You guys should do what tech report does and do some kind of live stream whenever you record the podcast and then distribute a recorded stream a few hours after recording.
Then you can clean it up and release higher quality audio in the podcast feed a while later.
It strikes a good balance between prompt release and high quality release.
I am still like baffled how the successor of SATA3 is evolving. Seams like everybody is scratching their head. All these form factors don't scale on desktop motherboards or entertainment systems for multiple U.2 , M.2 ,... Which is fine so why dont they just make it simple namely 1/2 U.2 slots. If you want scaling up you can include adapters that slide in a sata 4-16x port that can hold multipe U.2 drives. You can even keep the sata ports for older SSD/HDD drives. And for 80% of the users those EMPTY/USELESS 16x/8x/4x PCIe slots just turns into a U.2 adapter that can connect 2-4 drives paralell next to their graphics cards.
This feels the most easy scalabe route for all form factors. Everything else so far trys to keep to old IDE/SATA form factors. Why just not use the empty space. Am I missing something technically because I have no technical background to judge this? Why are we even talking about cables ?
PS google images M.2 adapter , then you know what i mean by adapter.
Hmm, dunno, I do hope that doesn't end up being the only viable route tho... As a gamer I don't really have the luxury of a bunch of empty slots (between two dual slot GPU and a sound card for Dolby Headphone)... I could spare one, maybe, on a decently spaced board.
I see myself needing 2-3 drives for a while too, unless PCI-E/M2 prices drop like a rock, would rather see SATA survive in some form even if it's just for an array of slower SSD dedicated to data.
The issue here is that some users actually use those 'useless' slots for other things. If we take the enterprise tack, the best way to sort it would be through a backplane, although if you look at many rack systems they still rely on cabling to a certain extent, usually through breakout cables to the backplanes.
M.2 is a form factor we're seeing being used on motherboards to fill up empty PCB space, so as prices of M.2 match that of SATA (for the most part they should be already), there will be some shift. The downside is that SATA is more universal, so if you need to migrate drives it's easy to take out - removing an M.2 drive when you have other hardware on top is not straightforward. M.2 is also not really hot-swappable.
U.2 bridges some of those issues, by providing a cabling system and moving up in terms of performance. But as we mentioned in the podcast, the big barrier to that right now is the number of devices (one consumer model out right now) and cabling cost ($2 each, rather than 0.1 cents for SATA). If you're still relying on cables, either consumer or enterprise, it factors into the cost of the build. Especially in enterprise, if you want 8-16 drives in a single rack unit. If you don't use a cable (or a backplane) for a 2.5/3.5-inch drive, hot swappableness is lost pretty much.
I guess you could have M.2 on a PCIe riser card, and if the PCIe lanes run through a hot-swappable capable PCIe switch, then it's doable. But you have to then think about riser cards and ease of replacing. PCIe switches aren't cheap either.
A lot of what we discussed on the topic in the podcast comes from the manufacturers themselves, and how they perceive this part of the industry.
Wondering out loud with some questions : Does anybody have numbers stating how many people actually fully use their expansions slots. My personal experience is limited to gaming friends and building PC's for family. The usage scenario's are one major graphics card , rarely one USB adapter and in more and more case no dedicated graphics card.
If you go SLI/Crossfire + audio card + USB adapter , I see possibly room to squeeze in one extra expansion slot? SLI-Crossfire builds tend to extend all the way over the sata ports that the only way to go then is cables.
How much would a PCEIe switch cost if you mass produce them vs cables? I would see them taking maximum 4 U.2 drives , cheapest M.2 riser card i could find for one is around 8$ dollars, if that scales , then this wouldn't be cheap either indeed.
It's too bad Dustin isn't here to talk about this, because the beginning reminds me of the one podcast that Dustin was on with Anand, and how Haswell had low overclocking headroom compared to Ivy Bridge and SB, and how in the future were we going to see this trend of lower peak OC continue. And sure enough, it does here. Instead of 4.6-4.8Ghz on SB, 4.4-.4.6 on IB, 4.2-4.4 on Haswell, and now 4.1-4.2Ghz on Broadwell. IPC goes up but peak Frequency continues to decrease over time. I'll have to see how skylake does, to determine if it's worth upgrading from my 2500k, and I haven't come across any situation where my 2500k isn't fast enough yet.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
Facebook
Twitter
RSS
9 Comments
Back to Article
ImSpartacus - Wednesday, July 8, 2015 - link
You guys should do what tech report does and do some kind of live stream whenever you record the podcast and then distribute a recorded stream a few hours after recording.Then you can clean it up and release higher quality audio in the podcast feed a while later.
It strikes a good balance between prompt release and high quality release.
Drazick - Wednesday, July 8, 2015 - link
Hi,What about Part II of the Broadwell 5775C review?
