Makestation

Full Version: Benchmarking some machines. WIP.
You're currently viewing a stripped down version of our content. View the full version with proper formatting.
I've recently ordered a new Thinkpad which should be arriving near the end of April if all goes well.
It's more or less a maxed out E595, only things that aren't maxed are:
  • RAM, since one 16GB module is enough to get me through the day and I can buy a used 16GB SODIMM for like 50$ on ebay to get the dual channel benefits
  • Drives, 1TB HDD is good enough because there's room for a second drive, and I've got like 5 SSDs lying around that I can pop in.
CPU is a Ryzen 3700U quadcore, and it only has integrated graphics.
See, that's what interests me. One of the reasons I bought my current laptop way back when, was because it had a dedicated GPU chip, not just the usual Intel UHD 620 or whatever was common at the time. This had an Nvidia 940M, and Nvidia being a decent company in terms of performance, it HAD to be good. So I bought it, needed a new computer for school anyway and didn't have much time to do all the research in the world.

The one I'm getting now only has integrated graphics with the Ryzen7: Vega 10.
According to Passmark, it's on the (very) low end of "high-end" cards, still doing about 6x worse than the relatively cheap <200$ GTX 1060 you can buy anywhere nowadays.
But, strangely enough, it's still rated about 50% better than the 940M.

Funnily enough, I don't doubt it at all. 
Japanese and Korean people buy the most Thinkpads in the world, I'm pretty sure (literally try to find Thinkpad reviews on YouTube and I shit you not 90% of the results are in JP or KR) and since both computer cultures there are relatively big on the gaming scene, most reviews I saw online had some sort of gaming demo in them by just pointing the camera at the screen.
And it looked alright. Not great, not terrible. Could play MW Remaster without too many hiccups at medium settings. Could play Fortnite 1080 60FPS on High, even though that's not really a good benchmark, I'm pretty sure my phone could as well.

Either way, one of the main issues I had with the current HP is that the thing would get HOT. I mean, boiling water no problem. GPU would get to 80C (176F) in a couple minutes playing almost anything on super low settings and lower resolution options. Then, it would throttle, shut down, go down to 1-2FPS for a good 10-15seconds, slowly go back to normal, then rinse & repeat. The worst part is, with this card and this CPU (7500U, U as in LOCKED) I can't change any options like voltage, so I can't underclock it to keep temps down. I can't overclock it to make it actually be useful. I can't remove throttling restrictions because, well, that would probably melt the case, but also because it would simply lock me out of those options to change any aspect of how the CPU operated.

So I decided my run with Intel+Nvidia has reached its end, and picked up this new Thinkpad.

In CAD... (1.35 * USD)
I paid 1250$+tax (total: $1400) for this HP.
I paid 1000$+tax (total: $1120) for the Thinkpad.

Pretty massive savings already. Keep in mind the HP is just over two years old now. Warranty expired not long ago.
But, usually when you pay more for something, you get a better quality product in return.
Maybe the HP is still going to be faster than the Thinkpad. Maybe Intel really does dominate the market of CPUs for a reason (that's probably going to drastically change this year, mind you.) Maybe this Vega 10 will actually be terrible, despite the Passmark numbers.

So, I've decided I'll be running my own tests.

As an added bonus, I'll also benchmark my old Acer that I'm using until the Thinkpad arrives.
I want to see how much laptops scam nowadays versus the actual sheer performance numbers.
I got the Acer as a Xmas gift from my parents back when I was like 13 or something. They got tired of me using the computer to play games or having to reinstall Windows because kiddo me would download 'free music' with a .exe extension. 
Thus, finding the price might be difficult, but maybe we have the receipt somewhere, or I can approximate its cost based on what old reviews said about very similar machines. At the very least, I could give my own appraisal of what it's worth today in terms of performance. I'm expecting AT LEAST a few 100.


THE TESTS:
First, I'll be using UserBenchMark since it's probably the easiest to actually run tests with, and takes no more than two minutes.
Then, ideally I also want to run PCMark 10 to get an idea of day-to-day use for productivity. It's got a basic free version that gives me scores, so that's nice.
Finally, to benchmark graphics performance, I'm gonna use three things:
1. 3DMark (Basic, Fire Strike + Night Raid).
2. Unigine Valley 
3. Unigine Heaven

I've never really used 3DMark in the past, but it also gives a straight score.
But, Unigine's Heaven benchmark has always shown me straight up how shitty my GPU was, so that'll be nice just to see what framerates are like, how many straight up get dropped, when it performs better, when worse, etc.
Finally, Unigine Valley is just fun. I'd straight up walk around in it for hours on end when I was younger and listen to some cool music. Just gotta enjoy its own beauty. But, yeah, it's a pretty cool benchmark too, especially for how many polygons it's got.

