Vega FE benchmarked by random dude

[Valentyn]

26 minutes ago, leadeater said:

But these are not launch reviews or reviews at all, user benchmarks are not the same thing and unless you are well, frankly utterly stupid, no one is going to mistake them as anything else. When the actual reviews come out they will get recycled as you say like always but these aren't those.

 

This is the actual issue.

 

PcPer are doing official review live, and the Fe at 1440Mhz is barely faster than a 1050Mhz Fury X so far...not good at all.

https://www.pcper.com/live/

[Flavio hc 16]

from the pcper  live: dirt rally

amd FE around 100 fps

titan xp: 155-160fps...

it's not looking good, sigh

[Taf the Ghost]

36 minutes ago, leadeater said:

But these are not launch reviews or reviews at all, user benchmarks are not the same thing and unless you are well, frankly utterly stupid, no one is going to mistake them as anything else. When the actual reviews come out they will get recycled as you say like always but these aren't those.

 

This is the actual issue.

While I don't think this is a good look at the moment, and do agree that they released the FE because they could & to hit investor targets, we saw the "I give up, AMD!" stuff post Computex with regards to Vega. Some anti-hype has its utility. 

 

Also, almost no one benchmarked the Titan X(p) or even mostly the Titan X as well. No one really talks about either for buying graphics cards.  The FE won't be too important since few are getting review kits.  RX Vega will be the big media push.  (Granted, if SIGGRAPH goes terribly, throw as much shade at the RTG marketing department as possible.)

[MageTank]

22 minutes ago, Valentyn said:

 

PcPer are doing official review live, and the Fe at 1440Mhz is barely faster than a 1050Mhz Fury X so far...not good at all.

https://www.pcper.com/live/

Man, that twitch chat is the most delusional bunch of people that I've ever seen. Some are downright uninformed as to how basic computing even works. 

[Valentyn]

1 minute ago, MageTank said:

Man, that twitch chat is the most delusional bunch of people that I've ever seen. Some are downright uninformed as to how basic computing even works. 

 

That's rather typical, as wccft linked the stream on their site; so I expect their denizens are all in that chat as well.

[MageTank]

1 minute ago, Valentyn said:

 

That's rather typical, as wccft linked the stream on the site; so I expect their denizens are all in that chat as well.

Yeah... I was trying to explain the concept of undervolting to achieve a higher clock speed, and was just cursed to the high heavens by a guy that clearly doesn't know how voltage works. My suggestion was simple: undervolt to reduce thermals (thereby increasing thermal headroom) and see if it boosts higher as a result. However, the poor lads brain only comprehends overclocking as "THROW MOAR VOLTS 4 ALL THE CLOCK SPEEDS!". I even invited him to look at my pascal overclocking on the voltage curve, and how I achieved a 120mhz higher overclock by actually using less voltage, but he was having none of it.

 

I simply wanted my theory to be tested, as I believe these cards are over-volted out of the box (RX480 style) and I wanted to see if someone could prove that. Sadly, far too many rabid fans in that chat to even get a word through. 

[Valentyn]

1 minute ago, MageTank said:

I simply wanted my theory to be tested, as I believe these cards are over-volted out of the box (RX480 style) and I wanted to see if someone could prove that. Sadly, far too many rabid fans in that chat to even get a word through. 

 

Exactly, on all the last major AMD cards undervolting has really helped.
 

Look at the RX 480, and Fury X, undervolt, and get more out of it in the end as well.

AMD has been bad at overvolting reference cards the last 2 years.

[JoeCoke]

Vega might suck, it might not, but jesus fucking christ people, wait for the actual cards to be reviewed by actually capable people who actually know what they're doing. Besides, who knows, the drivers might improve a lot before the real deal comes around. (NOT DEFENDING AYYMD, just urging people to think!)

[Valentyn]

5 minutes ago, 2Buck said:

Vega might suck, it might not, but jesus fucking christ people, wait for the actual cards to be reviewed by actually capable people who actually know what they're doing. Besides, who knows, the drivers might improve a lot before the real deal comes around. (NOT DEFENDING AYYMD, just urging people to think!)

 

PcPer are doing it right now, live.

