Why not MSI B450M Pro-VDH Max at this price?

Pro-VDH Max's core current are rated the same, with extra 1 phase.

This post has been edited by kingkingyyk: Jul 15 2020, 07:07 AM

Well, if you look at the spec sheet of the low side MOSFET, Pro-VDH Max is better.

B450 Pro 4 - http://www.sinopowersemi.com/temp/SM4336NSKP_datasheet.pdf (65A)
B450M Pro-VDH Max - http://www.sinopowersemi.com/temp/SM4503NHKP_datasheet.pdf (80A)

This post has been edited by kingkingyyk: Jul 15 2020, 11:43 AM

There are 2 revisions if you check the list.

P/S : I'm not disagreeing, but just to let TS to know more .

This post has been edited by kingkingyyk: Jul 15 2020, 02:09 PM

AMD shifted focus to AI accelerators, so there is no competition for NVIDIA. Super would probably be either slip in to fill in the price gap or replaces the existing product at the same price with slightly better performance.

Ryzen 5 5600G RM579
MSI A520M-A Pro RM299
Kingston Fury Beast 8GB DDR4-3200MHz RAM RM176
WD Blue SN570 500GB SSD RM189
Silverstone ST50F-ES PSU RM165
Cap ayam casing RM100

That price was from old stock. NAND flash industry predicted the demand wrongly earlier and had a lot of inventory. The clearance is done now.

Manufacturers want to recover back what were lost earlier this year?

Ryzen 5 5600 RM569
MSI B550M Pro-VDH WiFi RM559
Kingston Fury Beast DDR4-3200MHz 8GB RAM x2 RM176
Gigabyte RX6600 8GB GDDR6 Eagle RM999 / PNY RTX3050 8GB GDDR6 Verto RM988
WD Blue SN580 1TB RM308
FSP HV Pro 550W PSU RM219
Casing RM170

Ryzen 7 5800X3D RM1449
Thermalright PA120 Cooler RM159
MSI B550 Gaming Plus RM709
Kingston Fury Beast DDR4-3200MHz 16GB RAM x2 RM318
Sapphire RX7900XT Pulse 20GB GDDR6 RM4299
WD SN580 2TB SSD RM539
LianLi 205 Mesh Casing RM329
Silverstone DA 750W PSU RM399
Viewsonic VX2882-4KP (28"+4K+150Hz) RM1799

Total : RM10000

This post has been edited by kingkingyyk: Dec 29 2023, 11:00 AM

DRAM-less SSD isn't a monster when it comes with HMB.

That happens when you have insufficient amount of RAM and with 32GB RAM, what use case you think will cause the issue?

HMB simply lets the SSD to borrow the system memory as it's DRAM. It does have a memory portion to work with during the operation, except with some data round-trips. The general issue with DRAM-less SSDs are those usually paired with cheaper NANDs / less-tuned write/load-balancing algorithms / weaker controller. Not the case for SN570/SN580.

https://documents.westerndigital.com/conten...70-nvme-ssd.pdf
https://documents.westerndigital.com/conten...50-nvme-ssd.pdf

Both are rated same. WD Green SN350? That is indeed something to avoid.

If you think otherwise, please share your findings.

This post has been edited by kingkingyyk: Dec 29 2023, 12:26 PM

Just share something to support your statement, don't run away

AM4 is reaching end of life now, it is already much better than the initial launch.

Having built quite an amount of AM4 machines, as long as you don't push the out-of-spec stuffs to the limit (i.e. running XMP, undervolting), you don't really run into much troubles.

Comparing the Blue with those QLC or part-swapping models, it is surprisingly acceptable to pay a little extra for transparency. The performance is top among DRAM-less segment as well.

You missed the point.

You articles only talk about scenario where the SSD is indeed without DRAM to work with. HMB allows the SSD to have DRAM to work with, except they are borrowed from your CPU.

HMB already allows the controller to access preserved part of the DRAM attached to your CPU, to act as it's DRAM.
Some DRAM-less design, like WD Blue or Phison PS3111-S11 do come with a small amount SRAM to store the metadata table instead of thrashing the data. That is fixed since the algorithm ran in the controller has deterministic in terms of space complexity, i.e. 128GB of space has 16MB of metadata.

To add things into complexity, HMB only lets SSD controller to get an amount of RAM attached to your CPU, and it is up to the controller to store data it wants into that portion. The data structure & algorithm comes into place. With this layer, it can request size like 1GB and use it like a SSD with DRAM, but usually it is overkill. You only see they request like 32MB-64MB in real life, reason being :
- Windows does have write caching, which still writes the data into RAM first and flush it to the drive asynchronously in background
- SSD controller receives write request, it can first buffer the data in the HMB portion but this is useless since the data is already in RAM anyway (refer to the earlier step, but this can work if the OS doesn't work that way)

Next...
- SSD controller writes the data to SLC cache if it is not full yet, then only back off to TLC mode (where the speed starts dipping)
- Throughout this process, the SRAM or HMB portion of cached metadata can be used in searching the empty page / updating the metadata
- Periodically / every X round of writes / by strategy (i.e. LRU, FIFO, weighted access or some kind of smart AI you name it), the metadata will be flushed back to the NAND flash for persistent storage.

Here is an IEEE paper that shows it is entirely possible for using HMB as both metadata table and data cache, and what's stopping controller firmware engineer from implementing this? One of the author was from Samsung memory. Your "drive swap" problem is now solved.
https://ieeexplore.ieee.org/stamp/stamp.jsp...rnumber=8868151

It is too early (late, for now) to conclude that every single DRAM-less design is evil when you have so many things that come into play in the Write amplication ratio, which is the term that you didn't even mention in your post.

To add salt to the injury, WD offers both SN570/SN580 and SN850X 5 years warranty and rated them at the same TBW, so this clearly goes against what you mentioned about the shorter warranty thingy.

This post has been edited by kingkingyyk: Dec 30 2023, 08:25 PM

The 5600G comes with stock cooler so PA120 is not a must. Why not squeezing a small amount from other parts and go 5700G?

Try FPGA?

https://videocardz.com/newz/nvidia-recommen...tability-issues

https://fudzilla.com/news/58781-chipzilla-s...urned-in-droves

Looks like Intel got too desperate in pushing the chip to limit and eventually the chip becomes unstable even with the stock voltage curve?

Different context. This is the case of pushing thing out of spec, which is obviously normal as not guaranteed.

------------------------------

A thing to note now is both camp are trying to push the chip as high clock as possible by providing high voltage when out of factory. It is just the matter of time when the stock curve no longer can fulfill after using for some time. The argument is weak on saying Intel is more stable, and I showed the contradiction.

So now I have AMD server running 24/7 without failure after years and so can I claim AMD is more stable? I afraid not.