1TB Team Group MP33 M.2 NVMe PCIe 3D SSD

opps removing due to error in my post

Quote from arbitrager : FYI: has DRAM TEAMGROUP MP34Q M.2 PCIe SSD uses QLC Flash and PCIe Gen3x4 interface and supports SLC Caching technology with DRAM Cache Buffer. New era QLC Flash offers SSD a capacity of up to 8TB(8000GB).

This is the mp33 not the mp34... The mp33 is tlc not qlc. Also it doesn't have a dram cache, but it does use slc cache which means it uses a small portion of the tlc nand flash in slc mode and functions as the cache for acceleration.

Quote from Numus19 : This is the mp33 not the mp34... The mp33 is tlc not qlc. Also it doesn't have a dram cache, but it does use slc cache which means it uses a small portion of the tlc nand flash in slc mode and functions as the cache for acceleration.

So, uhhhh, this any good or what…? 😂

Quote from Dr.Polanski : So, uhhhh, this any good or what…? 😂

1tb, pretty decently high read/write speeds, 600tbw, for 65? yeah, for the price this is pretty fantastic for a budget drive.

Quote from Numus19 : This is the mp33 not the mp34... The mp33 is tlc not qlc. Also it doesn't have a dram cache, but it does use slc cache which means it uses a small portion of the tlc nand flash in slc mode and functions as the cache for acceleration.

I wish there was a note that could be automatically added to all SSD deals to prevent this confusion between SLC cache and DRAM.

For the 1500th time
1. DRAM is used to keep SSD translation tables in memory that is used to map read requests into locations where that data is kept in the SSD. The mapping is also required if you want to do internal wear leveling, etc. NVMe drives can use the memory in the host computer for this mapping called Host Memory Buffer. You don't want to keep this in the NAND itself because it requires constant updating which will make it slow and wear out the SSD faster. HMB is part of nvme protocol used by nvme SSDs.

2. SLC cache is a write buffer that almost every SSD has because writing to TLC/QLC directly is slow. This has nothing to do with DRAM usage and is not a substitute for it. Team Group puts it in the header to hide no DRAM and confuse people. Just saying has SLC cache isn't sufficient. How much cache there is also makes a difference because once the cache is filled up, write performance has a tendency to get near HDD speeds or worse. Budget SSDs skip on DRAM and have a small SLC cache to cut costs.

Quote from WingsOfF : I wish there was a note that could be automatically added to all SSD deals to prevent this confusion between SLC cache and DRAM. For the 1500th time 1. DRAM is used to keep SSD translation tables in memory that is used to map read requests into locations where that data is kept in the SSD. The mapping is also required if you want to do internal wear leveling, etc. NVMe drives can use the memory in the host computer for this mapping called Host Memory Buffer. You don't want to keep this in the NAND itself because it requires constant updating which will make it slow and wear out the SSD faster. HMB is part of nvme protocol used by nvme SSDs. 2. SLC cache is a write buffer that almost every SSD has because writing to TLC/QLC directly is slow. This has nothing to do with DRAM usage and is not a substitute for it. Team Group puts it in the header to hide no DRAM and confuse people. https://static.slickdealscdn.com/ima...lies/smile.gif Just saying has SLC cache isn't sufficient. How much cache there is also makes a difference because once the cache is filled up, write performance has a tendency to get near HDD speeds or worse. Budget SSDs skip on DRAM and have a small SLC cache to cut costs.

That is a lot of wrong information and misleading information.

There really is no difference between HMB and DRAM for the data map. All drives require this. Why are you talking about HMB with DRAM? They are 2 separate things, with DRAM-less cache using HMB to function as the DRAM cache (using a small amount of the CPU's dram). It is a little slower than DRAM directly on the chip, but it isn't significantly slower. DRAM-less NVMes now can do wear leveling just like drives with DRAM chips. The DRAM is also used as a write buffer, which is why the speeds are that much faster. Literally the reason they use an SLC cache is because in SLC the NAND can last upwards of 100,000 write cycles.

Quote from Numus19 : That is a lot of wrong information and misleading information. There really is no difference between HMB and DRAM for the data map. All drives require this. Why are you talking about HMB with DRAM? They are 2 separate things, with DRAM-less cache using HMB to function as the DRAM cache (using a small amount of the CPU's dram). It is a little slower than DRAM directly on the chip, but it isn't significantly slower. DRAM-less NVMes now can do wear leveling just like drives with DRAM chips. The DRAM is also used as a write buffer, which is why the speeds are that much faster. Literally the reason they use an SLC cache is because in SLC the NAND can last upwards of 100,000 write cycles.

You are very confused.

HMB is a substitute when you don't have DRAM for the translation tables. DRAM is on the SSD. If budget SSDs skip on-board DRAM for cost reasons, they can use memory on the host computer instead using the HMB feature of the nvme protocol (unlike SATA SSDs which can't do HMB). So nvme SSDs either have DRAM or they use HMB. Typically this is of the order of 64MB or so in size. Yes, the translation tables are necessary which is why nvme SSDs have dram for it or use HMB. They may or may not do wear leveling if they don't have DRAM. Depends on the controller and how much of a warranty they want to provide.