 

Tomb Raider 75 vs 120 on Titan X

https://www.pcper.com/live/

[leadeater]

43 minutes ago, Valentyn said:

 

PcPer are doing official review live, and the Fe at 1440Mhz is barely faster than a 1050Mhz Fury X so far...not good at all.

https://www.pcper.com/live/

Didn't instill much confidence in their judgement when he stated during his testing of GTA V that while FE is using older drivers it shouldn't matter since the game is old and is optimized well already.... yes that is true but this is a new GPU architecture so old drivers do matter. How much I don't know but that was kind of a dumb statement, heat of the moment I guess, I don't expect that same reasoning to make it through to the written article.

[MoonSpot]

25 minutes ago, MageTank said:

Man, that twitch chat is the most delusional bunch of people that I've ever seen. Some are downright uninformed as to how basic computing even works. 

You should go to their IRC chatroom while watching the stream, don't even bother with twitch and youtube chats....ever.

[Valentyn]

Just now, leadeater said:

Didn't instill much confidence in their judgement when he stated during his testing of GTA V that while FE is using older drivers it shouldn't matter since the game is old and is optimized well already.... yes that is true but this is a new GPU architecture so old drivers do matter. How much I don't know but that was kind of a dumb statement, heat of the moment I guess, I don't expect that same reasoning to make it through to the written article.

 

Yeah, especially after AMD told them this.
"Expect no performance delta between Gaming and Pro mode. Shipping driver is current, but is not comparable to RX Vega driver."
PCPer Ryan quoting AMD

 

They're doing PCIe power draw tests now. Although so far the performance of the Frontier Edition is just bad. Some games and tests it's barely able to beat a Fury X, others it's around GTX 1080.

[MageTank]

2 minutes ago, leadeater said:

Didn't instill much confidence in their judgement when he stated during his testing of GTA V that while FE is using older drivers it shouldn't matter since the game is old and is optimized well already.... yes that is true but this is a new GPU architecture so old drivers do matter. How much I don't know but that was kind of a dumb statement, heat of the moment I guess, I don't expect that same reasoning to make it through to the written article.

Yeah, I just turned the stream off. Going to wait for a written review and a better testing methodology. Watching him do fallout 4 also made me question their methodology as it looked like he was only half-way paying attention. The runs were pretty inconsistent. 

[leadeater]

3 minutes ago, Valentyn said:

They're doing PCIe power draw tests now. Although so far the performance of the Frontier Edition is just bad. Some games and tests it's barely able to beat a Fury X, others it's around GTX 1080.

 

Quote

We'll look at the data after these individuals get burned at the stake

Haha awesome quote.

 

Also damn missed some of it since I closed the stream, they given any power figures yet?

[Valentyn]

3 minutes ago, leadeater said:

 

Haha awesome quote.

 

Also damn missed some of it since I closed the stream, they given any power figured yet?

 

About 130-138 per 8 pin so far

[Marinatall_Ironside]

9 minutes ago, Valentyn said:

 

Yeah, especially after AMD told them this.
"Expect no performance delta between Gaming and Pro mode. Shipping driver is current, but is not comparable to RX Vega driver."
PCPer Ryan quoting AMD

 

They're doing PCIe power draw tests now. Although so far the performance of the Frontier Edition is just bad. Some games and tests it's barely able to beat a Fury X, others it's around GTX 1080.

Yeah power draw is close to the official 300W TDP rating, so AMD was pretty accurate with the figure.

 

They are also recording the power draw and FPS performance. It's also interesting on how they test power draw at the PCIe bus by using a riser card rigged with probes specifically to intercept the 3.3V and 12V lines at the bus and they have probes coupled into the 12V lines on the PCIe power plugs. That's pretty much how you eliminate all other variables down to whatever specific component that's being tested, the right way.

 

I'm not judging performance just yet, as we don't have the complete picture yet. Until the full test suite PC Perspective runs is complete, I'm reserving my judgement.

Edited June 29, 2017 by JurunceNK

[Notional]

Things PCPer has concluded (via own testing AND AMD official statements to them):

 

There is no difference between pro and gaming mode in performance. It is simply settings (and perhaps some support under pro) that is the difference.

