AMD Ryzen R9 3950X new build

[Illusion]

3900X= $550+$170+$90+$290+$400+$0+$85=$1585 @ ~45Mnps = ~ 28.4knps/$
3950X= $800+$170+$90+$290+$400+$90+$85=$1925 @ ~60Mnps = ~ 31.2Knps/$
3960X=$1575+$400+$100+$290+$400+$90+$85=$2940 @ ~90Mnps = ~ 30.6Knps/$
3970X=$2075+$400+$100+$290+$400+$90+$85=$3440 @ ~105Mnps=~ 30.5Knps/$


I am more than a tad confused and all the new hardware has me baffled. I am looking at http://www.ipmanchess.yolasite.com/amd- ... -bench.php and note that on my 9980 XE which I bought from Origin PC is getting me around 31-33M . This is so disappointing compared to some of the benchmarks I see. Am I looking at this correctly? Is there something I can do to get more nodes/sec. I had Origin overclock the CPU but they could not get a lot out of it.

Also, on a new system for NN engines, I see that some motherboards do have 2 slots, but don't support SLI (which is what I am used to). Do I just "plug in" 2 RTX video cards and the NN engine will see both and use them both?

Thanks....sorry I am not up to speed on this stuff. I would love to pay someone to advise me so I quit wasting money

[Dann Corbit]

Keep in mind that Ipman chess uses the assembly version of SF called AsmFish, which gets higher NPS than SF.
Try to download the kit he provides, and follow the instructions exactly. Then see what NPS you get

[brianr]

Also, on a new system for NN engines, I see that some motherboards do have 2 slots, but don't support SLI (which is what I am used to). Do I just "plug in" 2 RTX video cards and the NN engine will see both and use them both?

The NN engines like Leela (Lc0) will just use multiple GPU cards for playing games, which increases strength quite a lot, although the "backend" parameter settings can be a bit tricky to get just right. So, no SLI is needed. For playing only 2 CPUs are good (sometimes 3 for multi-GPU configs I think), although higher clock speed helps. Many CPUs can actually slow things down for Lc0.

However, for training nets, the current Lc0 training code only uses one GPU (although TensorFlow can use multi-gpu setups automatically, so Lc0 training could be modified at some point). For training nets lots of CPUs (and RAM) help.

Lastly, training nets is not the same as generating training games. With more than one GPU separate copies of the training client can run on each GPU to create games in parallel.

[Zenmastur]

I purchased a MB and a cooler for my Threadripper build. The ASROCK MB in my previous post wasn't my first choice. I was only looking at it because the board I wanted was out of stock. I checked it again this morning and it was still out of stock but they offered to back order it with a week delay which wasn't acceptable. I checked again a little while ago and it was back in stock but everything else in my shopping card had gone out of stock in the mean time.

I had decided to order memory instead of screwing around with the slow memory I already had. Finding memory that is on the MB's QVL is a pain in the ass. I found some earlier, but now it was out of stock. So, I didn't order any memory, I'll just use what I have until one of the dimm sets from the QVL is available again.

I had wanted a Noctua NH-U14S TR4-SP3 cooler but it was out of stock when I was filling my shopping cart. So I had chosen a FRYZEN cooler, but now it was out of stock as well! I was lucky the NH-U14S had come back into stock so I ordered it as well.

The MB I picked is the GIGABYTE TRX40 AORUS PRO WIFI. It's a very good MB and it's cheap. Here's a review:

I should get everything I've ordered, which is enough to get me up and running by saturday.

Regards,

Zenmastur

[Illusion]

Zenmastur,

I am sure many would love to see the build you selected, given that all the parts are obtainable....

I would model my new computer after that quite closely.

Dann Corbit too If you guys would be so kind....

[Dann Corbit]

I just went to this site, and added on the stuff that I wanted:
https://www.cyberpowerpc.com/system/AMD ... per-Xtreme

I went for a 1600 watt power supply and a 360mm water cooler, which limited my case options to just a few.

[Zenmastur]

I just went to this site, and added on the stuff that I wanted:
https://www.cyberpowerpc.com/system/AMD ... per-Xtreme

I went for a 1600 watt power supply and a 360mm water cooler, which limited my case options to just a few.

That actually looks like a "reasonable" build at a not too expensive price. I think I would change the OS. If you want to run Windows I think it's better to run Windows 10 Pro. IIRC there are restrictions that you will be bumping into with the home edition, although I would have to look it up to be sure.

I think I would change the memory if you can "IF" you are planning on upgrading it at a latter date. Other than that it looks like a completely "normal" and usable build.

Regards,

Zenmastur

[Dann Corbit]

I just went to this site, and added on the stuff that I wanted:
https://www.cyberpowerpc.com/system/AMD ... per-Xtreme

I went for a 1600 watt power supply and a 360mm water cooler, which limited my case options to just a few.

That actually looks like a "reasonable" build at a not too expensive price. I think I would change the OS. If you want to run Windows I think it's better to run Windows 10 Pro. IIRC there are restrictions that you will be bumping into with the home edition, although I would have to look it up to be sure.

I think I would change the memory if you can "IF" you are planning on upgrading it at a latter date. Other than that it looks like a completely "normal" and usable build.

Regards,

Zenmastur

Yes, I put Ubuntu on a 1 TB SSD thumb drive. Faster and better in every way. I found some that actually work and are fairly inexpensive. You have to be careful with cheap huge thumb drive. There are fakes out there that report a huge size, but are really tiny units with a fake directory structure

[Gino Figlio]

I just went to this site, and added on the stuff that I wanted:
https://www.cyberpowerpc.com/system/AMD ... per-Xtreme

I went for a 1600 watt power supply and a 360mm water cooler, which limited my case options to just a few.

Have you checked their own online reviews?
I was about to place an order but now I hesitate..

[Gino Figlio]

I forgot to post the link to the consumer affairs reviews:
https://www.consumeraffairs.com/compute ... er_pc.html