The Threadripper thread

[PetrolHead]

It would be interesting to see what Threadripper can do in scientific computing. Hardware Unboxed had one Excel Monte Carlo benchmark, but it would be nice to be able to put the CPU through some proper simulation runs with both custom codes and commercial simulation software.

 

[TimoneX]

The chip is a monster for physics maths and CPU based FP calculations in general. Geekbench 4 shows some great numbers in several areas. Given its pricetag, it could be a game changer that will open up levels of performance at price points we haven't seen before. Hopefully people give it a chance despite its uninspiring performance at facebook farming.

https://browser.geekbench.com/v4/cpu/3696775

 

[Super Nade]

Is the 1900 (not X) slated for release soon, or is it merely a rumor?

 

[macklin01]

I've been itching to try one of these for our computational biology simulations, where we simulate millions of cells (plus all their physics) in 3D. We parallelize by OpenMP, so in principle, we should get good gains out of something like this.

I've open sourced our simulation software, so if anybody wanted to benchmark with these CPUs as they come out, I'd be very interested in working with you!

We could create a standardized test problem ...

Examples on YouTube:

1) Simulation of colon cancer metastases in a large liver tissue

2) Simulation of a 3-D cancer assay:

3) 3-D simulation of DCIS (an early breast cancer):

More videos here: https://www.youtube.com/playlist?list=PL1IHi3Kb0zyn-HBXeMBLjTU_-rFEKqFKM

- - - Updated - - -

Timonex, thanks for pointing me to geekbench. Nice to see some benchmarks beyond gaming engines. (I've previously used matlab's "bench", but it's not great.)

 

[wingman99]

Hi macklin01 nice to see you again. I have not seen you in a long time.

 

[||Console||]

Im still waiting to see FAH PPD . If any one get one please let us know the ppd .

 

[macklin01]

Hi macklin01 nice to see you again. I have not seen you in a long time.

Thanks -- great to see you, too! And we can thank c627627 for getting my attention with a reported thread. Started browsing for the first time in ages. :-D

 

[Alaric]

At least one gaming review has been changed http://www.anandtech.com/show/11726...-game-mode-halving-cores-for-more-performance

 

[Mr Alpha]

Interesting how NUMA + SMT off seems to be a better choice than Game mode (aka NUMA + cores on one die disables).

 

[Alaric]

I find the whole debate amusing. If you're gaming with enough GPU horsepower for your resolution then you would about have to be using an i3 with no HT before the CPU actually screwed up the game. At 60 Hz on my 1080p display I could probably get away with a 2c/2t i3 and not drop below 60 fps in almost any game. The ones I play I probably wouldn't drop below 100 fps. Heck, I could manage with a new Pentium. I think most people gaming with a Threadripper rig will have some serious juice in the graphics department, or the difference between 150 fps and 120 fps just won't be a priority for them. Either way, Threadripper seems like a beast for its intended purposes and a bang per buck winner, too.

 

[mackerel]

I don't think the "modes" message was communicated well. This is a processor targeted at thread count so I'd imagine most would want to run it all out. Well, if I had one, I'd leave it in NUMA+SMT for best general multi-thread performance, and if some games take a minor hit, so be it. OTOH if you really did need high fps gaming performance, this isn't the platform to buy in the first place.

 

[Tír na nÓg]

I don't think the "modes" message was communicated well. This is a processor targeted at thread count so I'd imagine most would want to run it all out. Well, if I had one, I'd leave it in NUMA+SMT for best general multi-thread performance, and if some games take a minor hit, so be it. OTOH if you really did need high fps gaming performance, this isn't the platform to buy in the first place.

Absolutely right, like getting a draught horse for horse racing (well a draught horse on steroids though!).

And as you say, minor hit in games. Particularly for most of this forum users, who play more and more on 1440p and 2160p.

 

[PetrolHead]

Particularly for most of this forum users, who play more and more on 1440p and 2160p.

I don't think I've seen any benchmarks to support this conclusion. What are use basing this statement on? In general, the higher the resolution, the more the performance depends on the GPU and thus the smaller the differences between CPUs. The effect and magnitude of GM varies from benchmark to benchmark, but it is the most noticeable at 1080p and qualitatively it seems to give better 0.1% and 1% lows while lowering the average FPS slightly. Whether or not this is true for any particluar case depends on the game, the GPU and the settings. For example in Gamer's Nexus' reviw GTA V saw a clear boost in 0.1% and 1% lows at 1080p, but in Anandtech's review the GM didn't in practice affect 1% lows at all with a GTX 1080, which is what Gamer's Nexus was also using. However, the effect was seen on the R9 Fury and the RX 480, but with those also the average FPS increased, which was not the case in GN's bencmark.

