You missed that PCIe 4.0 provides double the bandwidth per lane over PCIe 3.0.
2 pcie 3.0 16x cards would work on pcie 4.0 8x 8x as if they were on pcie 3.0 16x 16x
Same applies to all devices.
-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
The end of THREADRIPPER?
- Thread starter Drumroll
- Start date
You missed that PCIe 4.0 provides double the bandwidth per lane over PCIe 3.0.
2 pcie 3.0 16x cards would work on pcie 4.0 8x 8x as if they were on pcie 3.0 16x 16x
Same applies to all devices.
I don't miss much.
Kinda irrelevant to my post video as this is about PCI-e 3.0 16x and 8x while PCIe 4.0 8x is equal to PCIe 3.0 16x bandwidth?
FYI Ryzen 3000 on X570 has 40 PCIe 4.0 lanes.
Last edited:
How is it irrelevant? If you have a card a year ago ALREADY saturating PCIe 3 16x then what do you think is going to happen going forward, or do you change platforms every year? The point is very simple: some people need to be able to run more than just 8x, especially because they also want pcie for nvme. Right now, Ryzen can't provide that, only Threadripper (for AMD). Without updates that would mean AMD shot themselves in the foot if they would kill TR.
If X570 has 40 PCIe 4 lanes, that's great news.
Though Nvidia has no PCIe 4.0 cards so again irrelevant.
You should've just told me you can't think past the present, would've saved me the trouble.
Yeah, that's not gonna happen.
Associate
- Joined
- 27 Apr 2007
- Posts
- 967
But will consumer tech actually utilise the extra bandwidth or is that there initially purely for data centre where they might utilise it?
I remember when SATA 3 had about roughly 4 times the data throughput of consumer HDDs.
But, there may be extra features alongside the extra bandwidth that can be useful for consumers? Thoughts!
Indeed it will be 2021 before PCI-E 5.0 hits HEDT, or Q4 ' 20 if they shift the TR releases.
I'd be very surprised if it was even that soon. PCIe 3 was delayed as implementers hit snags, it took 7 years between preliminary announcement of PCIe 4 and official ratification and then another 2 years to even show up on a motherboard. As of now, the 2-year old PCIe 5 spec hasn't even reached version 1.
5.0 should be hitting the final production revision/specification by Q4 '19, and hardware would be available inside of 2 years.
4.0 was finalised on 8th June 2017, and it's now being released into the first commercially available CPU in less than two years, there have been PCI-e 4.0 compatible devices on the market since Q3 last year, but no CPU's and boards to use them with commercially.
The delay in getting 4.0 out is precisely the reason why 5.0 will be implemented so quickly, it's a bit like Intel's 7nm coming at almost the same time at 10nm, separate development with one not hitting targets as soon as expected/predicted.
No. PCIe 4.0 is 2 years old. It arrived in 2017. PCIe 5.0 hit 0.9 in January this year, and 1.0 is expected to be finalised in the next few months.
The cycle between 4.0 and 5.0 has been hugely accelerated because there's the need for it in HPC, Server and various industrial applications (i.e. automotive). Plus, stakeholders are generally pulling in the same direction.
It's highly improbable that AMD don't do PCIe 5.0 for Zen 3 EPYC. Unless they change what they're doing now, then PCIe 5.0 will necessarily come to TR4, and there's no reason why it would be disabled for AM4 - indeed they need to keep up with Intel who will probably offer it.
PCIe 5.0 delivers significant latency improvements as well as further bandwidth improvements. Hence why 4.0 is pretty much a stopgap. Bandwidth won't do much for consumers over 4.0, but latency certainly will.
Edit: One final thing. I strongly suspect the reason why a lot of the X570 boards have fans on the heatsinks around the PCIe traces is that the intention is that they will support PCIe 5.0 next year. Just as some of the X470 boards are being given a BIOS upgrade to PCIe 4.0 compatibility now. PCIe 5.0 will take the bus speed to 32GHz.
