TR Forums

Hello all!

I must upgrade my aging FX8350 + MB + ECC RAM because the M/B has started having hardware issues (loses CMOS settings -- it's not the battery) and the extra performance is welcome. Budget is not a major issue, although I usually shop for price/perf.

Use is (a) casual desktop use, (b) photo editing and video compression for storage, (c) gaming 1440p, (d) heavily multithreaded bioinformatics analyses that can take hours or even days to complete. No overclocking is planned.

I have the following hardware (other than the FX8350)
- RX480 8GB Nitro OC edition
- 2xWD RE 4TB
- Samsung 840 500MB, Samsung 850 EVO 750 MB
- Antec P183
- Corsair H80
- Seasonic 750W Gold
- A bunch of high-end silent 12cm fans

I will get a CPU+MB+RAM as a minimum. Waiting for Zen, obviously, before I make my decisions.
Current candidates are either 7700K + nice MB + 2x16GB DDR4 3000 or Zen 8c16t @ < $500, nice MB and same RAM.

Questions
- This time I thought I'd go for non-ECC RAM. Finding ECC 16GB non-registered DIMMs is hard. They are more expensive and slower. I'd like 2x16GB at first and add 2x16GB later. Any comments?
- I will probably upgrade the Antec P183 to the Corsair Obsidian 750D Airwflow edition ($145 local price). What do you think?
- Since I'm going to have to reinstall windows, should I buy an M.2 SSD? Both my Samsung SSDs have the bug that reduces speed over time, so I'd like to avoid Samsung if possible, as a matter of principle. I noticed that if I want 1TB, prices hover around the $450 mark for good SSDs (ideally, non-TLC). This is a lot of money. Do you think SSD prices will fall a lot with Optane? Should I wait?
- Apparently, the Corsair H80 works with both the Socket 1150 and AM4 retention. Any other information?

Other ideas?

Wait for Zen reviews, then decide. 

Optane SSDs will be very pricey and probably wont trickle into consumer use for a few years. I haven't heard anything about Optane compatibility on Zen/AM4.  It also remains to be seen how much perceptible advantage Optane will provide in consumer workloads.  NVMe SSDs are seeing slow adption because in real world tasks, there isn't much difference between them and SATA SSDs.  NVMe M.2 SSDs are priced high (~$0.50/GB) because their raw performance is higher compared to SATA (see above related to perceptible performance differences).  In all reality, M.2 SSDs actually cost lest to manufacture.  As M.2 gains market share, expect prices to go down some, but as long as 2.5" SATA SSDs are on the market, they will set the low end of the cost curve.  At the very least, I would only recommend M.2 SSDs for use as an OS drive, if you need a second SSD, get a 2.5" SATA drive.

The large majority of consumers don't need ECC RAM.  If your usage truly requires it, you'd know though.  I'll leave that for someone else to comment.

Why start this thread now if you're waiting for Zen?
Wait for Zen!

RAM and M.2 questions cannot be answered until the Zen boards and the ASMedia chipsets have been reviewed. 
Case, airflow, and thermal questions cannot be answered until we know how hot Zen runs and whether you need more than you have.

Basically you're just creating yet another Zen speculation thread, of which there are already too many.

Yeah, like others have said, if you can wait until Ryzen benchmarks/pricing info comes out that would be best.

Several of your workflows look heavily threaded so being able to get a solid 8 core / 16 thread chip for far less than $1000 would be ideal.

Chrispy_ wrote:

Why start this thread now if you're waiting for Zen?
Wait for Zen!

RAM and M.2 questions cannot be answered until the Zen boards and the ASMedia chipsets have been reviewed. 
Case, airflow, and thermal questions cannot be answered until we know how hot Zen runs and whether you need more than you have.

Basically you're just creating yet another Zen speculation thread, of which there are already too many.

I said I'm waiting for Zen. The thing is, I would like to (a) plan my expenses 3-4 months in advance (1T SSD is a sizeable chunk of money) and (b) keep my eyes open for the occasional offer, if I have settled on a specific piece of hardware. For example, right now the Obsidian goes for $145 (instead of the usual $160). Should I get it?
I haven't yet settled on Zen (although I like what I heard) and I'm not asking any question about Zen because I don't think anyone has the answers.

