Folding Forum

I'd be more intrigued by the 24pcie lanes and pcie 4.0 support, even when running GPUs at pcie 3.0 speeds.

As for CPU folding, it'll be only slightly faster than similar Intel CPUs.
The magic lies in it's ability to run more GPUs at near to full speed, than Intel.

If pcie 4.0 can be ran simultaneously with pcie 3.0, it could mean one could transfer their current GPUs to the AMD motherboard, AND add a few RX5000 cards additional at pcie 4.0 4x/1x slot, without any pcie speed bottlenecks.
It would be interesting for those who would be folding in the 8+M PPD range (1000+ Watt).

It seems everything is leaning towards this summer.
The motherboards have been introduced, though no pricing yet,
The CPUs will be launching on 7/7, at which time AMD is also aiming for their RX graphics cards, supporting pcie 4.0, and competitors to an RTX 2070 or 2080, both in performance and possibly in power consumption as well.
7/7 could mean we'd see them as soon as August on major sites, like amazon or eBay.
Early adopters could share their experience, while it would make more sense for the rest of us to wait until September or October, when supply and demand has balanced out, and prices come down.

https://www.pcgamesn.com/amd/navi-rx-50 ... rmance?amp

These are interesting times!

Nathan_P wrote:PCIe 5.0 is supposed to be for the server market, don't expect it in the consumer space anytime soon.

This article says expect pcie 5.0 in the next 2 years:

https://videocardz.com/press-release/pc ... ifications


Though I've noticed that even an RTX 2060 runs without too much performance losses on a pcie 3.0 1x slot in Linux. A PCIE 3.0 1x slot starts to restrict a 2070 or higher in Linux.

Another indication that pcie 4.0 benefits most, those who are serious about folding, and run power draws close to 1kW; 6M PPD or up; as there are plenty of Intel motherboards with 3x full size pcie 3.0 slots at 8x/4x/4x speed, good for 3M to ~7M PPD, without restricting the GPUs.

PCIE 4.0 4x slots should also be future proof up to GPUs running twice to thrice the performance of an RTX 2080TI.
So PCIE 5.0 might not really be necessary for most people, unless a new type of GPU is developed that will cut power by 3 to 4x per card. I don't immediately see this happen, unless there was a new material invented to make CPU and GPU chips.

I'm holding off, my main rig is a 1920x on x399 so waiting for that to be announced. My gpu rigs just run gpu so no upgrades for the systems until the boards fail or cpu folding becomes a serious thing again. My testing gave me a 2% gain going from 2.0 x16 to 3.0 x16 so 4.0 will make little difference.

pcie 4.0 is more interesting for x1 risers on Linux. Or running pcie 4.0 x4 on Windows without performance loss.

A massive CPU like this, 6+ cores (with SMT, also called Multi-threading; AMD's variant of Hyper Threading), you don't really want to use for folding.
However, at 7nm it might be close to what older Pascal cards, or AMD Radeon cards can offer, in terms of efficiency.
Most CPUs fold at about the same speed as a budget graphics card of last generation (GT 1030, GTX 1050, RX550,...)


This thread, after only a week, seems already passe, now that PCIE 4.0 is on the horizon!

PCIE 3.0 will be enough for people running 2 to 3 graphics cards for folding.
I agree also that 5.0 will probably not be necessary for most regular people; as PCIE 4.0 is good enough for running 3 fast GPUs (like 2080 Ti), and 3 mid-tier cards like RTX 2060-2080s completely unrestricted, and that such a system would probably run at ~1600W continuous (~900W when power capped).
That's more than what most people are willing to run on their fuses.

MeeLee wrote:A massive CPU like this, 6+ cores (with SMT, also called Multi-threading; AMD's variant of Hyper Threading), you don't really want to use for folding.
However, at 7nm it might be close to what older Pascal cards, or AMD Radeon cards can offer, in terms of efficiency.
Most CPUs fold at about the same speed as a budget graphics card of last generation (GT 1030, GTX 1050, RX550,...)


This thread, after only a week, seems already passe, now that PCIE 4.0 is on the horizon!

