Old Hardware Benchmarks Running UCSF Chimera

There are two types of benchmarks: the volume benchmarks and the molecule benchmarks.

Volume benchmarks

The volume scores (Surface, Mesh, Contour, Solid, Recolor) give the edge size of a cube shaped volume data array such that rendering or contouring or recoloring can be performed 10 times per second. Data of this size can be easily manipulated in Chimera.

Doubling a surface or mesh rendering score means rendering 4 times as many triangles or lines per second. Doubling a contouring, solid rendering or solid recoloring score means processing 8 times as many data points per second.

Molecule benchmarks

The molecule scores (Null, Wire, Stick, BStick, Ribbon, Sphere, Ops) give the frames-per-second for a given molecule and representation. (See the benchmark extension documentation for details.)

Benchmark Extension

The benchmark extension can be started from the Tools/Utilities/Benchmark menu entry. If you send us (chimera-users@cgl.ucsf.edu) your scores and machine information, we will include them in the table below.

Scores for Chimera Versions >= 1892

Some of these scores are incorrect high values. Prior to 1.2509 Chimera version the benchmark code did not draw the models to the screen in in order to avoid syncronization to the display refresh rate (common with old CRT displays). Some drivers were able to take advantage of this and shortcut the full rendering giving incorrect timings.

Graphics	Processor	Clock (MHz)	OS Type	Surface	Mesh	Contour	Solid	Recolor	Notes	Test Date	Chimera Version	Machine	Graphics Memory (Mb)	Main Memory (Mb)	OS Version	Driver Version	Host name
Quadro FX 3500	2 dual-core Opteron 2220			842	846	153	477	153		8/07	2422	HP xw9400	256	8192		100.14.11	From Dave Konerding
				Molecule (Atoms) Null Wire Stick BStick Ribbon Sphere Ops 1F4H (33805) 46 114 9 8 (n/a) 25 21
Quadro FX 1500	Intel Core 2 6600	2400	Linux	839	809	179	435	161		9/07	2422		256	2048	Fedora Core 6		From Gabor Papai
GeForce 8800 GT	2 quad-core Intel Xeon	2800	Mac	766	754	184	406	239		3/08	2500(?)	Mac Pro	512	2048	Mac OS 10.5.2(?)		From Jeff Triffo
				Molecule (Atoms) Null Wire Stick BStick Ribbon Sphere Ops 1F4H (33805) 163 170 5 5 8 7 42
2 NVidia Quadro FX3400s w/SLI	Dual Xeon	3400	Windows	709	709	156	372	138	PCI Express	8/04	1980	3DBOXX 8206	256 each	2048	XP Professional	65.90 prerelease	SIGGRAPH04 booth
Radeon HD 2400 Pro	Dual Intel E2160	1800	Windows	648	554	168	248	115		4/08	2492	Dell Inspiron 530s	256	3072	Windows XP	6.14.0010.6764	From Sacha De Carlo.
				Molecule (Atoms) Null Wire Stick BStick Ribbon Sphere Ops 1F4H (33805) 93 171 8 7 (n/a) 14 30
Quadro FX 5600	2 Quad-core Xeon	2800	Mac	633	627	184	443	239	Out of main memory at solid 444.	3/08	2496	PowerMac	1536	2048	Mac OS 10.5.2		strawson.compbio.ucsf.edu
				Molecule (Atoms) Null Wire Stick BStick Ribbon Sphere Ops 1F4H (33805) 161 168 5 5 (n/a) 7 40
NVidia Quadro FX3400	Dual Xeon	3400	Windows	631	672	152	300	136	PCI Express	8/04	1980	3DBOXX 8208	256	2048	XP Professional	61.76	SIGGRAPH04 booth
Quadro FX 1400	Pentium M	2000	Linux	601	682	156	310	150		10/06	2199	Dell Precision M70 laptop	256	1024	Redhat Enterprise 4 WS	1.0-8774	CGL machine
ATI FireGL X1 256p	Dual Xeon	2400	Windows	592	594	142	298	85	2X multisampling	4/04	1917	Compaq Evo W8000	256	1024	XP Professional	7.98.5	whimsey.cgl.ucsf.edu
ATI FireGL X1 256p	Dual Xeon	2400	Windows	583	585	142	285	86	6X multisampling	4/04	1917	Compaq Evo W8000	256	1024	XP Professional	7.98.5	whimsey.cgl.ucsf.edu
Quadro FX 1400	Pentium M	2000	Windows	577	683	156	256	156	Solid size 300 - "not enough memory".	10/06	2255	Dell Precision M70 laptop	256	1024	Windows XP Pro	84.30	CGL machine
Nvidia Quadro FX 4500	Dual-core Xeon	2660	Mac							5/07	2382	Apple Mac Pro	512	2048	Mac OS 10.4.9		From Matthew Dougherty