Don't confuse this with SLC cache which is used to buffer writes. These are typically several GB in size and have nothing to do with the DRAM. You can either have a special SLC chip or more likely you use the TLC in SLC mode (called Pseudo SLC) so you can theoretically have up to a third of the available space as SLC cache. But budget SSDs which use simple controllers may fix the size of the cache and suffer from write slow downs when that small SLC cache is used up.

Those two are completely independent concepts.

So, it is incorrect to make statements like it doesn't have DRAM but does SLC cache instead.

Quote from WingsOfF : You are very confused. HMB is a substitute when you don't have DRAM for the translation tables. DRAM is on the SSD. If budget SSDs skip on-board DRAM for cost reasons, they can use memory on the host computer instead using the HMB feature of the nvme protocol (unlike SATA SSDs which can't do HMB). So nvme SSDs either have DRAM or they use HMB. Typically this is of the order of 64MB or so in size. Yes, the translation tables are necessary which is why nvme SSDs have dram for it or use HMB. They may or may not do wear leveling if they don't have DRAM. Depends on the controller and how much of a warranty they want to provide. Don't confuse this with SLC cache which is used to buffer writes. These are typically several GB in size and have nothing to do with the DRAM. You can either have a special SLC chip or more likely you use the TLC in SLC mode (called Pseudo SLC) so you can theoretically have up to a third of the available space as SLC cache. But budget SSDs which use simple controllers may fix the size of the cache and suffer from write slow downs when that small SLC cache is used up. Those two are completely independent concepts. So, it is incorrect to make statements like it doesn't have DRAM but does SLC cache instead.

https://www.google.com/url?sa=t&s...YZI_IVQd8s

Quote : The DRAM served as cache for writing data to the drive

Sorry but I am going to take western digitals word over yours

they have done extensive testing to show HMB + SLC Cache doesn't have a major disadvantage over dram caches on the nvme. you can direct your argument to them if you want

Quote from Numus19 : https://www.google.com/url?sa=t&s...YZI_IVQd8s Sorry but I am going to take western digitals word over yours they have done extensive testing to show HMB + SLC Cache doesn't have a major disadvantage over dram caches on the nvme. you can direct your argument to them if you want

You are still confused about the concepts and arguing something else. Do you understand that keeping translation tables and the write cache are two different concepts? If you aren't clear about this, then you should do some reading on how SSDs work. Nothing I have said is inconsistent with WD document. You aren't understanding it.

All recent generation SSDs SATA or NVMe with TLC or QLC NAND use a SLC or Pseudo SLC cache, otherwise their writes would be worse than a HDD. Nobody uses a DRAM as a write cache any more because the increased capacities of the SSDs require a lot more cache and it is not only expensive to put that much DRAM but it wouldn't even fit in a m.2 card.

This is purely about writes. So saying something has SLC cache isn't saying much. Everybody does. But how much SLC cache is available makes a difference. Budget SSDs may skimp on these. The smaller the SLC cache, earlier the write slowdown in sustained writes. This performance degradation is independent of DRAM in the SSD. Even SSDs with DRAM (used for translation tables) and high end ones have this problem for sustained writes beyond their cache size. Budget ones show this problem earlier.

The early generation Sata SSDs didn't have the HMB capability because it is not part of the protocol. So, the translation tables had to be on the device. DRAM was used for this and also as a small write cache because the small capacity of the SSD didn't require a large cache. But as the SSD capacities increased and large media files or game files were written, the write cache was moved to using part of the NAND chip itself rather than increase the DRAM chips. DRAMs are now used primarily for translation tables.

NVMe allows for use of HMB in the protocol so those SSDs can completely get rid of DRAM used for translation tables and use HMB instead. Not as fast as on board DRAM and the SSD cannot do offline processing for wear leveling etc without involving the host computer.

Vendors have tried to make the case that the lower cost dramless ones are "just as good" as without it but that is a self-interested marketing just like they do for CMR vs SMR or TLC vs QLC. We do know that there is a difference between those and not as peachy as the vendors would like to portray. So, take it with a grain of salt.

Here is an example of how the sustained writes slow down even in high end DRAM SSDs after their SLC cache gets filled up. How much SLC cache they have and how they refresh the cache in their controllers make some better than others. Budget SSDs typically have controllers that don't do any such advanced tricks and not likely to have big enough caches. Why saying something has an SLC cache doesn't make all of them the same in this write slow down issue.

And this is what happens with this budget SSD for the write slow down. Note the dips down to worse than HDD write speeds even though this one has a SLC cache.

Cheapest 1tb NVMe so far, so i guess as long as it doesn't suck, it'll go front page.
If they don't run out of stock in the next half hour.

Toshiba NAND + Phison E13 + HMB + 5 years warranty = good for daily use. This is much better option than a 1TB SSD.