Firestrike graphics vs passmark? Which is more accurate?

Hi I was just curious if anyone has info on how reliable fsg and psm are? What I mean is, ok specifically in fsg the 970m is at like 7300 and the 980m is at around 8000 and passmark it's 4224 vs 5954. That's a pretty big discrepancy between the two right? I'd throw unigine in there but I've yet to find any list of median scores by them for comparison.

Trying to decide between a 980m or 970m for my dell m6700 3940xm. The 980m is around 70 bucks more and obv 100w+ versus the 970m's 80w which seems to suggest that I could overclock the 970m if I wanted to pretty easily up to 980m stock performance(and find an elusive unlocked vbios). 980m OC would be obv pretty hot to OC. I've got the overhead I think with an oem 240w psu. I'm only at 1080p. Seems like if fsg is the more trustworthy then it wouldn't be worth the extra money for the 980m.

 

Use real games as benchmark. Unless you're not interested in playing any ;-)

 

That's kinda useless if you don't have 2 of the same laptops with two different cards though. The synthetics are there for a reason, to get an idea across different systems of what a graphics card will perform at. But yeah ofc I can just lookup the yt stuff.

 

As a followup I guess I was concerned about the FSG ratings between a 970m and 980m, the 970m was 7200 and the 980m was a 7900 which makes no sense, kinda like the "adjusted" passmark ratings which make no sense, when compared to the raw scores they post in most cases. I'm assuming bc they either use the median and many people seem to have badly under performing systems or they have some other statistical formula that doesn't work too well.

I see lately that FSG top cards has actually corrected this with a 980m at 9800 and a 970m at 7500 which makes more sense and is in the actual range I got on both cards. I was trying to decide between them when I first posted this thread. Notebookcheck has accurate ratings as an alt reference.

https://benchmarks.ul.com/compare/b...RE&reverseOrder=true&types=MOBILE&minRating=0

 

The non-SLI FSG score of each of my 980ms undervolted and running super cool at 1.2 GHz is around 11k.

Where did you get that 8000 figure?

In 3Dmark every run counts in the database and in turn affects the median so let's say you have a performance issue with your card and you run 3Dmark to check it, bam that goes into the database as a run.

Don't use 3Dmark median at all in getting accurate performance information of a GPU.

 

The 7900 was on the link I dropped above until about 2 weeks ago for the 980m, as I said they updated it to 9800. That was kinda the point of the original post. It was perplexing to me since it was only a few hundred points above the 970m at that time, which makes no sense, it's supposed to be 25-30% higher than the 970m.

I don't use the median I'm assuming FSG/UL uses it, there's no way for me to use it anyways as I only have my card to use not other cards and they don't post the last 5 raw scores like passmark does, which you could get a median between 5 scores. My stock 980m is at 9300. The 970m I had in here recently was 7200 stock and 7800 w the slight msi oc within range of the current fsg score of 7450 for the 970m. You must have an unlocked vbios to get that 11k I would think.

 

I have my own vBios, that I tuned myself to fit my cards.

Actually one of my cards has all 6 MOSFETs and can be pushed much further but for the two cards to perform well in SLI I had to make some compromises.

The 980ms in my opinion have much better potential than their weaker siblings the 970ms.

 

Wow 980m sli sounds pretty sick setup. Yeah I'm vanilla except for the min nvidia allows +135core and I have +200 memory, haven't done a fsg test on it like that yet I'd imagine like +500 points or so to around 9800. I had a quadro k4000m that I had oc'd using an unlocked svl7 vbios that was 64% higher fsg score from stock lol, nvidia really gimped those things and it was a 100w card. The little graph that fsg gives you with similar systems had my rating WAY over to the right/high than all the other k4000m's it was awesome! Don't think there's that much room with a 980m it's not gimped too bad lol

 

For 70 more I'd take the 980m.

*Full disclosue, I have a 980m 8GB ...and will until the RTX 4000 series...

**Full disclosue, ...unless my local microcenter has a stupid good deal on something with the 3000 series.

 

980m has had crazy longevity and will probably be great for gaming for another couple years, I envy the people who bought a 980m laptop 3 years ago

 

Flagship GPUs always tend to take longer before obsolescence, unless there is a major paradigm shift.

980M came out 5 years ago, so 3 years ago it would’ve already been replaced by Pascal.

And yeah, it was a good card for its time and closed the gap between mobile and desktop significantly, being about 70% of the desktop 980 performance, while the preceding 680M and 780M were only about 50% of desktop 680 and 780 performance respectively. My 980M had all 6 core FETs present and I ran it OCed to 1350/6000 at 1.1625V as my daily driver, which basically turned it into a substantially less power efficient mobile GTX 1060 (150W vs. 80W).

That being said, the GTX 1080 was a great upgrade for me as the 120Hz itch was too strong to resist and the 980M was starting to show its limitations there. Undervolted to around the same 150W as my previous souped up 980M, the 1080 is about twice as fast.

 

5 years.

As with my last laptop, I think my main problem with games in its final 12-18 months will be CPU related.

 

I think for 60 FPS your CPU will ride out this generation just fine. The issue with Clarksfield was that it was such a far cry from desktop CPUs, like half the clock speed. 10 years ago the power efficiency of mobile CPUs was severely lacking and you had to choose between more cores or higher clocks.

