[Official] New Egg shuffle thread

[Avacado]

28 minutes ago, BWG said:

This is me today!

 

https://www.newegg.com/msi-geforce-rtx-3070-rtx-3070-gaming-x-trio/p/N82E16814137603?Item=N82E16814137603

 

 

Grats bro, I entered too. Hope I win as well.  You got that winning alert fast, I usually don't get any emails until like 9pm. 

 

Edited May 6, 2021 by Avacado

[BWG]

Really neat experience. I got some ram in the combo. Also, a free MSI Gamepad.

[BWG]

@AvacadoI have an open pci-e x16 @ x4 2.0 slot on my x370 board which is kind of strange. Do you think it will bottleneck it? 

Edited May 6, 2021 by BWG

[Avacado]

15 minutes ago, BWG said:

@AvacadoI have an open pci-e x16 @ x4 2.0 slot on my x370 board which is kind of strange. Do you think it will bottleneck it? 

Well, I would recommend throwing one of your Ti's on that and swapping the 3070 to something higher bandwidth. Is it going to be your main GPU? or just for folding on? Short answer, no, I don't believe it would bottleneck a 3070 folding, but anything else, yea, for sure. I still have yet to land any 3 series cards, so I appreciate the ask, but I am not qualified to really answer that in lieu of proper experience. Maybe @Bastiaan_NL could render some advice on that. 

Edited May 6, 2021 by Avacado

[BWG]

I'll drop a 1070 down to that slot and put it in the top slot at 16x 3.0

 

I can already see this turning into another GPU bench situation lol. Looped in though since the 3070's an air card. Man, it's been a long time since I bought an air cooled GPU! I think the last ones I bought were dual 460's in SLI which ended up being converted to H2O

Edited May 7, 2021 by BWG

[u3b3rg33k]

6 hours ago, BWG said:

@AvacadoI have an open pci-e x16 @ x4 2.0 slot on my x370 board which is kind of strange. Do you think it will bottleneck it? 

I thought someone went through this way, way back and found that for folding basically a 1x slot was adequate bandwidth - still needs enough CPU time to keep it fed/polled.  I'm over here with PCIe 2.0 x16 on switches and i've never noticed any issues gaming or folding. I do notice issues gaming WHILE folding, especially if I forget to pause folding on my gaming GPU...

[Avacado]

5 hours ago, u3b3rg33k said:

I thought someone went through this way, way back and found that for folding basically a 1x slot was adequate bandwidth - still needs enough CPU time to keep it fed/polled.  I'm over here with PCIe 2.0 x16 on switches and i've never noticed any issues gaming or folding. I do notice issues gaming WHILE folding, especially if I forget to pause folding on my gaming GPU...

Each generation effectively doubles GB/s. Most likely will be fine if indeed he can get the PCI-e 4.0 interface to work as intended. My guess is that 4.0 1x will be "fairly"close to what we see at 3.0 4x speeds folding. Again, I have not had the pleasure of playing with 4.0 or any 3 series cards, so I couldn't tell you, that is why I asked Basstiaan to comment, I think he ran these experiments not to long ago. 

 

Of course you wouldn't notice any issues on 2.0 @ 16x, I wouldn't expect you to, but if you folded 2.0 @ 1x, compared to 3.0 @ 1x there should be a difference, we are talking about double the bandwidth between generations, this is why I asked what his intended purpose for the new 3070 would be.

 

 

Edited May 7, 2021 by Avacado

[tictoc]

12 hours ago, BWG said:

@AvacadoI have an open pci-e x16 @ x4 2.0 slot on my x370 board which is kind of strange. Do you think it will bottleneck it? 

 

I tested this last year with a 2080 Super.  Performance would be somewhere between the 3.0 x1 and 3.0 x4 results, so probably a 15-20% hit in ppd.

