News Adata Teases First PCIe 5.0 SSDs, up to 14 GBps of Throughput

[Admin]

Adata's XPG PCIe 5.0 SSDs appear to outperform previously spotted Kioxia prototype drives.

Adata Teases First PCIe 5.0 SSDs, up to 14 GBps of Throughput : Read more

 

[wifiburger]

that's nice, there's no x4 pcie5 for any cpu

alder lake has x16 pcie5 on the GPU side, but not sure it's worth dropping to x8 just to run 1 nvme at pcie5 and that's mostly on crazy expensive boards

This device could be great for PCIE4 since you'll be at max read/write vs current PCIE4 which may not reach max bandwidth.

But in all honesty, pc games are starting to approach some crazy sizes, like 200gb+ call of duty . Some other games, 40gb,60gb,80gb... etc

My last storage purchase was not nvme or sata ssd.
I went back to hdd, 12TB and does 200Mb/s+ which for games I can't tell the different vs my pcie4 nvme and no clean-up required for managing game installs

 

[salgado18]

that's nice, there's no x4 pcie5 for any cpu

alder lake has x16 pcie5 on the GPU side, but not sure it's worth dropping to x8 just to run 1 nvme at pcie5 and that's mostly on crazy expensive boards

This device could be great for PCIE4 since you'll be at max read/write vs current PCIE4 which may not reach max bandwidth.

But in all honesty, pc games are starting to approach some crazy sizes, like 200gb+ call of duty . Some other games, 40gb,60gb,80gb... etc

My last storage purchase was not nvme or sata ssd.
I went back to hdd, 12TB and does 200Mb/s+ which for games I can't tell the different vs my pcie4 nvme and no clean-up required for managing game installs

Today, dropping to x8 pcie 5.0 wouldn't hamper any GPU. I don't know the next generation, they seem to be aiming at huge leaps of performance, but today that's not an issue.

I agree, SSDs should focus more on capacity than speed at this point. New games will increasingly need fast storage to work properly, but even still the benchmark are consoles, with ~5000 MB/s reads, so pcie 4.0 is sufficient. I bet in some time your HDDs won't be enough for some of those games, since today they rely on disk only at load time, not in real time.

 

[Im__A__Blimp]

This is so stupid. Anything past 3500MBPS on PCIe3.0. .... you won't notice it. Your games won't load any faster or run any faster. Even if they did, the difference would be splitting hairs. That's why XBSX didn't use pcie4. They knew this. Focus not on speed, but on consistency, quality, capacity, life longevity. Speed is way overkill and no applicis going to legitimately need anything past PCIe 3 3500

 

[Alvar "Miles" Udell]

Innovations such as DirectStorage will also take away some of the requirement of a massively high read speed once they achieve more widespread use.

But yes, some company, especially Samsung, needs to focus on high-capacity M.2 format drives with PCIe 3.0 speed but high capacity. You have the Team Group MP34 4TB for $430, and that's not exactly a slow drive that TomsHardware hasn't tested yet, especially for the price, but all the other 4TB drives are near twice the price.

One could argue 2TB is all you need, with an NAS being much better suited towards bulk storage, but think outside of computers. Phones and tablets, for example, don't need massively fast storage, but they really do need to have 256gb as standard if not 512gb as file sizes increase. Mobile workstations don't need PCIe 4 speed, but they do need large amounts of storage for professional applications.

 

[hotaru251]

pcie 5 is again pointless.

gen 4 was nowhere NEAR capped.

it just inflates cost for stuff that "offers" it.

 

[Alvar "Miles" Udell]

pcie 5 is again pointless.

gen 4 was nowhere NEAR capped.

it just inflates cost for stuff that "offers" it.

Disagree. Gen 5's lower power consumption and higher bandwidth allow for much more connectivity with fewer necessary connections. You only need an x4 Gen 5 to equate to x8 Gen 4, which mean the next generation video cards from AMD and nVidia won't be bottlenecked. Also in servers they can get much more dense with compute power and not be bandwidth limited.

As far as storage goes, outside of datacenters and high end professional workstations, it is pointless.

 

[hotaru.hino]

Disagree. Gen 5's lower power consumption and higher bandwidth allow for much more connectivity with fewer necessary connections. You only need an x4 Gen 5 to equate to x8 Gen 4, which mean the next generation video cards from AMD and nVidia won't be bottlenecked. Also in servers they can get much more dense with compute power and not be bandwidth limited.

Except there's the issue that not everyone is going to be on the latest and greatest. So you'll have something like the RX 5500 where if it reaches a condition that requires a significant amount of PCIe traffic (e.g., VRAM is full and needs to start swapping), having 8 lanes at a reduced speed is going to impact performance. This would make more sense if you wanted to multiplex the lanes, but that adds complexity.

