cheapo Avexir ram so I'm not sure what the point of that ram is.

 

[Journey]

I'm running 16-18-18-18 @ 3466 on cheapo Avexir ram so I'm not sure what the point of that ram is.

To charge people more for something they can get cheaper, marketing at it's finest.

 

[BongoHunter]

The max is 128 GB, and if you install 2 kits of 32 GB (4 X 8 GB), you will have 64 GB.

From your signature, I see you already have so.

Yus - sorry I was not referring to that new kit that was posted (4*8gb), I was wondering if anyone had seen support for 32gb modules discussed anywhere. As I understand it TR supports way more than 128gb, but the limitation at this point is with the motherboard vendors (TR CPU supports 512gb per memory channel theoretically) - though that will never be achieved as no one offers TR board with support for registered memory.

But Samsung did recently announced 32GB unbuffered dimms (highest capacity yet achieved without rdimms):
https://www.samsung.com/semiconductor/dram/module/M378A4G43MB1-CTD/

Intel have stated they will support the new modules with a BIOS update, at least on some of their newest platforms like Xeon E for example - doubling the memory capacity of some of the boards to 128gb from the 64gb that most of the 4 bank boards currently support. Hoping AMD camp will be able to do something similar at some point

 

[4K8KW10]

AMD has confirmed that it is working hand-in-hand with Microsoft to address performance issues brought around by the use of non-uniform memory access (NUMA) in its high-end many-core Threadripper and Epyc processors under the Windows operating system.
https://bit-tech.net/news/tech/cpus/amd-microsoft-working-on-threadripper-epyc-performance/1/

 

[Undesirable]

G.Skill Announces AMD X399-optimized DDR4-3466 32GB (4x8GB) Trident-Z Kit

They must not hold AMD users in very high regard if they release a RAM kit with as slack timings as that. Bit of a joke unless it's ECC memory or something.

 

[smogsy]

AMD has confirmed that it is working hand-in-hand with Microsoft to address performance issues brought around by the use of non-uniform memory access (NUMA) in its high-end many-core Threadripper and Epyc processors under the Windows operating system.
https://bit-tech.net/news/tech/cpus/amd-microsoft-working-on-threadripper-epyc-performance/1/

wonder if this will help the 1950x/2950x in NUMA mode also

 

[HEADRAT]

Posting this here as I have a problem with a Threadripper 1950x, this is really now a cry for help, I've tried multiple MOBOs/PSU/Memory in countless combinations but I can't get this sucker to work. I've sunk litteraly 100s of hours and £££ into this I've been building PCs since the 1980s but honestly this has me beat!

So far I've tried

• Different Memory in many combinations
• Different PSUs
• Different Taichi Mobo
• Aorus X399 Mobo
  
• Different GFX cards
• Removing Mobo battery etc.

I'd convinced myself that the MOBO was the issue so I bought another board, plugged it all together and I get exactly the same error code (90)!!! (you could have knocked me down with a feather).

I've now purchased a Aorus X399 AORUS PRO and I still get the some Code 90 error!

Can't really believe my luck as it had been working up until I upgraded to the 3.60 BIOS (Taichi), the PC just crashed while unattended last and now fails to start! (error code 90, which I have had ever since)

HELP!!

Cheers

HEADRAT

 

[sandys]

Have you tried a bios flashback to an older version of BIOS? I used this trick to revive mine once.

 

[anthony99]

Sounds like flashback is your only hope

 

[Journey]

Going to sound really, really dumb, but did the post code change if your remove all of the PCI-E devices including the GPU?

 

[HEADRAT]

Going to sound really, really dumb, but did the post code change if your remove all of the PCI-E devices including the GPU?

Yes, if I remove different components it get stuck on something other than 90, e.g. if I pull all the memory it gets stuck on Q-Code 16.

In regards to flashback I'd try that if it was just one board (in fact the original board is in the process of being RMAd as it won't flashback) but I'v had the same error code (90) on two brand new MOBOs:

1 x Taichi x 399
1 x Aorus X399

I'm really leaning towards that the CPU could be knackered (it's a OEM chip so no warranty, hence why I'm trying everything to get it working).

Thanks all for your suggestions as I'm really left "scratching my head".

 

[FloppyPoppy]

Sounds like the cpu is goosed above all else. What's the warranty situation on it?