 

[Tír na nÓg]

I don't think I've seen any benchmarks to support this conclusion. What are use basing this statement on? In general, the higher the resolution, the more the performance depends on the GPU and thus the smaller the differences between CPUs. The effect and magnitude of GM varies from benchmark to benchmark, but it is the most noticeable at 1080p and qualitatively it seems to give better 0.1% and 1% lows while lowering the average FPS slightly. Whether or not this is true for any particluar case depends on the game, the GPU and the settings. For example in Gamer's Nexus' reviw GTA V saw a clear boost in 0.1% and 1% lows at 1080p, but in Anandtech's review the GM didn't in practice affect 1% lows at all with a GTX 1080, which is what Gamer's Nexus was also using. However, the effect was seen on the R9 Fury and the RX 480, but with those also the average FPS increased, which was not the case in GN's bencmark.

I think that I didn't make myself clear, as it is exactly what I mean: @1440/2180p, the threadripper lower perf doesn't impact the fps.

 

[PetrolHead]

Apologies, I put the emphasis on the wrong word and misunderstood. Mea culpa.

 

[Tír na nÓg]

No problem!

 

[TimoneX]

I've been itching to try one of these for our computational biology simulations, where we simulate millions of cells (plus all their physics) in 3D. We parallelize by OpenMP, so in principle, we should get good gains out of something like this.

I've open sourced our simulation software, so if anybody wanted to benchmark with these CPUs as they come out, I'd be very interested in working with you!

We could create a standardized test problem ...

Examples on YouTube:

1) Simulation of colon cancer metastases in a large liver tissue

2) Simulation of a 3-D cancer assay:

3) 3-D simulation of DCIS (an early breast cancer):

More videos here: https://www.youtube.com/playlist?list=PL1IHi3Kb0zyn-HBXeMBLjTU_-rFEKqFKM

- - - Updated - - -

Timonex, thanks for pointing me to geekbench. Nice to see some benchmarks beyond gaming engines. (I've previously used matlab's "bench", but it's not great.)

Yw, I like to see benchmarks that go a bit beyond fps in one shoot em up or another. lol

I have a 1950x and an x399 board I'm currently working on setting up. Well, the board is running on a bench now, but the water loop is taking some time to setup. I'd be happy to run a few benchies and supply the results.

Is the 1900 (not X) slated for release soon, or is it merely a rumor?

Just a rumor so far as I know. The 1900x comes out end of the month. It's easy to imagine AMD will start announcing non-x parts after all the fancy XFR chips are out in the wild...lest they have another 1800x/1700 fiasco.



Quad channel ram is still @ 2133mhz. I'll fiddle with that next.

 

[TimoneX]

The best I can get out of my "value ram" 64GB kit is 2933Mhz 14-14-14-14-34 1T with current AGESA code. This however has resulted in some nice gains. Still tuning as time allows, but this is my favorite power consumption/noise vs performance compromise thus far. Still waiting on threadripper optimized water blocks to become available. The borrowed cheapy block I have strapped on there now is insufficient to remove heat quickly enough at higher(> 1.35) voltages and sustained full loads. AMD isn't kidding when they claim threadripper cores are the top 5% bins. They get higher speeds than Ryzen at lower voltages.



As you can see, R15 came up with nearly the same result despite the system being clocked 100Mhz slower(3.9Ghz), so the boost going from loose 2133Mhz RAM timings to tight 2933Mhz timings appears significant. I've done a bit of testing on latencies with UMA/NUMA switching at these settings as well and the reductions there are also significant. The rig is dead silent with these settings as the 420mm and 3x140mm SP fans have no trouble keeping up. Hopefully I can push back up to 4Ghz while remaining silent with a full cover water block. This rig may well end up serving double duty as a HTPC centerpiece so silence has been a goal I've maintained throughout the component selection process. Fractal Design 140mm venturi fans kick ***...jis sayin.

 

[Super Nade]

The 1900X has been officially unwrapped. I don't think you can find it in retail just yet.

http://www.techradar.com/news/amd-releases-budget-8-core-ryzen-threadripper-1900x-processor
http://www.anandtech.com/show/11794/now-shipping-amd-ryzen-threadripper-1900x-ryzen-pro-cpus

I am looking to upgrade soon. Gotta decide on which route to take 1700x or 1900x? Main applications are number crunching (MATLAB/LabVIEW) with some gaming thrown in (not a priority).

 

[wingman99]

The 1900X has been officially unwrapped. I don't think you can find it in retail just yet.

http://www.techradar.com/news/amd-releases-budget-8-core-ryzen-threadripper-1900x-processor
http://www.anandtech.com/show/11794/now-shipping-amd-ryzen-threadripper-1900x-ryzen-pro-cpus

I am looking to upgrade soon. Gotta decide on which route to take 1700x or 1900x? Main applications are number crunching (MATLAB/LabVIEW) with some gaming thrown in (not a priority).

What does MATLAB/LabVIEW run for a CPU temperature?