Poster | Thread |
megol
|  |
Re: News about Vampire and Apollo Posted on 3-Jan-2017 20:59:41
| | [ #961 ] |
|
|
 |
Regular Member  |
Joined: 17-Mar-2008 Posts: 355
From: Unknown | | |
|
| @Samurai_Crow
Yes but the thing with Altivec was that it was designed to be generally useful (within limitations) and it took Intel ages to make something comparable (IIRC until SSE3). More registers are useful but it doesn't help when the instruction set chosen is obviously inferior. With that said for basic SIMD support a 3 operand MMX version isn't too bad, I just hope there are spaces in the encoding for improving the SIMD support in later versions. Judging by a post by Gunnar (that was quickly removed) describing one (potential?) encoding there should be...
|
|
Status: Offline |
|
|
Overflow
 |  |
Re: News about Vampire and Apollo Posted on 4-Jan-2017 2:41:55
| | [ #962 ] |
|
|
 |
Super Member  |
Joined: 12-Jun-2012 Posts: 1628
From: Norway | | |
|
| A more general comment/copy regarding the improvements in the Vampire core below: Another comment was the IDE speed for the upcoming V1200 IDE being quoted to be 12mb/sec, which is about 4 times the original speed.
The comment about ADoom not being optimized for the Vampire, yet, is a bit meh. If any ID games should be optimized it should be Quake 1 or Quake 3 Arena since they are actually competative games that are still played today. It would be the perfect games to port for a lively gaming scene on the Vampire. I realise Quake 3 Arena would be a massive undertaking tho. Quake 1 is already running, altho slow (until FPU is incorporated).
Quote:
Gunnar von Boehn (Apollo Team Member) Posts 2288 03 Jan 2017 19:52
Quote:
Jan Vonka wrote:
ADoomWOS "timedemo demo1": CyberstormPPC 604e/333 + 060/50 320x200: 77,03 fps 800x600: 16,79 fps Sonnet Crescendo 7200 G3/400 (512k L2) 320x200: 110,63 fps 800x600: 25,35 fps |
Can anybody benchmark Vampires 600 and 500+ right now? Latest Vampire2 Core reaches at DOOM in 800x600 = 19.8 FPS. So its somewhat faster than CyberstormPPC with PPC 604e@333MHz. This is the normal 68k DOOM executable, its not tuned for AMMX or for APOLLO (yet).
With video playback the Vampire is even better. RIVA video player which has some AMMX tunings, the VAMPIRE2 is more than twice as fast in video playback than the CyberstormPPC 604e@333MHz.
Both the Sonnet and Vampire play our Testvideo equally well. And the Sonnet PCI with PPC-G3@400Mhz scores today a little bit faster than the Vampire in Doom.
Of course the tuning of the APOLLO core has just started. And after GOLD1 comes GOLD2 and the there will come a GOLD3 release.
And is likely that the Vampire will beat the Sonnet PCI also in DOOM in the future. |
Last edited by Overflow on 04-Jan-2017 at 02:42 AM.
|
|
Status: Offline |
|
|
OneTimer1
|  |
Re: News about Vampire and Apollo Posted on 4-Jan-2017 17:28:24
| | [ #963 ] |
|
|
 |
Super Member  |
Joined: 3-Aug-2015 Posts: 1148
From: Germany | | |
|
| @Thread
When the Intel came up with MMX it was recommended using some kind of code asking first if this extension is available or not, so you could use the same code on CPUs with or without MMX.
Maybe Intel wasn't clear if they would keep it, the original MMX features became obsolete fast, because they where limited to integer processing. Today most GPUs could do more complex operations faster than MMX could, I don't even know if today’s 64Bit CPUs support this poorly used command set.
https://en.wikipedia.org/wiki/MMX_(instruction_set) Last edited by OneTimer1 on 04-Jan-2017 at 05:30 PM.
|
|
Status: Offline |
|
|
pavlor
|  |
Re: News about Vampire and Apollo Posted on 4-Jan-2017 18:55:54
| | [ #964 ] |
|
|
 |
Elite Member  |
Joined: 10-Jul-2005 Posts: 9660
From: Unknown | | |
|
| @Overflow
Doom is good application benchmark for Vampire: shows integer and 2D performance. Vampire can outshine even 603e/604e + PCI GFX there (thanks to really fast memory for both CPU and GFX). Quake 3 is out of question - requires 3D acceleration. |
|
Status: Offline |
|
|
Seiya
|  |
Re: News about Vampire and Apollo Posted on 4-Jan-2017 21:58:11
| | [ #965 ] |
|
|
 |
Super Member  |
Joined: 19-Aug-2006 Posts: 1479
From: Italia | | |
|
| @pavlor
so, at this point it needs some doom benchmarks with standard processor (not overclocked)
Last edited by Seiya on 04-Jan-2017 at 09:58 PM.