DPete27 wrote:

Optane SSDs will be very pricey and probably wont trickle into consumer use for a few years. I haven't heard anything about Optane compatibility on Zen/AM4.  It also remains to be seen how much perceptible advantage Optane will provide in consumer workloads.  NVMe SSDs are seeing slow adption because in real world tasks, there isn't much difference between them and SATA SSDs.  NVMe M.2 SSDs are priced high (~$0.50/GB) because their raw performance is higher compared to SATA (see above related to perceptible performance differences).  In all reality, M.2 SSDs actually cost lest to manufacture.  As M.2 gains market share, expect prices to go down some, but as long as 2.5" SATA SSDs are on the market, they will set the low end of the cost curve.  At the very least, I would only recommend M.2 SSDs for use as an OS drive, if you need a second SSD, get a 2.5" SATA drive.

The large majority of consumers don't need ECC RAM.  If your usage truly requires it, you'd know though.  I'll leave that for someone else to comment.

I read that prices would go down with Optane, but didn't really read into it. Maybe they are talking about the high end. Would suck buying a $500 M.2 SSD to find it at $250 a few months later.

All the press releases I've seen have been something along the lines of 10x cheaper than SDRAM and 1000x faster than NAND.  Notice that doesn't imply that 3DXPoint will be cheaper than NAND. For a quick and dirty: sticks of DDR4 RAM are running about $5/GB and large capacity SSDs are running close to $0.25/GB.  There's still some breathing room there, even if you stick to the 10x cheaper claim (which I doubt in at least the first year). 

3DXPoint will give NAND competition, yes.  How much that competition is and how much it will actually affect NAND prices remain to be seen.  Remember that NAND will still control the majority of the market for quite a few years.  NAND isn't only used in SSDs you know!  

I generally avoid being an early adopter of any technology. It's usually buggy, not will supported, is quickly improved, and expensive. I wouldn't consider xpoint a selling point at this point in time. If it drives down prices it won't be for a while, either.

The thing that makes this thread pointless is that you're trying to plan in advance but every aspect you try to plan is impossible because there is no pricing for Zen yet, and no pricing for 3DXPoint either, with the caveat that I know it will DEFINITELY be far more expensive than NAND. It's going to replace the top-tier performance SLC enterprise SSDs.

[size=100][font=Source Sans Pro, Trebuchet MS, sans-serif]When they say it'll be cheaper than SDRAM, 256GB of SDRAM will set you back about $1500. So maybe 3DXPoint will only be $1000 making it 16x more expensive than NAND. It's going to be enterprise-only at first, and I suspect the first consumer parts won't appear until 2019 or 2020 at the earliest, when they'll be like PCIe NVME storage is now - available to consumers but more expensive than the mainstream stuff by a non-trivial margin.

Chrispy_ wrote:

The thing that makes this thread pointless is that you're trying to plan in advance but every aspect you try to plan is impossible because there is no pricing for Zen yet, and no pricing for 3DXPoint either, with the caveat that I know it will DEFINITELY be far more expensive than NAND. It's going to replace the top-tier performance SLC enterprise drives. Expect $10/GB as a very rough ballpark figure.

When someone says it'll be cheaper than SDRAM, that's like saying that a Learjet is cheaper than the Space Shuttle; Neither are particularly cheap vehicles in the first place.

Well, even though you are stuck on Zen (on which I asked absolutely no question), your answer on 3dXPoint is genuinely useful to me and not pointless. Anyway, based on what has been discussed here, my conclusion is to keep my current SSD because M.2 is probably not much faster in everyday use and the Optane/3DXPoint stuff too expensive in the near term.
Nobody commented on the Corsair Obsidian case, but the reviews are quite good. I think I'll get it if I see a nice offer.
Thanks to all for your answers.

I was updating my reply above as you were replying with more info - but yeah, it's a long way off before 3DXPoint is even remotely competitive with NAND when it comes to cost/GB.