Now, that's all Windows stuff, but my last (bonus), laptop runs Linux exclusively (remember when I installed Debian a few days ago?)
None of the above benchmarking tools work on Linux, sadly, but there's another option I can use that will just spit out a number for me to use: GeekBench.
It's also proprietary/paid, but that's whatever. There's a free version I can use, and it's not an install, just some binaries you download so I'm not too concerned about it taking over my system.

So I'll run GeekBench on all machines to get an idea of how well old tech holds up against new.


SPECS:
Code:
          |  HP Envy x360                            |  Thinkpad E595                       |  Acer Aspire 5755G
==========================================================================================================
CPU       | Intel 7500U, Up to 3.5GHz, FAKE Quad *   | AMD Ryzen7 3700U, Up to 4.0GHz, Quad | Intel 2670QM, 3.1GHz Quad
GPU       | Nvidia 940M + Intel 620 UHD              | AMD Radeon Vega 10 (Integrated)      | Nvidia GT540M + Intel UHD ****
RAM       | 12GB DDR4 (2133MHz, 4GB Sol. + 8GB Mod.) | 16GB DDR4 (2400MHz) **               | 8GB DDR3 Sol. *****
Screen    | 1080P, 60Hz, 17", Touch                  | 1080P, 60Hz, 15.6"                   | 1366x768, 60Hz, 15.6"
Disk      | 512GB Samsung NVMe                       | 1TB WD(?) 5400 RPM + SSD ***         | 480GB Corsair Force LE
Remind me to never make a table again. I think I'm dying.

NOTES:
* - Their site lists it as a Duo, but it spoofs itself under Linux to make it look like it's a quad core, rather than appearing as 2C4T, it's spoofed to 4C4T under /dev.
** - Although the RAM spec. is actually 2666MHz, the CPU only supports up to 2400MHz in its factory settings state. I may overclock to make this worthwhile, may as well. Also note that it's a single module, so I won't be able to take advantage of the dual-channel performance boost (roughly 20% based on other tests.)
*** - It comes with an HDD. I'll be throwing an SSD in later. I'll update if necessary.
**** - I don't know the exact model, but I'm pretty sure it's 4000-series, even though old reviews I've found online mention 3000-series. Then again, their reviews also mention the Nvidia 630M, so it sounds like a slightly different model, or maybe they bought the year later.
***** - Don't know the clock speed either, but it's probably somewhere around 1200-1600MHz.

The drive is probably the most useless part of this benchmarking series, since it's the easiest to swap out anyway PLUS UBM gave me a +460% for the HP NVMe. 
In other words, nothing will beat it. I'm pretty sure that's what actually saved its overall score, just how fast the drive was compared to everything else.
Even if I can max out the SATA speeds in other computers, that little NVME is still going to be a good 3x faster after looking at its CrystalDiskMark grades. No exaggeration there.

I'll begin posting benchmarks tomorrow once I have some spare time, and hopefully won't be too high. I've run the tests before already just to make sure they work, and I need to fix UBM a bit first since there's some slight issues with using dual GPUs (i.e. 940M + UHD.)
THE NUMBERS: HP ENVY x360
(Still WIP, will update continuously)

Here's the current HP x360 Envy's scores.
PCMark 10 - 3084
Unigine: Heaven (PasteBin)
Unigine: Valley (PasteBin)
3DMark 11: Fire Strike - 1440
GeekBench 5: CPU

OVERCLOCKS:
3DMark 11: Fire Strike - 1604 (OVERCLOCKED GPU)
 PCMark 10 - 2970 (OVERCLOCKED GPU)
GeekBench 5: GPU [CUDA]
GeekBench 5: GPU [OpenCL]