_________________
|
|
Status: Offline |
|
|
pavlor
|  |
Re: News about Vampire and Apollo Posted on 5-Jan-2017 16:35:17
| | [ #966 ] |
|
|
 |
Elite Member  |
Joined: 10-Jul-2005 Posts: 9660
From: Unknown | | |
|
| @Seiya
I have ADoom WOS results for 320x200 (classic demo3 and shareware WAD):
Amiga 3000, Cyberstorm PPC604 (68060 60 MHz) 604e 233 MHz, VooDoo3/2000, 144? MB RAM Fast: 56.5 FPS Amiga 1200T, BlizzardPPC (68060 50 MHz) 603e 150 MHz, Voodoo 3, 256 MB RAM, AmigaOS 3.9 BB2: 29.6
Now ADoom (1.3 I think; 68k): Amiga 3000, Cyberstorm PPC604 (604e 233 MHz) 68060 60 MHz, VooDoo3/2000, 144? MB RAM Fast: 36.6 FPS Amiga 4000D, Cyberstorm MKIII 060 @ 50MHZ, 128 MB Fast RAM, Cybervision 64/3D: 24.4 FPS
As you see 68k ADoom seems to be more optimized than WOS version (or there are other bottlenecks). |
|
Status: Offline |
|
|
Seiya
|  |
Re: News about Vampire and Apollo Posted on 6-Jan-2017 0:27:50
| | [ #967 ] |
|
|
 |
Super Member  |
Joined: 19-Aug-2006 Posts: 1479
From: Italia | | |
|
| Quote:
pavlor wrote:
I have ADoom WOS results for 320x200 (classic demo3 and shareware WAD):
Amiga 3000, Cyberstorm PPC604 (68060 60 MHz) 604e 233 MHz, VooDoo3/2000, 144? MB RAM Fast: 56.5 FPS
Now ADoom (1.3 I think; 68k): Amiga 3000, Cyberstorm PPC604 (604e 233 MHz) 68060 60 MHz, VooDoo3/2000, 144? MB RAM Fast: 36.6 FPS
|
604e: 56 FPS WOS 36 FPS 68k
wos version is much faster than 68k version
and the ohter case is maybe 603 very worst for slow memory access.
Last edited by Seiya on 06-Jan-2017 at 12:31 AM. Last edited by Seiya on 06-Jan-2017 at 12:29 AM.
_________________
|
|
Status: Offline |
|
|
pavlor
|  |
Re: News about Vampire and Apollo Posted on 6-Jan-2017 15:31:28
| | [ #968 ] |
|
|
 |
Elite Member  |
Joined: 10-Jul-2005 Posts: 9660
From: Unknown | | |
|
| @Seiya
Quote:
wos version is much faster than 68k version |
FPS/MHz:
604e/ADoomWOS - 0.24 68060/ADoom 68k - 0.61 |
|
Status: Offline |
|
|
gregthecanuck
|  |
Re: News about Vampire and Apollo Posted on 6-Jan-2017 17:55:51
| | [ #969 ] |
|
|
 |
Cult Member  |
Joined: 30-Dec-2003 Posts: 846
From: Vancouver, Canada | | |
|
| @pavlor
That FPS/MHz ratio is highlighting the difference in performance/clock of 68K vs. PPC. PPC has simpler instructions (hence RISC) but can be clocked higher.
With the "68080" Apollo core this is becoming even more apparent. Clock for clock the 68K architecture is far more performant than PPC. Last edited by gregthecanuck on 06-Jan-2017 at 05:56 PM.
|
|
Status: Offline |
|
|
WolfpackN64
|  |
Re: News about Vampire and Apollo Posted on 6-Jan-2017 18:06:21
| | [ #970 ] |
|
|
 |
Regular Member  |
Joined: 8-Oct-2016 Posts: 300
From: Unknown | | |
|
| @gregthecanuck
Depends, the core the Apollo core is compared against is quite old. CISC chips are for complex instructions indeed faster then RISC chips, but RISC chips will be faster when performing simple, shorter instructions that the architecture was build for.
The 68080 is one hell of a chipdesign for such a small team, but I wouldn't say the 68k architecture is surperior to PPC, especially modern POWER chips would hammer the 68080 on a perf/clock basis.