				Molecule (Atoms) Null Wire Stick BStick Ribbon Sphere Ops 1F4H (33805) 154 166 8 8 (n/a) 13 38
Radeon X1900	2 Dual-core Xeon	2660	Mac	567	566	192	350+	134	Solid size >350 - "not enough memory".	1/07	2304	Apple Mac Pro	512	2048	Mac OS 10.4.8		From Jeff Speir
				Molecule (Atoms) Null Wire Stick BStick Ribbon Sphere Ops 1F4H (33805) 150 167 6 6 4 10 38
GeForce 7900GT	Intel Core 2 Duo Extreme 6800	2900	Linux	549	472	194	489+	172	Out of memory on solid after size 489 at 12.5 fps.	10/06	2199	Custom built.	256	4192	SUSE 10.1	x86_64 8774	From David Konerding.
Radeon X1600	Intel Core Duo	2000	Mac	547	543	157	256	75		10/06	2255	iMac	256	1024	Mac OS 10.4.8		oscar.cgl.ucsf.edu
Radeon X1600	Intel Core 2 Duo	2330	Mac	541	541	170	272	93		3/08	2496	MacBook Pro	256	2048	Mac OS 10.4.11		feyerabend.cgl.ucsf.edu
				Molecule (Atoms) Null Wire Stick BStick Ribbon Sphere Ops 1F4H (33805) 159 174 4.5 5.7 13 5.2 34
GeForce 7900GT	Intel Core 2 Duo Extreme 6800	2900	Windows	507	489	211	349	172		10/06	2199	Custom built.	256	4192	Win XP 32-bit	91.47_forceware	From David Konerding.
Radeon HD 2600	2 Quad-core Xeon	2800	Mac	468	466	182	325	217		2/08	2470	Apple Mac Pro	256	6144	Mac OS 10.5		From Jeff Triffo
				Molecule (Atoms) Null Wire Stick BStick Ribbon Sphere Ops 1F4H (33805) 152 159 6 5 (n/a) 8 40
3dLabs VP990 Pro	P4	3000	Windows	431	428	156	250+	102	Driver set for max texture performance.	4/04	1917	Dell OptiPlex GX270	512	1024	XP Professional	3.01-0739	Greg's computer
3dLabs VP990 Pro	P4	3000	Windows	427	421	156	127	96	No multisampling.	4/04	1917	Dell OptiPlex GX270	512	1024	XP Professional	3.01-0739	Greg's computer
ATI Mobility Radeon X1300	Intel Core Duo T2300	1700	Windows	394	486	141	256	88	Laptop.	4/06	2199	Dell Inspiron E1505	128	512	Windows XP Media Edition		From Daniil Zaitsev
ATI Mobility Radeon 9700	G4	1670	Mac	379	382	116	256	80	Laptop.	3/05	2065	PowerBook	128	?	10.3.8 ?		Tom Ferrin's computer
3dLabs VP990 Pro	P4	3000	Windows	357	264	156	76	89	With 4x multisampling.	4/04	1917	Dell OptiPlex GX270	512	1024	XP Professional	3.01-0739	Greg's computer
Radeon X1900	2 Dual-core Xeon	2660	Windows	315	315	186	350+	129	Solid size >350 - "not enough memory".	1/07	2304	Apple Mac Pro	512	2048	XP		From Jeff Speir
				Molecule (Atoms) Null Wire Stick BStick Ribbon Sphere Ops 1F4H (33805) 92 169 10 10 25 18 42
Radeon 9800 Pro	Dual G5	2000	Mac	306	306	142	256	94	Chimera compiled on G4.	4/04	1917	Power Mac	128	1536	10.3.3		mach.cgl.ucsf.edu
Quadro4 500 GoGL	P4 Mobile	2200	Linux	296	170	138	217	108	Mesh (10) and solid (11) unexpectedly slow.	4/04	1917	Dell Precision M50 laptop	64	1024	Redhat 8.0	4363	feyerabend.cgl.ucsf.edu
Quadro4 980 XGL	P4	3000	Windows	287	287	158	256	120		4/04	1917	HP Workstation xw4100	128	1024	XP Pro	5664	exhibitus.cgl.ucsf.edu
Quadro NVS 285	Dual P4	3400	Windows	261	116	144	110	95		9/07	2422	Dell Precision WorkStation 380	64	1024	XP Pro	7.1.8.4	Dr. Dave's computer
ATI FireGL 8800	P4	2400	Linux	234	227	142	256	107		10/05	2176	HP XW5500 workstation	128	512	Gentoo GNU/LINUX	fglrx 8.14.30	from Mingfeng Yang
GeForce4 MX	Athlon 2800	2100	Windows	163	153	131	>250	118		3/05	2090	Emachines T2825	64	512	XP Home, SP2	6.6.9.3	TG's computer
Radeon Mobility M7	Pentium M	1400	Linux	147	127	133	128	68		10/05	2176	IBM T41	32 (?)	512	Ubuntu GNU/LINUX 5.10	Xorg-x11 6.8.2	from Mingfeng Yang
Intel 965GM	Intel Core 2 duo	1600	Linux	130	129	158	128	129		8/07	2422	Thinkpad x61t	256	1024	2.6.22	xf86-video-intel 2.1.1	From Damir Perisa
				Molecule (Atoms) Null Wire Stick BStick Ribbon Sphere Ops 1F4H (33805) 97 53 1 1 (n/a) 2 28
Radeon Mobility 7500	P4	1600	Windows	119	122	110	99	63		10/05	2172	Compaq Evo N800V	32	512	XP		from Boaz Shaanan
Intel GMA 950	Intel Core 2 Duo	1830	Mac	112	121	151	211	98	No hardware T&L.	10/06	2255 ?	iMac. Same GPU in Mac Mini.	64	512	10.4.8		from Greg Couch