As for the Obsidian 750D, it's a good case. I don't have any personal experience with it but I have used the 350D mATX version of it on numerous occasions and I've no real complaints at all. The 750D is (I think) the same basic design/style/material/quality/features - but just bigger. It is actually VERY big, even compared to the P183 which I used to have.

I can't comment on the 750D specifically, but I do have a 550D. It's a reasonable case -- has done the job well enough for me, although if I was in the market for a new case I would probably end up looking at some of the Fractal Design cases (Define 4 or Design S, depending on if I went ATX or mATX). 

I don't see much point in getting ECC ram for home use. Save the money and get faster / cheaper non-ECC. 

January/February/March is generally worst time to buy. Prices always go through the roof. Only people buying right after the holidays are unplanned emergencies/no choice and retailers know it.

Normally i'd say forget Ryzen because you should never get first generation of anything but it is about the only thing changing in PC market, dont expect anything new from Intel. (Except for PCI-Express 4.0 which i personally would wait for, but likely out of your time window)

I would definitely forget your 8xx experience with Samsung. Really the only decent choice for storage are the Samsung EVO and Pro SSDs. If your PC has any non-entertainment roles.

Even after you get a new system, the old FX8350 +ECC one should still be very useful for keeping around so it's worth fixing.  The flash in the BIOS chip can indeed run out of P/E cycles and wear out so I'd spend the ~$15 to try a replacement BIOS chip.  If it's dual-BIOS you could even try just lifting some legs on the primary BIOS chip to disable it.

I stick with AMD whenever I can and whenever it makes sense, but if I were in your place and I can't live without a PC I just might be forced to get Intel because Zen isn't out yet and AMD has been quiet regarding a firm launch date, AFAICT. Pricing is another unknown factor. Yes, Zen is just around the corner but I'd rather not be a tester of a completely new platform or be an early adopter for several reasons. Given these, I might be tempted to go with Skylake but I know I'll regret jumping hastily especially if Zen turns out to be great and priced well as they purport it to be. Just so I'll have a PC to use I'll probably just get a cheap APU like the A8-7600 and pair it with a cheap mobo to tide me over, or maybe get a cheap AM3+ board with which I can pass on the FX to someone else later. Everything else carries over. Then after Zen has been out in the market for a while and prices have stabilized and the kinks have been ironed out (not to mention we would have learned a lot about it by then), I'll make my decision.

blahsaysblah wrote:

January/February/March is generally worst time to buy. Prices always go through the roof. Only people buying right after the holidays are unplanned emergencies/no choice and retailers know it.

I hadn't thought about that. In fact, this explains why the prices of several items I keep in my online shopping cart have increased significantly. I can still wait a little bit.

bfg-9000 wrote:

Even after you get a new system, the old FX8350 +ECC one should still be very useful for keeping around so it's worth fixing.  The flash in the BIOS chip can indeed run out of P/E cycles and wear out so I'd spend the ~$15 to try a replacement BIOS chip.  If it's dual-BIOS you could even try just lifting some legs on the primary BIOS chip to disable it.

That is a very good piece of advice! I had to order a replacement BIOS chip in the previous century after a failed firmware flash attempt and it worked fine. I didn't think of doing the same here, but I could. Had cost me something like $20 back then. Know any reputable sites for that? Ones that ship to Europe?
Also could unplug the main BIOS and see what happens with the secondary.

bfg-9000 wrote:

Even after you get a new system, the old FX8350 +ECC one should still be very useful for keeping around so it's worth fixing.  The flash in the BIOS chip can indeed run out of P/E cycles and wear out so I'd spend the ~$15 to try a replacement BIOS chip.  If it's dual-BIOS you could even try just lifting some legs on the primary BIOS chip to disable it.

Unless he's been flashing the BIOS on a daily basis for the past few years I seriously doubt he's anywhere near hitting the P/E limit on the BIOS chip.

If the battery tests good (should be 3 volts, give or take a tenth of a volt or so), then either there's a bad connection somewhere between the battery and the chipset, or the chipset itself has gone bad. (The CMOS and clock are incorporated into the chipset these days instead of being a separate device.)