I think in many cases, it's not really a good comparison to make. How much software is there on OS3 that even uses the PPC accelerators? |
|
Status: Offline |
|
|
pavlor
|  |
Re: News about Vampire and Apollo Posted on 6-Jan-2017 18:15:02
| | [ #971 ] |
|
|
 |
Elite Member  |
Joined: 10-Jul-2005 Posts: 9660
From: Unknown | | |
|
| @gregthecanuck
Quote:
That FPS/MHz ratio is highlighting the difference in performance/clock of 68K vs. PPC. PPC has simpler instructions (hence RISC) but can be clocked higher. |
In CPU intensive tasks 604e shows better performance/MHz than 68060. Difference in ADoom performance/MHz means simply more optimized 68k port, I think. |
|
Status: Offline |
|
|
gregthecanuck
|  |
Re: News about Vampire and Apollo Posted on 7-Jan-2017 9:26:55
| | [ #972 ] |
|
|
 |
Cult Member  |
Joined: 30-Dec-2003 Posts: 846
From: Vancouver, Canada | | |
|
| @WolfpackN64
Sortbench seems to be used as a benchmark recently.
Here is a result from a more modern PPC chip from http://www.apollo-core.com/sortbench/results/result_ppc_7447.txt
CPU PPC 7447 Clock 1.42 GHz L2 512 KB System ibook
SORTBENCH 1.1 (Gunnar von Boehn) ------------------------------------------------------------- 1 K Element : 1750.24 MB/sec 2 K Element : 1784.18 MB/sec 4 K Element : 1784.84 MB/sec 8 K Element : 1779.10 MB/sec 16 K Element : 1782.16 MB/sec 32 K Element : 1765.12 MB/sec
That works out to just over 1.2 MB/sec / MHz on 32K element test.
Here are some recent POWER (Power 7) results from http://www.apollo-core.com/sortbench/results/result_ppc_power7.txt:
./sort.aix ------------------------------------------------------------- SORTBENCH 1.1 (Gunnar von Boehn) Its a CPU benchmark that stresses CPU, DCache and branch prediction. ------------------------------------------------------------- 1 K Element : 4418.00 MB/sec 2 K Element : 4353.62 MB/sec 4 K Element : 4429.73 MB/sec 8 K Element : 4464.52 MB/sec 16 K Element : 4278.30 MB/sec 32 K Element : 4223.28 MB/sec Processor Implementation Mode: POWER 7 Processor Version: PV_7_Compat Number Of Processors: 9 Processor Clock Speed: 3864 MHz
That works out to 1.09 MB/sec / MHz. Slightly less performant than 7447 on this benchmark. But made up for by many cores, higher clock and threading.
Recent benchmark on Sonnet PPC G3/500 from http://www.apollo-core.com/knowledge.php?b=2¬e=3369&x=0&z=EHTMx6 ... The WarpOS optimized SortBench gives the following corrected numbers. 1k = 634.82 MB/sec 2k = 635.04 MB/sec 4k = 635.16 MB/sec 8k = 635.18 MB/sec 16k = 538.31 MB/sec 32k = 515.31 MB/sec
That works out to 1.03MB/sec / MHz on 32K test.
Recent results from November against GOLD2 Apollo core (currently WIP but due "soon") from http://www.apollo-core.com/knowledge.php?b=2¬e=2923&x=2 ... 1k = 341.79 MB/sec 2k = 344.56 MB/sec 4k = 345.69 MB/sec 8k = 346.17 MB/sec 16k = 313.34 MB/sec 32k = 292.73 MB/sec
The clock speed on this core is somewhere around 100MHz. Getting roughly 3MB/sec / MHz. *Significantly* faster than any PPC/Power core shown above.
For grins here are results from an Intel Xeon E5 class http://www.apollo-core.com/sortbench/results/result_x86_xeon.txt:
Intel(R) Xeon(R) CPU E5-2680 0 Clockrate 2.70GHz cache size : 20480 KB bogomips : 5399.21
SORTBENCH 1.0 (Gunnar von Boehn) ------------------------------------------------------------- 1 K Element : 7659.10 MB/sec 2 K Element : 7750.36 MB/sec 4 K Element : 7756.81 MB/sec 8 K Element : 7813.89 MB/sec 16 K Element : 7612.14 MB/sec 32 K Element : 7568.92 MB/sec
Roughly 2.8 MB/sec / MHz on 32K test.
