-
Ever wanted an RSS feed of all your favorite gaming news sites? Go check out our new Gaming Headlines feed! Read more about it here.
VentureBeat: Xbox Series X Seagate Expansion Card Teardown
- Thread starter KodiakGTS
- Start date
Yeah because even looking at their controller its looks like its something totally custom, and the board also shows something from samsung that looks like some kind of Ram.
????????????????????
Microsoft has stated 2.4GB/s
Someone did a teardown of the SSD.
They found a controller capable of faster speeds than what Microsoft has advertised.
Why would anyone think "OH, THE SPEED OF THE SSD IS ACTUALLY FASTER, BUT MICROSOFT ARE KEEPING IT A SECRET!"
I just looked but this is still not saying the max speed is 3.7gb/s This still does not make sense.
To answer your first question
It's in the link that this thread is about, the teardown that this thread is about, shows that part.
And if that still doesn't "make sense"
In the link it also states that the max read speed of this part.
They shouldn't have bothered with CFExpress, it drives the price up for nothing. Who needs to hotswap them?
You just destroyed the entire USB and portable HDD marketing with that comment. Also all the fairies died.
When Ms announced the console, they said that the speeds given for the console and in this case, the SSD were minimum/guaranteed speeds for developers and not the maximum.
This was a very clear message they gave about them not provideding boosted or overlooked specs for the machine - alluding that is what is happening elsewhere or what is commonly done.
This is going to be easy enough to test as I'm sure within about a week of these units becoming available, someone will benchmark one.
Last edited:
I'm not sure if this is sarcastic or anything I'm just saying I would have prefered a cheaper SSD that you would need an internal cradle or something since it's not really a memory card, you plug it up and leave it there for years.
You basically said you couldn't understand why anybody would portable data.
I think we can all relate to taking discs or cartridges to friends houses as kids, this is something that isn't facilitated by a next gen machine where a 200gb game resides only internally within a console.
I mean if you prefer to pay premium for that more power to you I guess, I don't.
just because a part is rated as such does not mean it his that even with the controller. we will see soon enough in tests but they cant just leap to that conclusion based on the theoretical max of the controller, vs the stated maxes at raw.
I havea feeling this nails it. THe SSD can max out at 3.75 but Microsoft has all been about hitting sustained speeds. THey will need headroom for this and cooling. So the solution they have gone with allows them to hit the desired sustained speed of 2.5GBs
But I would hope you could understand why someone else would.
Yea, you can't have that console experience where that theoretical situation of streaming in assets as the camera rotates - if it will only get there some of the time.
Likewise there is no point planning a journey to work based upon the top speed of your car, if you have various other external factors that made mean you can travel at that speed the entire time.
It's a shame, otherwise my commute would take 20minutes instead of an hour
A single number for reads and writes of SSDs is also always going to be problematic, given that sequential and random reads have drastically different behavior.
Since nobody wants to actually explain how this works.
Controllers come in different tiers. Some support 4, 8, 16 channels. Each channel can be divided into 2 and can carry a certain amount of data per channel.
Say you want an SSD that can reach 4GB/s. You can buy a controller that has 4 channels. Each channel provides 1GB and the combination of all channels with Flash chips capable of 1GB/s each give you a total of 4GB/s. If you put 4 flash chips only capable of 500MB/s, you only get 2GB/s regardless of how fast the controller is. But you can put 8 flash chips capable of 500MB/s to those 4 channels and get 4GB/s.
The combination of all flash chips and controller determines the max capability of the SSD.
Take PS5 for example
12 chips, hooked to 12 channels. To attain 5.5GB/s, each chip only has to contribute approximately 460MB/s.
Durability is practically the least concern for modern SSDs. Especially for these which are mostly used for storage so they don't see very frequent writes. To put into perspective my Intel 600p NVME which is 4 years old and serves as my OS drive thus very frequent writes, only have ~50TB writes, the SSD itself is rated for 144TBW, I'm not even reaching half of its lifespan.
I doubt that using CFExpress drives the price up by that much. It's largely just the same interface as PCIe. It may need some additional ESD protection for the pins. And the CFExpress connector is designed to provide power to the card prior to data wires being connected. But really, that seems to be the biggest difference. The enclosure can't cost any more than the heat sinks that come on NVMe drives.