 

OS: Arch Linux

Kernel: 5.6.14

Driver: 440.82

GPU: RTX 2080 Super

 

p13406 (core_22)

PCIe 3.0 x16: TPF - 01:09 | ppd - 2,512,601 || PCIe Utilization - 18% | GPU Utilization - 97% | GPU Power - 220W | Clocks - 1935core|7500mem

PCIe 3.0 x4: TPF - 01:12 | ppd - 2,357,211 || PCIe Utilization - 34% | GPU Utilization - 95% | GPU Power - 217W | Clocks - 1935core|7500mem

PCIe 3.0 x1: TPF - 01:28 | ppd - 1,744,509 || PCIe Utilization - 50% | GPU Utilization - 84% | GPU Power - 190W | Clocks - 1935core|7500mem

PCIe 2.0 x1: TPF - 01:50 | ppd - 1,248,271 || PCIe Utilization - 55% | GPU Utilization - 77% | GPU Power - 168W | Clocks - 1965core|7500mem

Edited May 7, 2021 by tictoc

[BWG]

I decided to switch the whole idea around and utilize the test bench instead for my 4.0 main rig at x1. 3080 on slot 1, 3070 on slot 2, 1080ti on slot 3, 1080ti on x1 to x16 riser. I won't need a new PSU then. I have 6 plugs on the 1050w and 4 plugs on the bench. Perfect!

 

Props to DHL!  EKWB shipped the plexi tops for the ti's on Wednesday from Slovenia, and they arrive today for only 16 bucks in shipping cost.

 

3070 did not get to UPS yesterday, so now it will arrive on Monday. Thermal pads and fan controller Saturday. This gonna be dope! lol

[Bastiaan_NL]

Sorry for the delay, a 'nice' project at work is taking most of my time and energy.

 

I only looked at the bus interface load on different cards at a few different systems but I never did a decent test because I could not get the FAHBench to work properly with a copy of a work unit. 
The unit on my main system is finishing as we speak, after that I'll give it another try to see if I can get something to work. 

 

 

4 hours ago, Avacado said:

Again, I have not had the pleasure of playing with 4.0 or any 3 series cards, so I couldn't tell you, that is why I asked Bastiaan to comment, I think he ran these experiments not to long ago. 

I'd let you do some testing with the cards if I was living in the dungeon

 

 

3 hours ago, tictoc said:

 

I tested this last year with a 2080 Super.  Performance would be somewhere between the 3.0 x1 and 3.0 x4 results, so probably a 15-20% hit in ppd.

 

OS: Arch Linux

Kernel: 5.6.14

Driver: 440.82

GPU: RTX 2080 Super

 

p13406 (core_22)

PCIe 3.0 x16: TPF - 01:09 | ppd - 2,512,601 || PCIe Utilization - 18% | GPU Utilization - 97% | GPU Power - 220W | Clocks - 1935core|7500mem

PCIe 3.0 x4: TPF - 01:12 | ppd - 2,357,211 || PCIe Utilization - 34% | GPU Utilization - 95% | GPU Power - 217W | Clocks - 1935core|7500mem

PCIe 3.0 x1: TPF - 01:28 | ppd - 1,744,509 || PCIe Utilization - 50% | GPU Utilization - 84% | GPU Power - 190W | Clocks - 1935core|7500mem

PCIe 2.0 x1: TPF - 01:50 | ppd - 1,248,271 || PCIe Utilization - 55% | GPU Utilization - 77% | GPU Power - 168W | Clocks - 1965core|7500mem

I want to do the same with the 3090.
It's interesting that the TPF is 41 seconds higher with 55% PCIe utilization, I expected a higher utilization for such a significant increase in TPF.

[Avacado]

I honestly don't think I will ever win a shuffle, because i'm unwilling to buy bundles I don't need like most others. Sure I could click them all, but why? The only bundles I click are the ones with PSU's, cause who doesn't need more PSU's. I did win that 3060, but that doesn't count right? I need a 3070 at minimum to replace a 2080Ti. 

Edited May 7, 2021 by Avacado

[u3b3rg33k]

8 hours ago, tictoc said:

 

I tested this last year with a 2080 Super.  Performance would be somewhere between the 3.0 x1 and 3.0 x4 results, so probably a 15-20% hit in ppd.