PCIE 3.0 will be enough for people running 2 to 3 graphics cards for folding.
I agree also that 5.0 will probably not be necessary for most regular people; as PCIE 4.0 is good enough for running 3 fast GPUs (like 2080 Ti), and 3 mid-tier cards like RTX 2060-2080s completely unrestricted, and that such a system would probably run at ~1600W continuous (~900W when power capped).
That's more than what most people are willing to run on their fuses.

That depends on what voltage you are running, in the UK 1600w on one socket is nothing.

In Germany it is max 3500 watts 16A at 230V in one circuit and each home has several of these.

Common household circuits have a !0A or 15A, world-wide ... but in North America, the mains are typically 120V, 60 Hz. For heavy duty devices (electric stove, air conditioner, electric car refueling, etc.) we can install a 240V circuit with a special plug or meet other needs with other types of special circuits. Even portable electric space heaters are typically limited to 1000W.

Chargers for laptops/phones/etc. adapt to various national standards. (read the label).

bruce wrote:Common household circuits have a !0A or 15A, world-wide ... but in North America, the mains are typically 120V, 60 Hz. For heavy duty devices (electric stove, air conditioner, electric car refueling, etc.) we can install a 240V circuit with a special plug or meet other needs with other types of special circuits. Even portable electric space heaters are typically limited to 1000W.

Chargers for laptops/phones/etc. adapt to various national standards. (read the label).

Also remember (at least in the US) that the Standards organization (NEMA here) recommends you do not load the circuit over 80% as a continuous load, so while a Hair Dryer can be 1875 Watts as you don't use it long, your Folding computer should not be over 1500 watts on a dedicated circuit, less if you have other items plugged in as well. (monitor?)

... and most are not dedicated circuits. An older house might have two or three rooms on a single circuit except for the kitchen or laundry room where there may be a circuit for each major appliance. Nowadays, single circuits for those appliances is mandatory.

Yes, it's possible for people in USA, to run a secondary phase, and run a 2000-3000W system
However, most households have more than just a PC running.
Especially heaters and AC units, as well as stove, ovens, hairdryers, or electric car chargers, they all consume a massive amount of power.

Not to mention, the cost of running 1000W continuously, is about $100 a month extra.

Theodore wrote:Not to mention, the cost of running 1000W continuously, is about $100 a month extra.

Not in my house. Solar panels have paid for themselves over the years.

If I ever get to buy a home of my own, I'll install solar panels as well.
Some panels give a 12-16V DC output, which is great to directly feed the 12V line of graphics cards with (with the aid of some voltage regulators); in parallel with the PSU's 12V lines.
It'll lower the PSU load.
No DC to AC conversion losses, like when trying to convert the 12V DC to 120V AC, to feed the PSU, which downconverts it back to 12V DC anyway.
I'm currently running 650 Watts, and if my system could take another graphics card, I would, but it's only a quad core.

If it takes about 2 years for PCIE 4.0 to be adopted, It'll probably take another 2 years, before manufacturing will be ready for PCIE 5.0 GPUs and hardware.

So I won't hold back, waiting for PCIE 5.0 motherboards.
And like mentioned, PCIE 4.0 1x slots are fast enough for RTX 2060 and 2070 cards.

The only question I might have, for linux users, is that Ubuntu 18.04 or 18.10, with Linux kernel 4.x will probably not support PCIE 4.0 out of the box.
We might need Linux kernel 5, or higher; which would be Ubuntu 19 or higher; both of which will most likely not support Python 2.
Linux users might be left out, and Windows does need at least PCIE 4.0 2x or 4x slot speeds for RTX graphics cards.
Then again, most motherboards on the market support either full sized PCIE slots, or an m.2 slot (which can be converted with a pcie 4x adapter), so it'll still allow one to plug in up to 3 full size GPUs and up to 3x GPUs with PCIE 4x slots.

Seems like I made an error in a few previous assumptions, which I'd like to correct here; as well as share some more findings:


1- It seems that all AMD PCIE 4.0 motherboards made today, use active cooling of up to 15 Watts. Some go even as far as to cool parts on the motherboard (not CPU or GPU) with liquid cooling!
I don't think this is very interesting for folding 24/7 to be using active (nor liquid) cooling, unless the overall power consumption of the system is so high that the extra consumption would disappear in the total.
I wouldn't buy such a system with the purpose for 24/7 folding on a single or double graphics card setup, as the 15 watts is unnecessary overhead.
Although many cooling solutions will use less than 15Watts.