In the past, CFExpress cards have been expensive simply because they are niche, professional items and are rarely purchased/used.
Thank you that makes sense Here. So its determined not by the controller but by what the flash chips+the controller.
it's Gen 1, Microsoft already confirmed that so the same as One X IIRC
Thats what it is packaged as but each nand die has 1 dedicated channel. 2 64GiB die packaged as 1 128GiB module × 6.
Whatever the maths is, I find it easier to believe that the MAX is higher so they can hit a sustained speed of 2.5GHz over the max is absolutely 2.5GHz and they just push for that all the time. From working with PC tech it doesnt seem to make the best kind of sense to me.
Like we know that theres a high chance that the Xbox GPU could hit above 1850MHz but they would need to add more power and spin the fans up to keep the cooling down, just likea PC GPU, but for the profile that MS wants to target it doesnt make sense to do that, becuase you would increase power draw, heat and noise.
yeah there would be more power drawn etc, the controller is pretty much a processor so yeah all power concerns would be there.
The max pretty much is the controller plus the speed/ amount of chips used. Like in the 12 channel case it all adds up to the 5.5 speed vs the 2.4 speed but the controller can help to sustain that If im understanding it correctly.
Ah, interesting. I have read the data sheets for a couple flash chips, but hadn't caught on that they were multi chip packages.
So with this xbox card we need to wait until someone speed tests it to find out or can you take a guess at the max Read with the data shared?
PHISON Electronics Corp. - PS5019-E19T
www.phison.com
No DRAM and Dynamic SLC support as a caching solution.
SLC Cache vs. Direct TLC Write
To resolve the slow read/write performance of TLC SSDs, SLC cache is utilized “SLC cache” is not the same as “SLC NAND” but behaves like it by storing only one bit of data per cell
That is the max speed of the controller. That is one part in the chain. One element of the SSD. That controller can be paired with many different types of flash chips of different speeds etc.
It's like looking a the PSU of a computer, seeing it can do up to 1000W and therefore deriving that the CPU, GPU all plugged into it must pull 1000W under load.
It's not how it works.
2.4GB/s sequential read is one thing, but a controller without a DRAM cache for its LBA->PHY address lookup table will perform significantly slower in random read latency and throughput tests.
The Sony storage patent (that was broken down and analysed on this forum previously) that was consistent with the Road to PS5 talk by comparison starts off in its premise that even an external DRAM cache wasn't quick enough for what they wanted to do, and the patent deals with a way of shrinking the address look up table to be able to fit within an in-controller SRAM pool (not the one on the IO Complex).
In sequential read the difference will be something like 2.4GB/s vs 5.5GB/s. In random read performance the difference will be much, much, much greater.
It would be much greater if PS5 merely opted for an external DRAM cache for their controller. But if the patent is true and they have implemented on-chip SRAM to store the table, it will be significantly quicker in random read.
The second half of the patent deals with Sony's equivalent of DirectStorage API, which they've named File Archive API. A hash table look up of a file name immediately gets the LBA address of that file which can be sent straight to the controller, with no metadata layers to traverse.
It's surprisingly readable once you understand the basics:
INFORMATION PROCESSING DEVICE, ACCESS CONTROLLER, INFORMATION PROCESSING METHOD, AND COMPUTER PROGRAM - Sony Interactive Entertainment Inc.
<div p-id="p-0001">Disclosed herein is an information processing device including a host unit adapted to request data access by specifying a logical address of a secondary storage device, and a contro
www.freepatentsonline.com
The story is far more than 2.4 vs 5.5 numbers that are thrown around suggest.
I think Matt was right when he said it was on another level from what he saw.
I'll go with the figure Microsoft gave which is 2.4GB/s sustained read. There's been some benchmarks but its a very small sample size so we can't really draw much conclusion from that.
Likewise it didn't occur to me at first looking at the 6 flash packages on the PCB until someone found the particular package they were using. TH58LJT0T24BA4C
I don't even need to leave the house for it to be useful. We have three xboxes here at home each with a different user, and we'll have two of the new ones come Christmas. Being able to hotswap the card and transfer games for a user instead of having to repeatedly download them, as well as easily move room to room and play them myself is super useful. The price on these cards is a lot but being able to move the drive between consoles without having to reformat isn't a mistake.