In summary, what this shows is that the enhancements being integrated into the Apollo core are resulting in a very efficient 680x0 design. Also keep in mind the core has no L2 cache and is running on a low-end FPGA. The V1200 is rumoured to be designed for a newer FPGA (Cyclone V vs. III) with more/faster memory.
|
|
Status: Offline |
|
|
WolfpackN64
|  |
Re: News about Vampire and Apollo Posted on 7-Jan-2017 10:42:18
| | [ #973 ] |
|
|
 |
Regular Member  |
Joined: 8-Oct-2016 Posts: 300
From: Unknown | | |
|
| @gregthecanuck
The benchmark looks to be pretty accurate for Amiga use. But I have my doubts when a Core Duo outperforms a POWER7 chip.
In threaded workloads, the POWER chip should outperform the Core by a significant margin (but on the other hand, not a single Amiga platform to date really makes use of multiple cores and threads). |
|
Status: Offline |
|
|
gregthecanuck
|  |
Re: News about Vampire and Apollo Posted on 7-Jan-2017 10:46:57
| | [ #974 ] |
|
|
 |
Cult Member  |
Joined: 30-Dec-2003 Posts: 846
From: Vancouver, Canada | | |
|
| @WolfpackN64
The benchmarks are testing single-core/thread performance from what I understand. There is no internal multi-threading. Last edited by gregthecanuck on 07-Jan-2017 at 10:48 AM.
|
|
Status: Offline |
|
|
WolfpackN64
|  |
Re: News about Vampire and Apollo Posted on 7-Jan-2017 11:00:37
| | [ #975 ] |
|
|
 |
Regular Member  |
Joined: 8-Oct-2016 Posts: 300
From: Unknown | | |
|
| @gregthecanuck
Then the tests seem to be quite reasonable. The Apollo's perm/watt is quite an achievement for such a small team. They certainly put Intel's chip design team to shame. |
|
Status: Offline |
|
|
gregthecanuck
|  |
Re: News about Vampire and Apollo Posted on 7-Jan-2017 11:14:40
| | [ #976 ] |
|
|
 |
Cult Member  |
Joined: 30-Dec-2003 Posts: 846
From: Vancouver, Canada | | |
|
| |
Status: Offline |
|
|
WolfpackN64
|  |
Re: News about Vampire and Apollo Posted on 7-Jan-2017 11:21:34
| | [ #977 ] |
|
|
 |
Regular Member  |
Joined: 8-Oct-2016 Posts: 300
From: Unknown | | |
|
| @gregthecanuck
I hope they get to an ASIC design one day. I can see a lot of embedded applications for something like this (next to the classic systems of course). If only something similar existed for embedded PPC :) |
|
Status: Offline |
|
|
Overflow
 |  |
Re: News about Vampire and Apollo Posted on 7-Jan-2017 13:52:28
| | [ #978 ] |
|
|
 |
Super Member  |
Joined: 12-Jun-2012 Posts: 1628
From: Norway | | |
|
| Gunnar von Boehn (Apollo Team Member) Posts 2298 07 Jan 2017 10:48
Quote:
Ian Parsons wrote:
Is that confirmation of a USB controller on the Vampire 1200? If so it sounds like the board will make (almost) everybody very happy. The V1200 has 2 expansion ports. These ports are much faster than old Clockport. |
There will be USB, Network, and Wifi expansion for them. We are right now testing / developing drivers for those expansions. |
|
Status: Offline |
|
|
pavlor
|  |
Re: News about Vampire and Apollo Posted on 7-Jan-2017 14:46:53
| | [ #979 ] |
|
|
 |
Elite Member  |
Joined: 10-Jul-2005 Posts: 9660
From: Unknown | | |
|
| @gregthecanuck
Sorry, but this sort of benchmarks is meaningless for us. Real world application benchmarks would be far more interesting.
Quote:
The V1200 is rumoured to be designed for a newer FPGA (Cyclone V vs. III) with more/faster memory. |
I hope it will not rise price of the board. |
|
Status: Offline |
|
|
gregthecanuck
|  |
Re: News about Vampire and Apollo Posted on 7-Jan-2017 21:03:15
| | [ #980 ] |
|
|
 |
Cult Member  |
Joined: 30-Dec-2003 Posts: 846
From: Vancouver, Canada | | |
|
| @WolfpackN64
Quote:
I hope they get to an ASIC design one day. I can see a lot of embedded applications for something like this (next to the classic systems of course). |
ASIC comes up all the time. Everyone wants more speed. :)
However the core still has a long way to go before ASIC is considered. In the meantime even on a low-end FPGA this is a very cool project.
|
|
Status: Offline |
|
|