Last edited:
And PCIe4 wasn't implemented on a non-enterprise motherboard until Q3 2019. So your point is what, exactly? Intel will jump straight to PCIe 5, but it won't happen whilst they're bogged down with 14nm++++++++++++++++++++++++++. And however cool it is for AMD to be pushing the boundaries, mainstream and HEDT consumers barely need PCIe4, let alone 5, and Zen 3 is unlikely to be a significant departure from Zen 2 to warrant jumping into PCIe 5 within the space of a year.
PCIe 5 will come in 2021 when AMD move to Zen 4 and their new GPU architecture "goes big" for compute cards, Intel get their 10nm/7nm **** together and Arctic Sound reaches maturity, and Nvidia drop some new monster Tesla cards to bury AMD's new compute.
That's why I said unless they shift the release dates of the TR4 platform, since it looks like Zen3 is going to be a Q3 or Q4 '20 release date for EPYC/Ryzen and if the TR4 platform is lagging behind as indicated then it maybe Q1 '21 or later for TR4. I guess we'll need see how delayed the Zen2 iteration of Threadripper is, hopefully not too long.
I think you are missing the target market, DC, Compute and such are a big deal, they are bandwidth starved per core of CPU power presently and when you have 64/96/128 cores on a single socket server you are wasting a huge amount of potential being stuck with 3.0/4.0 if 5.0 is available and the fact being first to market with it would be a great selling point into these scenarios/markets.
Let's not forget you can still push 5.0 down to 3.0, meaning a 2P EPYC Zen3 could potentially host the equivalent of 640 lanes of PCI-E 3.0 (160x2x2) bandwidth devices, and a 1P around 480 lanes equivalent. That is a massive benefit in power reduction and TCO, as well as space saved.
Unless the v1.0 of 5.0 is delayed very significantly, I can't see AMD not implementing it for Zen 3 EPYC ... and then there's every reason for them to do it for the rest of the related product stack.
Indeed, they might well push it wider than PCIe 4.0 stuff, and more quickly. Some potential embedded customers will want it ASAP.
Also, we're assuming the Zen 2 architecture isn't or won't be 5.0 compatible. I wouldn't be 100% sure about that. Since embedded usually lags in terms of architecture, and customers will go elsewhere if they can't get 5.0.
Indeed, they might well push it wider than PCIe 4.0 stuff, and more quickly. Some potential embedded customers will want it ASAP.
Also, we're assuming the Zen 2 architecture isn't or won't be 5.0 compatible. I wouldn't be 100% sure about that. Since embedded usually lags in terms of architecture, and customers will go elsewhere if they can't get 5.0.
Will they? Will consumers actually go elswhere if they cant get pci-e 5? I don't think they will, I don't think that the large majority of businesses care either (I say majority as there will be some out there running serious amounts of GPU's) as most probably wont be using much in the way of pci-e for storage and will be running a proper san, but your standard consumer... not a clue. We are not even into the lifecycle of 4 - You're mad bro!
Actually AMD supports PCIe 4.0 on 300 and 400 Series motherboards, and is up to the motherboard manufacturers to publish the bios update to unlock it for Ryzen 3000!!!!
https://www.guru3d.com/news-story/pcie-4-feature-pops-up-in-x470-motherboard-bios.html
And original quote from here also
https://www.tomshardware.com/news/amd-ryzen-pcie-4.0-motherboard,38401.html
Yes I know, perhaps I should've said "PCIe 4 wasn't implemented on a non-enterprise system until Q3 2019". But the point I was making is that the spec existed in a ratified form for 2 years before it came to the consumer market, so any notion of PCIe 5 hitting consumer-level next year is slim to none.
This news about the X570 basically confirms Threadripper lives on. Thank fwck.
Tbh think I'll sit this one out anyway. No point in a switch until we get DDR5.
Tbh think I'll sit this one out anyway. No point in a switch until we get DDR5.