@ptsant - Are you absolutely sure the battery is good, i.e. you checked the voltage with a voltmeter? Just because you put a new one in doesn't automatically mean it is OK; I've had ones that were bad straight from the package before. And (stupid question) are you positive that the CMOS jumper isn't in the "clear" position?

just brew it! wrote:

Unless he's been flashing the BIOS on a daily basis for the past few years I seriously doubt he's anywhere near hitting the P/E limit on the BIOS chip.

I've actually run out of P/E on firmware EEPROM chips before, and even with no flashing of updates ever.  First of all such chips have far fewer P/E cycles available than other flash chips, there are no wear-leveling mechanisms, and most importantly they may suffer from stupid programming where the same cell is written to over and over to update something as dumb as power-on hour count.

I still have a monitor that hasn't been able to store setting changes for more than ten years (I work around this with the video driver) and sure enough the power-on hours haven't changed ever since.  Similarly, I had a router burn out its NVRAM just from logging Bandwidth and IP traffic monitoring to it.  It's important to note that long before a flash chip completely goes read-only it gets increasingly flaky and unable to store settings long-term.

That said I've been using the BIOS to dual and triple boot from different hard drives for nearly 20 years (since that commonly became a convenient option) and haven't worn out a motherboard BIOS chip yet.  And some of these boards definitely did store power-on count and hours to flash (in particular, the Abit ones).

bfg-9000 wrote:

just brew it! wrote:

Unless he's been flashing the BIOS on a daily basis for the past few years I seriously doubt he's anywhere near hitting the P/E limit on the BIOS chip.

I've actually run out of P/E on firmware EEPROM chips before, and even with no flashing of updates ever.  First of all such chips have far fewer P/E cycles available than other flash chips, there are no wear-leveling mechanisms, and most importantly they may suffer from stupid programming where the same cell is written to over and over to update something as dumb as power-on hour count.

I still have a monitor that hasn't been able to store setting changes for more than ten years (I work around this with the video driver) and sure enough the power-on hours haven't changed ever since.  Similarly, I had a router burn out its NVRAM just from logging Bandwidth and IP traffic monitoring to it.  It's important to note that long before a flash chip completely goes read-only it gets increasingly flaky and unable to store settings long-term.

That said I've been using the BIOS to dual and triple boot from different hard drives for nearly 20 years (since that commonly became a convenient option) and haven't worn out a motherboard BIOS chip yet.  And some of these boards definitely did store power-on count and hours to flash (in particular, the Abit ones).

Well, yeah... brain-damaged firmware that repeatedly rewrites the same flash sectors will definitely cause problems. But a BIOS on a motherboard (which has a CMOS memory to save stuff that gets updated semi-frequently) should have no such issues since there's no reason to write to the EEPROM unless the BIOS is being updated.

It's also not necessarily true that BIOS EEPROMs have less P/E cycles than the flash used in thumbdrives and SSDs. AFAIK (last time I checked) they were still SLC, with P/E counts at least in the thousands (if not tens of thousands). You're correct that there's no wear leveling though, so brain-dead firmware like you mentioned above will wear them out eventually.

just brew it! wrote:

Well, yeah... brain-damaged firmware that repeatedly rewrites the same flash sectors will definitely cause problems. But a BIOS on a motherboard (which has a CMOS memory to save stuff that gets updated semi-frequently) should have no such issues since there's no reason to write to the EEPROM unless the BIOS is being updated.