Code:
PCMark 10:   |  Stock GPU     |  OC GPU        |  % Delta
===================================================================
Overall:     |  3084          |  2970          | -  3.7%
Essentials:  |  7024          |  7018          | -  0.0%
* App Start: |    9346        |  9517          | +  1.8%
* Web:       |    6710        |  6499          | -  3.2%
* Video Con. |    5526        |  5590          | +  1.2%
Productivity:|  5568          |  4863          | - 12.7%
* Spreadsh.  |    6304        |  4753          | - 24.7%
* Writing    |    4918        |  4976          | +  1.2%
DCC:         |  2036          |  2085          | +  2.4%
* Video Edit:|    2830        |  2844          | +  0.5%
* Photo Edit:|    2312        |  2425          | +  4.9%
* Rendering: |    1291        |  1315          | +  1.9%

Code:
3DMark 11: FS  |  Stock GPU     |  OC GPU        |  % Delta
===================================================================
Overall:       |  1440          |  1604          | + 11.4%
Graphics:      |  1648          |  1807          | +  9.6%
* T1 Avg. FPS: |    8.12        |  8.24          | +  1.6%
* T2 Avg. FPS: |    6.42        |  7.51          | + 16.9%
Physics:       |  4638          |  4886          | +  5.3%
* Avg. FPS:    |    14.73       |  15.51         | +  5.3%
Combined:      |  484           |  563           | + 16.2%
* Avg. FPS:    |    2.25        |  2.62          | + 16.2%

UL/FutureMark claim that with consecutive testing, results should typically fall into a ~3% delta. So, with those standards, anything that has a % delta (last column) of -3% to +3% should be disregarded.
I increased my GPU clock by 135MHz, and Memory clock by 185MHz. That's +14% and +14.5%, respectively.
I know that changes don't increase linearly with the frequency increases, as Hz is reciprocal, but I would hope for some bigger gains than whatever PCMark is throwing back at me.
Perhaps I'll re-run the tests.

Notes:

PCMark 10 Wrote:Overall pretty satisfied with the numbers. Yeah, the overall score is low; it even lists it as lower than your typical Office machine. Granted, but what's actually bringing the score so low? Two things: Rendering (1291) & Digital Content Creation (2036.) Photo Score gets an honourable mention (2312) for the top 3. 
But the other scores are well above average, for instance the Apps score at a whopping Over 9000. 
It basically measures how fast things can start up, so the speedy NVMe drive certainly helps. But Web/Essentials are also pretty high. 

So I'm confident that if the GPU was better, this score would be DRASTICALLY higher.

3DMark 11 Wrote:The fans started maxing even while starting the application. Chrissake.
The software warned me that I should disable G-SYNC before running any of the tests, as it could negatively affect performance. 
I decided 'f*** it' since I ran every other benchmark already and want as much control variables as possible. So I left it on.

Overclocking Wrote:In the past I've mentioned how difficult it is to overclock this d*** GPU.
And it is.
But magically, using a tool from 2014 called Nvidia Inspector, I got it to +135MHz for processing clock and +180MHz for the Memory clock. 
Apparently, those are probably the safest performance vs. stability numbers out there.
And it looks like a bad(?) Nvidia Driver update broke MSI Afterburner that for their OC. I use a ? because they probably did it to stop people from OC'ing.

I suspect that something went seriously wrong during the OC PCMark 10 Spreadsheet test. I'll have to look into it, but the only thing I did differently was OC the GPU, which should have little to no effect at all.

Invalid GPU Wrote:Both 3DMark and PCMark are owned by UL, and their sysinfo can't seem to figure out what my GPU is.
When the application is open here I see the name and all the info, but on the results page there's that little 'Invalid' warning. For whatever reason, the results formally display as 'Generic VGA' or whatever.
I think that might have something to do with faulty/outdated Intel drivers. Since HP never approved an update past 2018 for the 620, I need to manually install the files and hope my whole computer doesn't blow up.
THE NUMBERS: ACER ASPIRE 5755

As promised, I ran Geekbench 5 on the thing (done in console instead of GUI but still publishes results online.
And here's the results link:
GeekBench 5 - CPU
GeekBench 5 - GPU (Coming soon)

Code:
.            |  HP Envy x360    |  Acer Aspire    |  Delta
  CPU:       |==================|=================|================                        
  --Single:  |  837             |  688            |  -17.8%    
  --Multi:   |  1959            |  2591           |  +32.2%
  GPU:       |==================|=================|================
  --OpenCL:  |  7132 (OC)       |                 |  
  --CUDA:    |  6872 (OC)       |                 |


CPU Wrote:So single-core score is about -17.8% compared to the HP. But what's interesting is that the multicore/multithreaded tests scored MUCH higher at +32.2%, despite the processor being a 2nd gen i7 and the HP having a 7th gen i7. You'd think that Intel improved SOMETHING over five years of optimization, but wow, that's quite the difference. I guess they improved single-core performance, but shit, it's not really by a huge amount, and neither even managed to break the 1000 base-line score that GeekBench has configured and tests against.