The card will indeed run at 1600mhz.

---

 

Now, this is NOT a gaming card. Its focus is still on content creation, and AMD has stated that gamers should wait on the RX version with RX drivers.

Any logical person should understand that this is not a gaming card, when it at 50% higher clock speed with same amount of stream processors, is slower or equal to Fiji. There would be no point to make the Vega architecture if its IPC is much much worse than the previous Fiji chip.

[Valentyn]

Just now, JurunceNK said:

Yeah power draw is close to the official 300W TDP rating, so AMD was pretty accurate with the figure.

 

They are also recording the power draw and FPS performance. It's also interesting on how they test power draw at the PCIe bus by using a riser card rigged with probes specifically to intercept the 3.3V and 12V lines at the bus and they have probes coupled into the 12V lines on the PCIe power plugs. That's pretty much how you eliminate all other variables down to whatever specific component that's being tested, the right way.

 

Yeah, nice to see their method and components.

Also the power seems to jump around a lot in tests, and clock speeds are not adjusted with those jumps and dips.

[cj09beira]

2 minutes ago, JurunceNK said:

Yeah power draw is close to the official 300W TDP rating, so AMD was pretty accurate with the figure.

 

They are also recording the power draw and FPS performance. It's also interesting on how they test power draw at the PCIe bus by using a riser card rigged with probes specifically to intercept the 3.3V and 12V lines at the bus and they have probes coupled into the 12V lines on the PCIe power plugs. That's pretty much how you eliminate all other variables down to whatever specific component that's being tested, the right way.

 

I'm not judging performance just yet, as we don't have the complete picture yet. Until the full test suite PC Perspective runs is complete, I'm reserving my judgement.

a question is how do we tackle the fact that nvidea uses the cpu for scheduling so part of the energy is wasted there 

[Marinatall_Ironside]

Just now, Valentyn said:

 

Yeah, nice to see their method and components.

Also the power seems to jump around a lot in tests, and clock speeds are not adjusted with those jumps and dips.

Yeah it is all over the place. Once they complete their tests they'll analyze and find out what's causing the power consumption figure to not be very consistent by also looking at clock speed changes over time, since that provides a pretty good (but not perfect) explanation of the power changes.

 

They just disabled AMD Chill as of writing this.

[Hunter259]

5 minutes ago, Notional said:

Things PCPer has concluded (via own testing AND AMD official statements to them):

 

There is no difference between pro and gaming mode in performance. It is simply settings (and perhaps some support under pro) that is the difference.

The card will indeed run at 1600mhz.

 

---

 

Now, this is NOT a gaming card. Its focus is still on content creation, and AMD has stated that gamers should wait on the RX version with RX drivers.

Any logical person should understand that this is not a gaming card, when it at 50% higher clock speed with same amount of stream processors, is slower or equal to Fiji. There would be no point to make the Vega architecture if its IPC is much much worse than the previous Fiji chip.

Acting as if AMD has never done that before... *Looks at Bulldozer vs K10*

[Mira Yurizaki]

On 6/28/2017 at 2:20 PM, valdyrgramr said:

Some random guy isn't credible.

A sample size of one doesn't make anything definitive. But it's still a sample.

[Taf the Ghost]

19 minutes ago, MageTank said:

Yeah, I just turned the stream off. Going to wait for a written review and a better testing methodology. Watching him do fallout 4 also made me question their methodology as it looked like he was only half-way paying attention. The runs were pretty inconsistent. 

The lack of using standard deviations in game benchmarking still bugs me... a lot.

[cj09beira]

1 minute ago, Hunter259 said:

Acting as if AMD has never done that before... *Looks at Bulldozer vs K10*

maybe the fact that they changed the cache could have a bad effect on software made for older gcn?

[Mira Yurizaki]

6 minutes ago, cj09beira said:

maybe the fact that they changed the cache could have a bad effect on software made for older gcn?

It's speculated that Vega is using tiled rasterization, similar to what NVIDIA did when they made Maxwell 2.

 

So that shouldn't really have an effect on anything considering Maxwell 2 performed admirably well across the board.