 

For 1080p older games 980m works great tbh, esp since I'm not a bells and whistles player no problem turning down bl3/rdr2/ow to 900 or 768p for 60fps could do 1080p in those but I don't like over 75% my gpu on my system and I like 60fps. My 980m I got as "not working" for 110$, guy was just fed up in a couple of games he had problems with it probably a driver issue or something i've had no problems with it. I'm one of those guys that still has a galaxy s5 works great for my basic smartphone needs etc Trying to cut way back on games, you know that life thing lol

 

My last laptop had a 1st gen i7 720qm. It choked on some games that needed more speed. With 8 efficient multithreaded cores in the consoles, I don't expect to go unscathed at all if I go through to 2021.

 

I know, I had a 740QM myself. Even with a lowly (compared to desktop standards) 5870M GPU, that sub-2GHz first-gen i7 still bottlenecked me in games like BF3.

You have to keep in mind that the next-gen games will have a large jump in GPU requirement as well. So if you have to lock to 30 FPS because your 980M can’t keep up, the load on the CPU will go down accordingly.

 

well your 1080 will last you another 3 years at least so your laughing as well

 

Not if more games continue the trend of running especially poor on Pascal like RDR2 does.

Some of these huge performance deltas between previously comparable Maxwell/Pascal and Turing/AMD cards in recent titles is giving me a sense of deja vu to what happened with Kepler circa 2014/15.

 

yeah thats true.....I agree

 

I had the same GPU. I ran AC:4 on low and high/v.high in the same 24-27fps. Luckily, Wolfenstein really only bogged down in the safehouse between missions and once at an escape cutscene. Ran DA:O fine with a little of everything incl godrays ...not enough vram for full foliage though. [One and only artifact, yellow rectangles].

GPUs usually hang in through launch era releases relatively well if they were doing OK going in ...they're mostly just cleaned-up last-gen titles. I prefer graphics over 60fps, so I'm not easy on the CPU. It might be a little easier this round as last time IPC needs basically doubled [Passmark single-thread score of 750 going to 1500 approximately], while I'd guestimate something scoring over 2000 now is common place [My 4710HQ hits 2200 w/OC]. I once did some math 2-3yrs back that figured to nextgen console ports wanting CPUs at 3300-3400 sngl-thread passmark score, but while we're about on track for that timing to nextgen launches, I think a score of 2800 should be enough ...though 3000 or better is where I'd want to be for something I want to keep all the way through.

 

Yeah, Clarksfield had 4C/8T but they were clocked so low (1.73GHz all-core turbo on the i7-720QM--yikes!). So while it was generally fine in well threaded engines like DA:I's Frostbite and W:TNO's id Tech 5 (virtual texture transcode was very efficient at utilizing threads), in a poorly threaded engine like AC4's AnvilNext, the performance tanked as you saw. And our GPU did the CPU no favors either, as AMD's inefficient DX11 driver adds additional CPU overhead. Mobile Sandy Bridge tore Clarksfield a new one, with all-core turbos in excess of 3GHz and a 25% IPC uplift on top of that.

A lower frame rate is actually easier on the CPU since like I said, CPU utilization is directly related to FPS. If you turn up the eye candy and lock a game to 30 FPS, it really doesn't take much CPU power to sustain that.

Eh, I'm really not sure about that one. Just look at what happened to Kepler. It was fine through about the first year of the current-gen era, but after that started to lag significantly behind Maxwell and GCN in many multiplat AAA releases, and that was a big controversy at the time. This gen isn't even over yet and I'm already starting to see the same trend on Maxwell/Pascal.

In terms of single threaded performance or IPC, your CPU isn't that far off-pace from the latest and greatest because IPC and clock speed increases have stagnated in recent years. CPU advancements have mostly focused on increased core/thread count (and bigger caches and more mem BW). Haswell is only about 15% behind Zen2 in CB R15 normalized to 3GHz:



So a 3.5GHz all-core 4710HQ isn't exactly a slouch as far as the speed of its individual cores compared to the low-3GHz Zen2 in the next-gen boxes. If your CPU does hold you back next-gen, it won't be because of IPC/ST performance (the problem we had with Clarksfield), it will be because of core/thread count, as game engines have gone wider instead of faster in terms of processor requirement in recent years

 

Yeah I was aware that fps is cpu thing but you can't discount gpu in that imo esp w the same cpu but upgraded gpu, when I went from heavily OC'd k4000m to a 980m that's a huge step up and I can't believe fps isn't vastly improved by that w the same cpu.

Once again fsg has it wrong, they now changed back the 980m to a 7900, only a step above the m4000m which is the quadro eqv of the 970m. My 980m stock gets 9300. I just wonder why the swings it was 9800 previously that's a huge difference. Yeah checking the yt games benchmarks much more accurate and notebookcheck. I'm thinking it's all just automated no human involved kinda like when gmail locks you out bc you changed wifi or travel, garbage. 970m was around 7900 previously now it's back to 7200, gtx should beat quadro lol

https://benchmarks.ul.com/compare/b...RE&reverseOrder=true&types=MOBILE&minRating=0

 

Yeah some of the results on that page are utter nonsense. My FSG score is about 10K higher than what they list for the GTX 1080N, not to mention they rank the 1080MQ way above the 1080N.

 

Just saw that 3dmark11 is being retired when w7 loses it's m$ support jan 20, not sure what they are going to replace it with though, they would still need a dx11 compatible test I would imagine. Or I guess they'll give it away free.

 

Fire Strike, Ice Storm, Cloud Gate, and Sky Diver are DX11.

 

Yeah but I assumed the whole 3dmark11 package included those, might not though.

 

3DM11 is DX 11 benchmarks, but replaced by some of the already mentioned benchmarks in the 3DM Suite. See... 3DM Suite contain both DX11 and 12 benchmarks.