2- Another interesting thing to note, is that no one knows yet, if PCIE 4.0 (which no doubt will be faster in throughput), has increased latency compared to PCIE 3.0?
For folding, raw throughput of a PCIE 3.0 1x (or higher) slot on Linux, isn't as important as the latency, as an RTX 2060 will perform almost as fast on a PCIE 3.0 1x slot, as it does on a PCIE 3.0 8x or 16x slot.
Running these cards on PCIE 4.0 compatible motherboards, will still cap them at PCIE 3.0 speeds; but no word on additional latencies on the newer hardware.
It remains yet to be seen, if PCIE 4.0 will accelerate the folding process by sheer throughput, and eliminating any overhead bottlenecks, compensating for any potential latencies.

3- One error I'd like to correct on my previous assumption, is that Intel does support more than just 16x PCIE lanes!
It's just not so well documented as AMD.

On current AMD offerings (using the PCIE 3.0 standard), AMD uses 16 PCIE lanes for graphics cards (to the two main GPU slots on the motherboard), and the additional lanes for m.2 or other PCIE slots on the board (whatever configuration the manufacturer made on the motherboard).
The first 16 PCIE lanes shared between the first two full size slots on the motherboard, are PCIE 3.0 compatible.
The other lanes are all PCIE 2.0.
The PCIE 2.0 1x slots (250MB/s) DO slow down Turing graphics cards, but m.2 2.0 4x slots (1GB/s) offer sufficient bandwidth for up to an RTX 2080.
(for reference, an RTX 2060 folding at PCIE 3.0 1x (500MB/s) gets 1.04-1.2M PPD, while it only gets 875-980k PPD on PCIE 2.0 1x (250MB/s) slot.
That's a 10-15% loss in PPD! While running this card on PCIE 3.0 4x slots or higher, nets only a few percent higher PPD, not as much as the 10% drop seen before)


Intel’s 6th Generation processors (Skylake) and up, allegedly supports the same amount of PCIE lanes, but all at 3.0 speeds, sources:
PCIE lanes explained under subsection "I/O Interface Technology"
Intel 6th gen processors WIKI

It's not well documented, but Intel supports the same amount of PCIE lanes as AMD, starting from Skylake (6th gen) CPUs; with the difference that ALL the PCIE lanes on Intel Skylake (6th gen) and up, run at the full PCIE 3.0 speeds!

This will actually put Intel in an favorable light, and AMD in a disadvantage, and unless PCIE 4.0 offers significant improvements over PCIE 3.0 for folding (which I don't think it would), I wouldn't choose an AMD system for a 24/7 folding computer; but I'd stick with Intel CPU PCs for now!
And preferably a 9th generation Intel CPU (but could also be 8th, 7th or 6th gen), of low TDP (preferably a mobile CPU of 25 or 45W, if possible, or a 65W desktop CPU), of 3 to 4Ghz, with 1 core per GPU if you're running Linux; and try to stay clear of the 95+ Watt, multi-core, high frequency CPUs that will only drain power out of the wall, without adding any folding performance gains.

(Side note: Due to chip shortages, at the time of writing, Intel CPUs have increased by 10 to 15% in price compared to the last month or two. I'm expecting those prices to go down, as soon as production meets demand).


4- I don't think there's a lot of documentation out there, on PCIE speeds and what modern graphics card is limited by them.
From my findings, an RTX 2060 or 2070 needs a PCIE 2.0 2x slot speed, or PCIE 3.0 1x speed.
Since 2x slots aren't an industry standard, these cards will need a PCIE 2.0 4x slot or higher to operate on.

A 2080 needs a PCIE 2.0 4x slot (or 3.0 4x slot) or higher; and a 2080 ti, needs a PCIE 3.0 4x speed slot or higher.
Even most gaming benchmarks show that the 2080 ti is more severely limited on a PCIE 2.0 4x slot than on a 2.0 8x, or 3.0 4x slot.

If you want a low power cpu to run your gpu folding rig look at xeon, especially older socket 115x xeon, those chips can go down as low as 25w.