Honestly kinda pathetic. I reckon if I reapplied thermal paste, it would be like another 10% faster. I've never reapplied thermal on this machine since I never knew that was a thing when I was a kid, and I've had this laptop since I was like 12.
New machine arrived.

Got scammed (i.e. I didn't think the performance hit would be too bad and only realized the SSD slot is M.2 only so I can't use any of my SATA drives) and the HDD is 5400RPM so loading anything is incredibly slow, but once it loads it's pretty decently fast. Need to use a modded Ryzen Master for OC though.

Not going to use it today for benchmarking as a result. I ordered a PM981 (970 Evo Plus OEM from China that's like half price) and another 16GB DDR4 2666MHz stick of RAM both for 200$CAD total, or about 1/3 of the price Lenovo would have charged me to put them both in for customization. Once they arrive, I'll pop 'em in, put the liquid metal cooling in, and run the benchmarks then. I'm expecting pretty fast speeds.

Tuesday at the very latest, but maybe the weekend if Amazon gets them to me my tomorrow with their 1 day shipping from Prime.
(May 7th, 2020 at 9:13 PM)Lain Wrote: [ -> ]Got scammed (i.e. I didn't think the performance hit would be too bad and only realized the SSD slot is M.2 only so I can't use any of my SATA drives)  and the HDD is 5400RPM so loading anything is incredibly slow, but once it loads it's pretty decently fast. Need to use a modded Ryzen Master for OC though.

WTF. I really wish they could get hard drives and the connections all on some universal configuration. I know, pipe dream, especially with all the different requirements for each.

Its 2020, why can't someone figure it out?
(May 7th, 2020 at 10:17 PM)Guardian Wrote: [ -> ]
(May 7th, 2020 at 9:13 PM)Lain Wrote: [ -> ]Got scammed (i.e. I didn't think the performance hit would be too bad and only realized the SSD slot is M.2 only so I can't use any of my SATA drives)  and the HDD is 5400RPM so loading anything is incredibly slow, but once it loads it's pretty decently fast. Need to use a modded Ryzen Master for OC though.

WTF. I really wish they could get hard drives and the connections all on some universal configuration. I know, pipe dream, especially with all the different requirements for each.

Its 2020, why can't someone figure it out?

I mean I could replace the HDD with a 500GB SSD I have laying around, but the extra storage is nice, and the PM981 is pretty cheap so it's not a huge hit to my wallet.
Not to mention it's tax return season so I might get like 900$ soon from the gov't after I apply for every benefit imaginable Smile
Don't mind paying for my own upgrades, Lenovo would have charged me 300$ for an OEM SSD which would probably end up being slower anyway, and it was like 400$ for the RAM upgrade as well (unless I got 2x8GB which would be the same price but I wanna have 32GB just to say I have 32GB and flex on all the other kids at school  Cool )

I'm glad Thinkpads are relatively open-source for upgrades. Their hardware manuals walk through any possible upgrade really, and there's enough community support if you basically want the f*** schematics LMAO. Might even coreboot if I can port it over and figure out which pins to short.
THE NUMBERS: Thinkpad E595

But first, Notes:
-- I threw a 970EVO Plus in there. Naturally, this is an insanely faster drive than most other drives on the market. It would take nearly a minute to boot before. Now, probably less than five seconds. It really does make a difference.
-- I added another 16GB of RAM but this should be negligible because the HP also used dual-channel RAM. Naturally, this should only (at best) get it back up to speed.

Now, Benchmarks.


UserBenchMark
Unigine: Heaven [Pastebin]
PCMark 10 - 3547
3DMark - Firestrike - 2510
GeekBench 5 - CPU
GeekBench 5 - GPU - 9040

Notes:
  • I tried some overclocking with RyzenAdj and making sure it worked with uProf, but honestly many of the settings don't seem to have any effect. Seems that either the BIOS is reverting changes on its own, or AMD drivers are doing that instead. You can't OC with the actual Ryzen Master software, since it won't run on mobile CPUs.
  • In some of the benchmarks, it seems that the RAM frequency is lower than it should be. For instance, it'll be listed as 1866MHz instead of 2400MHz (the max. allowed by the CPU.) Some people think that it's a result of mismatched CL ratings (i.e. CL17 and CL19) but they are actually the same. It seems that AMD is throttling the frequency when it isn't full-boosted. On top of that, looks like the boost isn't actually going up to the full 4GHz as anticipated, only hovering somewhere between 3.2-3.5. I reckon playing with TDP settings in RyzenAdj will fix that, but come on now.
  • RAM caching is the only way to get those really high SSD scores. I tried disabling it and it tanked the write speed by about 4x. Sadly, lots of benchmarking software doesn't allow for RAM cached drives to be tested, but seems I've had a couple flukes for it to have run.

