Discussion AMD Drops Third-Gen Threadripper From Latest Roadmap

Page 2 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.

akamateau

Reputable
Jun 8, 2015
88
5
4,635
It appears the consensus is that EPYC adoption rates are going to demand all RYZEN 3000 TR HEDT chiplets. This may be so but from perhaps another source entirely.

You see both Amazon AWS and Microsoft AZURE are the two finalists for the $10BILLION DOD JEDI cloud services contract which incidentally requires the construction of at least 6 new data centers for Secure and non-secure cloud networks.

https://www.fbo.gov/index?s=opportu...0314ec7c83cd0ace1636b5474a1&tab=core&_cview=0

Despite some drama from Oracle, the GSA has been directed to continue with the award not sooner than July 19.

https://federalnewsnetwork.com/defe...f-jedi-lawsuit-decision-expected-by-mid-july/

In fact Microsoft is buying up land ostensibly to demonstrate the ability to immediately move forward. Amazon likely has been as well, perhaps some of this has been under the cover of locating a new headquarters.

AMD however is a partner with both AWS and AZURE as they are both EPYC fans.

Perhaps AMD is seeing a unexpected demand for EPYC silicon due to one or the other JEDI wins?

For AMD, JEDI might be a WIN-WIN.
 
Last edited:
How do you know it's not if you don't know what it is?
Certifications involves the company getting some sample CPUs and testing the code they will run on their own so that they exactly know what they are dealing with.

Because nothing uncertified ever makes it to servers.
search for certified servers or software to find results.

I was just making a point that they will buy anything that would work for them if ARM is something they could use they would buy it.


Dude, quit before you make a total fool out of yourself.
 
QFT. But then again the kind of people who really need TR bandwidth who aren't professionals (with an EPYC budget) are a small portion of the population. The later statement is pure conjecture on my part.

While the notebook & homebuilt market is picking up, and flocking to AMD at the same time, I think the number of people owning Thread rippers who need a core update with the available bandwidth would be small. Finding a prebuilt TR system from the big players is like finding teeth on a hen. It's like those guys running E3 XEONs. Few and far between need that kind of power for home use.

Not that I don't empathize with you guys who do things like 3D render, rip movies or run VMs at home. Maybe when there are more 7nm chips available, you guys will get your updates.

Threadripper does have quite an enthusiastic following among those who run software that can take advantage of Massively Parallel Processing. Thousands of TR's have been sold to DIY enthusiasts who are professional chessplayers. Chess is one game that can fully apply upwards of 64 cores and 128 threads to deep analysis. And chessplayers always look for the fastest silicon and they are multiple buyers.

While it may not be necessary to provide yearly core upgrades to TR that simply happens naturally as a result of EPYC MCM production. The only "new" silicon would be the active interposer that is the substrate for TR. It would not be in AMD's best interest to over produce chiplets simply to maintain an inventory of TR over the years.

One could rightfully state that in general, the current state of the art of programming does not make use of MPP as raised to the level of TR.

Well if there is no TR then there is also no demand to raise the programming bar to meet it. That is the whole point of Field of Dreams. Build it and they will come.

I remember when the first 100mb hard drive came into the office I worked back in the early 90's! What were we going to put in all that space?

I also remember the massive debate regarding the usefulness of 64 bit programming. It took years to make the transition from 32 bit to 64. That never would have happened if 64 bit did not get built in.

If it is ANYTHING, TR is the tail that wags the dog.

TR marketing could be compared to Jeep's Grand Cherokee Trackhawk. How many folks are going to spend $90k on a 4x4 packing a 707 horsepower 6.7L Hemi V-8? Is it necessary for soccer mom's to take their kids to practice? NO!!! But I bet it is a whole lot of fun to drive!

So Trackhawk like Threadripper may not be suited for all workloads but, it does have a place in the market!

And there will be a new one for 2020 too!
 
Last edited:
Certification is a long process that dtarys with analysis amd then a test cluster group. But it usually and basically boils down to to total cost of ownership (tco) and reliability. The lifetime of a clister swap for a white or grey room can be up to six years or more.

Im sure AMD practically threw epyc racks at aws, google, and azure to get the process started. And I dont blame them. I convinced someone in our it dept of over 100k people to go to virtual instances of our compilers using TR. You had no idea what a tooth pull that was even as a lead engineer.

It departments tendto be slow on i stalling new server hardware even if the need is critical.