Could you test its compute capabilities?
Could you add MATLAB performance into your CPU review suite?
Thank You.
tsk2k - Thursday, July 9, 2015 - link
I too am waiting for that.plopke - Wednesday, July 8, 2015 - link
I am still like baffled how the successor of SATA3 is evolving. Seams like everybody is scratching their head. All these form factors don't scale on desktop motherboards or entertainment systems for multiple U.2 , M.2 ,... Which is fine so why dont they just make it simple namely 1/2 U.2 slots. If you want scaling up you can include adapters that slide in a sata 4-16x port that can hold multipe U.2 drives. You can even keep the sata ports for older SSD/HDD drives. And for 80% of the users those EMPTY/USELESS 16x/8x/4x PCIe slots just turns into a U.2 adapter that can connect 2-4 drives paralell next to their graphics cards.This feels the most easy scalabe route for all form factors. Everything else so far trys to keep to old IDE/SATA form factors. Why just not use the empty space. Am I missing something technically because I have no technical background to judge this? Why are we even talking about cables ?
PS google images M.2 adapter , then you know what i mean by adapter.
Impulses - Wednesday, July 8, 2015 - link
Hmm, dunno, I do hope that doesn't end up being the only viable route tho... As a gamer I don't really have the luxury of a bunch of empty slots (between two dual slot GPU and a sound card for Dolby Headphone)... I could spare one, maybe, on a decently spaced board.I see myself needing 2-3 drives for a while too, unless PCI-E/M2 prices drop like a rock, would rather see SATA survive in some form even if it's just for an array of slower SSD dedicated to data.
Impulses - Wednesday, July 8, 2015 - link
As a single guy I also have zero need for a NAS or an external storage bay, Intel is at the center of the whole mess, give us more lanes already!Ian Cutress - Thursday, July 9, 2015 - link
The issue here is that some users actually use those 'useless' slots for other things. If we take the enterprise tack, the best way to sort it would be through a backplane, although if you look at many rack systems they still rely on cabling to a certain extent, usually through breakout cables to the backplanes.M.2 is a form factor we're seeing being used on motherboards to fill up empty PCB space, so as prices of M.2 match that of SATA (for the most part they should be already), there will be some shift. The downside is that SATA is more universal, so if you need to migrate drives it's easy to take out - removing an M.2 drive when you have other hardware on top is not straightforward. M.2 is also not really hot-swappable.
U.2 bridges some of those issues, by providing a cabling system and moving up in terms of performance. But as we mentioned in the podcast, the big barrier to that right now is the number of devices (one consumer model out right now) and cabling cost ($2 each, rather than 0.1 cents for SATA). If you're still relying on cables, either consumer or enterprise, it factors into the cost of the build. Especially in enterprise, if you want 8-16 drives in a single rack unit. If you don't use a cable (or a backplane) for a 2.5/3.5-inch drive, hot swappableness is lost pretty much.
I guess you could have M.2 on a PCIe riser card, and if the PCIe lanes run through a hot-swappable capable PCIe switch, then it's doable. But you have to then think about riser cards and ease of replacing. PCIe switches aren't cheap either.
A lot of what we discussed on the topic in the podcast comes from the manufacturers themselves, and how they perceive this part of the industry.
plopke - Thursday, July 9, 2015 - link
Wondering out loud with some questions :Does anybody have numbers stating how many people actually fully use their expansions slots. My personal experience is limited to gaming friends and building PC's for family. The usage scenario's are one major graphics card , rarely one USB adapter and in more and more case no dedicated graphics card.
If you go SLI/Crossfire + audio card + USB adapter , I see possibly room to squeeze in one extra expansion slot? SLI-Crossfire builds tend to extend all the way over the sata ports that the only way to go then is cables.
How much would a PCEIe switch cost if you mass produce them vs cables? I would see them taking maximum 4 U.2 drives , cheapest M.2 riser card i could find for one is around 8$ dollars, if that scales , then this wouldn't be cheap either indeed.
cruzinforit - Sunday, July 26, 2015 - link
It's too bad Dustin isn't here to talk about this, because the beginning reminds me of the one podcast that Dustin was on with Anand, and how Haswell had low overclocking headroom compared to Ivy Bridge and SB, and how in the future were we going to see this trend of lower peak OC continue. And sure enough, it does here. Instead of 4.6-4.8Ghz on SB, 4.4-.4.6 on IB, 4.2-4.4 on Haswell, and now 4.1-4.2Ghz on Broadwell. IPC goes up but peak Frequency continues to decrease over time. I'll have to see how skylake does, to determine if it's worth upgrading from my 2500k, and I haven't come across any situation where my 2500k isn't fast enough yet.