Yes, settings are lost when the battery is removed so they are definitely volatile.  But the power cycle statistics I mentioned are not--those are flashed.  If it's just corruption causing the trouble then flashing the BIOS again (even the same version) to overwrite should fix it.  Or it could just be faulty.  It's probably still worth trying to reuse the FX8350 and ECC DDR3 though which aren't usable for a new build, so the question is if it's worth gambling on a BIOS chip when a new $55 mobo would definitely fix the problem (probably a lot more in Europe but the OP didn't seem too concerned about cost).

ptsant wrote:

 Know any reputable sites for that? Ones that ship to Europe?

eBay is likely the most convenient place.  The chips themselves are commodity items and aren't flashed until you order them.  It costs a lot more from the OE manufacturer in case you are leery of the NSA or FSB possibly running the shop but some places have been in business for many years with a good reputation.  First try a reflash and then the backup BIOS though.

ptsant wrote:

I have the following hardware (other than the FX8350)
- RX480 8GB Nitro OC edition
- 2xWD RE 4TB
- Samsung 840 500MB, Samsung 850 EVO 750 MB
- Antec P183
- Corsair H80
- Seasonic 750W Gold
- A bunch of high-end silent 12cm fans

Questions
- Since I'm going to have to reinstall windows, should I buy an M.2 SSD? Both my Samsung SSDs have the bug that reduces speed over time,

To the best of my knowledge the 850EVO is not affected by slow reads like the 840 series were.Totally different type of NAND 840 planer TLC.
850EVO-3D TLC.And one of the better drives on the market.

just brew it! wrote:

bfg-9000 wrote:

Even after you get a new system, the old FX8350 +ECC one should still be very useful for keeping around so it's worth fixing.  The flash in the BIOS chip can indeed run out of P/E cycles and wear out so I'd spend the ~$15 to try a replacement BIOS chip.  If it's dual-BIOS you could even try just lifting some legs on the primary BIOS chip to disable it.

Unless he's been flashing the BIOS on a daily basis for the past few years I seriously doubt he's anywhere near hitting the P/E limit on the BIOS chip.

If the battery tests good (should be 3 volts, give or take a tenth of a volt or so), then either there's a bad connection somewhere between the battery and the chipset, or the chipset itself has gone bad. (The CMOS and clock are incorporated into the chipset these days instead of being a separate device.)

@ptsant - Are you absolutely sure the battery is good, i.e. you checked the voltage with a voltmeter? Just because you put a new one in doesn't automatically mean it is OK; I've had ones that were bad straight from the package before. And (stupid question) are you positive that the CMOS jumper isn't in the "clear" position?

Thanks for the advice. I did both of those things: clear CMOS jumper is at default place and battery measured at 3.27V. I also dusted the battery contacts.
I didn't realize the CMOS was on the chipset. Anyway, ordered a BIOS chip online ($15 with a tool to remove the old one). We'll see what happens.

bfg-9000 wrote:

just brew it! wrote:

Well, yeah... brain-damaged firmware that repeatedly rewrites the same flash sectors will definitely cause problems. But a BIOS on a motherboard (which has a CMOS memory to save stuff that gets updated semi-frequently) should have no such issues since there's no reason to write to the EEPROM unless the BIOS is being updated.

Yes, settings are lost when the battery is removed so they are definitely volatile.  But the power cycle statistics I mentioned are not--those are flashed.  If it's just corruption causing the trouble then flashing the BIOS again (even the same version) to overwrite should fix it.  Or it could just be faulty.  It's probably still worth trying to reuse the FX8350 and ECC DDR3 though which aren't usable for a new build, so the question is if it's worth gambling on a BIOS chip when a new $55 mobo would definitely fix the problem (probably a lot more in Europe but the OP didn't seem too concerned about cost).

ptsant wrote:

 Know any reputable sites for that? Ones that ship to Europe?

eBay is likely the most convenient place.  The chips themselves are commodity items and aren't flashed until you order them.  It costs a lot more from the OE manufacturer in case you are leery of the NSA or FSB possibly running the shop but some places have been in business for many years with a good reputation.  First try a reflash and then the backup BIOS though.

I looked at MBs, a decent one is about $120 here (cheap ones usually don't support ECC) and the one I currently have is a top-of-the line Asus Crosshair Formula V. Anyway, the computer is usable, it just takes a small ritual to make it power on (hard power-cycle, clear CMOS, re-enter BIOS settings), so I thought I'd rather wait a couple of months.
I already tried a reflash. No luck. For the BIOS chip, I didn't look far: picked the top Google search result (https://www.biosflash.com/e/BIOS-Chips-ASUS.htm) which looked quite decent and expect to receive in ~5-7d.