 

OS: Arch Linux

Kernel: 5.6.14

Driver: 440.82

GPU: RTX 2080 Super

 

p13406 (core_22)

PCIe 3.0 x16: TPF - 01:09 | ppd - 2,512,601 || PCIe Utilization - 18% | GPU Utilization - 97% | GPU Power - 220W | Clocks - 1935core|7500mem

PCIe 3.0 x4: TPF - 01:12 | ppd - 2,357,211 || PCIe Utilization - 34% | GPU Utilization - 95% | GPU Power - 217W | Clocks - 1935core|7500mem

PCIe 3.0 x1: TPF - 01:28 | ppd - 1,744,509 || PCIe Utilization - 50% | GPU Utilization - 84% | GPU Power - 190W | Clocks - 1935core|7500mem

PCIe 2.0 x1: TPF - 01:50 | ppd - 1,248,271 || PCIe Utilization - 55% | GPU Utilization - 77% | GPU Power - 168W | Clocks - 1965core|7500mem

do you have more numbers? i'd be curious to see PCIe 2.0 x16 vs 3.0 x16.  I'm thinking this is a transaction latency thing more than raw bandwidth. 

 

also how did you get PCIe Utilization numbers? do you have a thread with all this posted I could look at?

[Bastiaan_NL]

51 minutes ago, u3b3rg33k said:

do you have more numbers? i'd be curious to see PCIe 2.0 x16 vs 3.0 x16.  I'm thinking this is a transaction latency thing more than raw bandwidth. 

 

also how did you get PCIe Utilization numbers? do you have a thread with all this posted I could look at?

We should start a thread about this, and test a few different videocards on as much settings as possible. And make sure we use the same method to do so, to get comparable results.

I don't think FAHBench is a good way to test, for some reason I don't get any load on the bus with a preprogrammed unit and a copied WU from the Fahclient doesn't work.

 

 

2 hours ago, Avacado said:

I honestly don't think I will ever win a shuffle, because i'm unwilling to buy bundles I don't need like most others. Sure I could click them all, but why? The only bundles I click are the ones with PSU's, cause who doesn't need more PSU's. I did win that 3060, but that doesn't count right? I need a 3070 at minimum to replace a 2080Ti. 

I wouldn't buy a deal with most of the PSU's either, only if it's 1kw or more. And even a 3060 is welcome here, to replace the 1650

[BWG]

Mine did come with ram.

[Avacado]

2 hours ago, u3b3rg33k said:

do you have more numbers? i'd be curious to see PCIe 2.0 x16 vs 3.0 x16.  I'm thinking this is a transaction latency thing more than raw bandwidth. 

 

also how did you get PCIe Utilization numbers? do you have a thread with all this posted I could look at?

There shouldn't be any difference between 16x 2.0/3.0, 2.0 16x its the same bandwidth as 3.0 8x, and 8x 3.0 does not suffer any penalty in bandwidth related to PPD. 

Edited May 7, 2021 by Avacado

[tictoc]

Sorry for continuing the OT.

 

4 hours ago, u3b3rg33k said:

do you have more numbers? i'd be curious to see PCIe 2.0 x16 vs 3.0 x16.  I'm thinking this is a transaction latency thing more than raw bandwidth. 

 

also how did you get PCIe Utilization numbers? do you have a thread with all this posted I could look at?

 

What @Avacado said for 2.0 x16 vs 3.0 x16.  I don't have a thread, but I did some testing for the first Community Folding Project thread.  I use nvidia-smi to log GPU data.  Some of the queries will be dependent on the GPU architecture, the driver that is being used, and the OS (Windows or Linux) being used.

Once upon a time PCIe utilization by percentage was reported by the driver (via nvidia-smi), but that is no longer the case (NVIDIA said it was noisy and not accurate).  Now you can only get the Tx and Rx throughput in MB/s. 

 

[u3b3rg33k]

"in theory" 

 

these should be the same:

Version 1.0    x1: 250 MB/s  1.0 x32 =8GB/s (lol)

Version 2.0:  x1: 500 MB/s  2.0 x16 = 8GB/s

Version 3.0   x1: 985 MB/s  3.0 x8 = 7.88GB/s

Version 4.0:  x1: 1.97 GB/s  4.0 x4 =7.88GB/s

 

but I assume there's other benefits to the multiple lanes than just raw bandwidth (transactional overhead from CPU/GPU)

Edited May 8, 2021 by u3b3rg33k