Also you can have more than 1 with different games, on PS5 is just an expansion of the internal drive, only 1 at a time and non portable.
Is it safe to break the thermal paste connection and then use a random thermal paste to glue it all together?
That is exactly what I am doing. We will have a Series X in my Game Room and a Series S in the living room when my wife wants to play. I will play from time to time in the living room and being able to pull out the expandable and plug it into my Series S is going to be so much easier than having to download the game.
This is good info (although probably in the wrong thread), thanks for sharing.PHISON Electronics Corp. - PS5019-E19T
No DRAM and Dynamic SLC support as a caching solution.
SLC Cache vs. Direct TLC Write
To resolve the slow read/write performance of TLC SSDs, SLC cache is utilized “SLC cache” is not the same as “SLC NAND” but behaves like it by storing only one bit of data per cellwww.atpinc.com
That is the max speed of the controller. That is one part in the chain. One element of the SSD. That controller can be paired with many different types of flash chips of different speeds etc.
It's like looking a the PSU of a computer, seeing it can do up to 1000W and therefore deriving that the CPU, GPU all plugged into it must pull 1000W under load.
It's not how it works.
2.4GB/s sequential read is one thing, but a controller without a DRAM cache for its LBA->PHY address lookup table will perform significantly slower in random read latency and throughput tests.
The Sony storage patent (that was broken down and analysed on this forum previously) that was consistent with the Road to PS5 talk by comparison starts off in its premise that even an external DRAM cache wasn't quick enough for what they wanted to do, and the patent deals with a way of shrinking the address look up table to be able to fit within an in-controller SRAM pool (not the one on the IO Complex).
In sequential read the difference will be something like 2.4GB/s vs 5.5GB/s. In random read performance the difference will be much, much, much greater.
It would be much greater if PS5 merely opted for an external DRAM cache for their controller. But if the patent is true and they have implemented on-chip SRAM to store the table, it will be significantly quicker in random read.
The second half of the patent deals with Sony's equivalent of DirectStorage API, which they've named File Archive API. A hash table look up of a file name immediately gets the LBA address of that file which can be sent straight to the controller, with no metadata layers to traverse.
It's surprisingly readable once you understand the basics:
INFORMATION PROCESSING DEVICE, ACCESS CONTROLLER, INFORMATION PROCESSING METHOD, AND COMPUTER PROGRAM - Sony Interactive Entertainment Inc.
<div p-id="p-0001">Disclosed herein is an information processing device including a host unit adapted to request data access by specifying a logical address of a secondary storage device, and a contro
The story is far more than 2.4 vs 5.5 numbers that are thrown around suggest.
I think Matt was right when he said it was on another level from what he saw.
Yeah I can definitely see the advantages for having removable storage. It also means instead of buying a larger SSD you can just buy a second smaller one if you run out of space and save some money.
This is me right here. Much prefer the solution MS provided here tbh. Being able to easily move these around the house and take it to a friends house when needed is huge for my case use.
Kinda off topic, but what are the best external SSDs that I can buy for the Series X/S in terms of speed?
Yeah I love it, as games grow in size I can see myself have multiple cards to organize things easier - kids card, racing card, RPG card.
But they should've had the slot at the front, at that point we could've seen some crazy publisher launch their game on an expansion card, essentially as if it was a cartridge game. Missed opportunity.
As this only applies to bc games and loading is also still dependant on the CPU I would go for any SSD that you can get. They will all be faster than a standard external disk. Whether an external nvme might be worth it for you is a trade-off between performance gains and cost, especially if you want to have more than 1tb of storage.
DF had a great video about storage solutions on XSX.
Just out of curiosity, why are you worried about the speed? I'm fairly certain the USB speed is going to cap you at close to the speed of a normal SATA SSD.
SATA SSD's run at around 550MByte/s, not Mbit (which is right around 5Gb/s).
Argh, sorry, you are right. Brain didn't boot up properly yet :-\
The people who worked on it said the speeds they gave were conservative and could be much higher. Not a secret - anything above 2.4 probably just isn't regular.
Hmm... The slot being in the back doesn't really take away from what you suggested. That's going to be one hell of an expensive cartridge though. But the possibility is definitely there and intriguing.
And as time passes and prices drop having multiple cards for different genres or whatever and being able to easily swap them is a great idea and another reason why this solution is much more appealing to me.