What kind of heavily threaded bioinformatic analyses? Like Tuxedo Suite (i.e. RNA-seq) stuff? I've never really found any very good benchmarks for any of the common genomic tools, since most people use servers more than desktops. (I'd love if you could PM if you know of any)

I'm wondering if 8C/16T- or 12C/24T Xeon-D boards make sense for this, since all I've read is that core count matters a lot more than clockspeed since these tasks parallelize well. For example see this HPC benchmark here. And it'd get you into 10 Gbe, which might become relevant as data increases? Is everything actually on that spinning rust?

Edit: also kicks ass in NAMD

double edit: I thought I read "light gaming" - nevermind, 2 GHz isn't gonna cut it for gaming 
cheers!

Yeah, general consensus seems to be that the 850 EVO is decent. I'm still avoiding it on principle, over the 840 EVO fiasco. If they manage to keep their nose clean for another product generation, I may take their SSDs off my "avoid" list.

From what I have read Optane is currently being billed as a cache point between the hard disk (HDD or SSD) and the RAM just like the first tiny SSDs were back in the day. I really doubt anything will come of this other than people who NEED the bleeding edge. Maybe after a while when that same tech expands to competitively sized SSDs and is used as primary storage and not a hard drive cache.

I think you ought to wait for the Zen vs Kaby comparisons in the next few months and look into M2 drives if you're saturating the SATA bus.

just brew it! wrote:

Yeah, general consensus seems to be that the 850 EVO is decent. I'm still avoiding it on principle, over the 840 EVO fiasco. If they manage to keep their nose clean for another product generation, I may take their SSDs off my "avoid" list.

This is exactly what I'm thinking. Current alternatives I am considering are Intel and Plextor drives, because they have a reputation for reliability.

llisandro wrote:

What kind of heavily threaded bioinformatic analyses? Like Tuxedo Suite (i.e. RNA-seq) stuff? I've never really found any very good benchmarks for any of the common genomic tools, since most people use servers more than desktops. (I'd love if you could PM if you know of any)

I'm wondering if 8C/16T- or 12C/24T Xeon-D boards make sense for this, since all I've read is that core count matters a lot more than clockspeed since these tasks parallelize well. For example see this HPC benchmark here. And it'd get you into 10 Gbe, which might become relevant as data increases? Is everything actually on that spinning rust?

I have access to "heavy macinery" at work, but I use the home computer for development, meaning I can test (and break) whatever software I want. RNAseq is one of the things we do (STAR, Hisat), exome variant calling (BWA, GATK, Mutect2 or VarScan) is another. The fundamental algorithm, sequence alignment, is very well parallelized with practically perfect scaling but some of the other stuff is heavily IO bound (for example, sort 150GB of files and find duplicate entries/"reads"). My own code is often parallelized for stuff that takes time, but runs at a high level (in R). Usually, I can spawn anything from 2 to 32 threads without difficulty if I want to, so I expect I can get quite good scaling if I don't run out of memory.
I don't have benchmarks beyond the machines I actually own, which are an i7 Macbook Pro (Haswell), the FX8350 and a Kaveri Athlon 860K. I remember when I got the FX8350 that I found an improvement from +40% to 100% in my workloads compared with the Phenom II 965 it replaced, so I was quite happy, but that was based on old software. Depending on the exact algorithm, compile flags (optimizations) and the like, the single-threaded performance is usually best in the FX8350 (bigger cache, I suppose) but sometimes the Kaveri has the upper hand, possibly because of newer architecture and 1866MHz non-ECC RAM (vs 1333MHz ECC on the FX8350). The Macbook single threaded performance is comparable, at a much lower frequency, but with multithreading it accounts to (in total) slightly less than the performance of the 4c Kaveri (~90% if I remember correctly), which is sort of amazing when you consider the difference in thermal footprint.
I will try to prepare a small test suite for the exome pipeline so that I can quantify the change from the FX8350 to the new CPU.