GeekBench Wrote:All I can say is WOW. I was already expecting higher multithreaded scores for the CPU since although Intel does make CPUs that can handle extremely high single-threaded turbo'ing, their multithreaded scores were always lacking. And AMD has always been known to have mediocre single-threading but gleaming multithreaded scores. But WOW. Single-thread is just as fast, and multithread is 50% faster.

And for the graphics, AGAIN almost 50% increase. That's remarkable for an integrated graphics chipset. You can't make jokes about that stuff anymore, unless it's targetted at IntLEL because FFS, the performance really is just that much faster, and it doesn't have a 'high performance discrete Nvidia chipset' to try and boost itself with.

UnigineHeaven Wrote:As noted in its notes, DX11 performs way better. I wish I had the option of running it in DX12 to see if that would give me another boost, but maybe I'll have to figure it out with the help of 3DMark.

Still, almost double the score this time around, and I wasn't having a complete f*** aneurysm while watching the benchmark do its thing.

PCMark 10 Wrote:I was actually expecting a higher score for this one, since while watching it, it looped over application Startup forever. Must have watched it open Chrome, Firefox, GIMP, and LibreOffice at least 20 times and its individual score was still lower.

But again, in the graphics-related tests, WAY better.

I reckon it has something to do with not testing anything with multithreading. I'll be playing with it more since it really is a more of a 'productivity test' than 'melt my f*** computer by pushing 12million pixels with physics and high-res textures.'
Very nice thread. I refrained from replying right away so that the first few posts would be yours in the thread. You've divided it up in sections (I wish there was a way for thread authors to re-arrange posts so that updates showed up at the beginning). I benchmark pretty much everything I touch, so I always love seeing benchmarks of various other hardware. Especially when the process is documented (as opposed to random Geekbench Browser reports, where someone could be running a web page that's hogging an entire core during the bench mark, for all we know. Finna)

I have noticed that with dual channel RAM, there is very rarely any significant performance improvement EXCEPT for gaming. In anything graphically intensive, it's usually very heavily bottlenecked by the RAM for integrated graphics. Outside of that, it's usually only a few percent of an improvement, which is worthwhile nonetheless, but not necessarily "noticeable."

Memory performance matters though. AMD server chips, for example, have smoked Intel even on a lot of single threaded tasks now, but some tasks are still very behind for AMD chips. This is largely because AMD's latency to access main memory is significantly larger. Yes, they have more cache, but if it's not in cache at all, Intel can access it faster.

Very nice work overclocking as well. Especially on a laptop. Big Grin
Only saw reply now sry for late response

I've done the research on dual-channel memory and you're more or less right on that. And even then, the most performance increase you'll get is around 20% and only when you're trying to address something at 128bits (2x 64bit).

But this is why it's incredibly important to consider from my results:

In the UserBenchMark results, you can actually see the history of all the tests I did to see what would happen if I changed some settings, and most notably, my SSD seemed to be performing absolutely f*** terribly during my first few tests:

https://www.userbenchmark.com/UserRun/27878517

Yes, it's still significantly faster than the average, but notice that it's at the bottom 3% of performance (3rd percentile) among other 970 EVOs.
Why?
RAM Caching was disabled.
In other words, the CPU was the bottleneck in writing to the disk.

Turn RAM caching on, and it gets to almost the exact mid-point for performance among the same model, and almost exactly double the total speeds.
Because it has double the addressing capability that a 64bit CPU can't provide natively.

Now, you can f*** with it even more by using memory and disk compression, but that causes ridiculous CPU load which I don't need in a laptop, so that's gone.
But, the performance benefits are certainly there to make the most of the disk.

And yeah that's why people only mention the benefits while gaming; games require a f*** of reading data from assets and writing data to the display memory segments. So naturally you'll see that performance boost especially with the cache enabled.