While this doesn't really matter for storage since most consumer use-cases are hardly impacted between SATA 6Gbps and PCIe 3.0 x4 bandwidth, you can't really say the same for video cards.

EDIT: In case I got it backwards that the slot is of a slower version and the device a faster version, it'd still be a problem if the slot was a faster version and the device a slower version, since PCIe negotiates to the slower of the two. While higher end graphics cards haven't seen a significant performance impact at 8 lanes of their "native" PCIe speed, if you wanted to say cut it down to 4 lanes, that's when things may get noticeable.

 

[Co BIY]

Comments not showing under the article and showing a Zero comment count even with 7 comments . Comments are only visible by following the link to Comment in the Forum.

 

[jkhoward]

Expandable VRAM when? /s (I know it’s been done before but 14 GBps would make a killer buffer for video cards).

 

[hotaru.hino]

Expandable VRAM when? /s (I know it’s been done before but 14 GBps would make a killer buffer for video cards).

It's your system RAM. If the application starts needing more space in VRAM than there's VRAM available, stuff will start swapping from VRAM into system memory (which eats into the application's memory pool)

 

[escksu]

I have to say all it does is looks good on sequential benchmarks. Under normal usage, your SSD is nowhere near that speed. I/O is mostly reading/writing small files with very low queue depths and serial (1 file at a time). So you are mostly in the 20-40MB/s range. This is also why PCIE 4.0 SSDs don't reduce your windows/app loading times by 1/2 compared to PCIE 3.0. Even Nvme SSD is only slightly faster than SATA ones under normal usage (only sequential transfer are significantly faster).

However, thats not to say that PCIE 5.0 is useless. You can have 1x PCIE 5.0 and its as fast as 4x PCIE 3.0. 2x is fast as 4x PCIE 4.0. This reduces trace counts and cost. It also reduce the number of PCIE lanes required.

So, next time, we could have just 4-8 lanes of PCIE for GPU instead of the usual 16x.

 

[escksu]

Expandable VRAM when? /s (I know it’s been done before but 14 GBps would make a killer buffer for video cards).

VRAM is expandable but quite a fair bit of soldering and electronic skills required. Manufacturers can easily double the amount of RAM for GPUs by using 2GB modules instead.

 

[Alvar "Miles" Udell]

Expandable VRAM when? /s (I know it’s been done before but 14 GBps would make a killer buffer for video cards).

Radeon™ Pro SSG GPU - Solid State Graphics | AMD

 

[watzupken]

SSD speed on the surface is improving, i.e. sequential read/ write speed, but as a whole, it has stagnated. Not to say that sequential read/ write speed is useless, but most people won't feel the difference whether they are using a SATA3 SSD vs PCI-E 5.0, unless you are transferring huge files very frequently. I mean if you do a blind test between these 2 SSD types, you generally don't feel that the SATA3 SSD is that slow when it comes to loading games or applications. The difference is perhaps a couple of seconds.
The chase for sequential read/write speed is like those days where camera makers were chasing after high megapixel count. It is good for those who use it for big prints, but most people are printing it on 3R/ 4R size, which don't matter. Instead, the sensor remains tiny and image quality is lacking even with extremely high megapixel count then.

 

[scottsoapbox]

No consumer PC is going to have to the queue depth to actually hit those numbers.

 

[hotaru251]

You only need an x4 Gen 5 to equate to x8 Gen 4, which mean the next generation video cards from AMD and nVidia won't be bottlenecked

80% unlikely even next gen gpu will saturate the 3.0 (and if so it would ONLY be the flagship card)

 

[Alvar "Miles" Udell]

80% unlikely even next gen gpu will saturate the 3.0 (and if so it would ONLY be the flagship card)

It depends on what you are doing. Can't say about gaming because there hasn't been a 3090 gaming test done by a reputable source, just a 3080 by TechSpot, and in their test there was no bottleneck, but if next generation video cards are much faster, you can't draw any conclusions from that test, other than the fact that most all home use applications will not be bottlenecked.

Puget Systems did a PCIe scaling test with an RTX 3090 on the same Ryzen 5950X system testing from PCIe 1.0 to 4.0. As it stands now the 3090 is noticeably limited by PCIe 3.0 vs PCIe 4.0 in more professional tasks. If next generation flagships are to be upwards of twice as fast as the 3090 and 6900XT, which is possible if due to nothing else than larger dies and higher clocks due to the higher power limit (Expect To See Nvidia's 12-Pin Power Plug On Future PCIe 5.0 GPUs | Tom's Hardware (tomshardware.com) ), then PCIe 5.0 will be a requirement to prevent a bottleneck.

PCI-Express 4.0 vs 3.0 Video Card Performance (pugetsystems.com)