Scores for Chimera Versions 1711 to 1891

Chimera versions 1892 and later use two light sources instead of one, which reduces the surface and mesh rendering scores.

Graphics	Processor	Clock (MHz)	OS Type	Surface	Mesh	Contour	Solid	Recolor	Notes	Test Date	Chimera Version	Machine	Graphics Memory (Mb)	Main Memory (Mb)	OS Version	Driver Version	Host name
Quadro4 900 XGL	Xeon	2800	Windows XP	384	375	150	256	123		6/03	1741	hp workstation xw6000	128	1024	XP Professional	4.0.7.2	cgl-test9.cgl.ucsf.edu
Radeon 9800 Pro	Dual G5	2000	Mac	381	378	138	256	93	Chimera compiled on G4.	6/03	1747	Power Mac	128	2048	10.2.7		WWDC demo machine
Radeon 9800 Pro	Dual G5	2000	Mac	374	374	142	256	103	Chimera compiled on G4.	11/03	1872	Power Mac	128	1536	10.3		mach.cgl.ucsf.edu
GeForce 4 Ti 4600	Dual G4	1000	Mac	349	347	102	256	67		6/03	1747	Power Mac	64	1280	10.2.6		mach.cgl.ucsf.edu
Quadro4 500 GoGL	P4 Mobile	2200	Linux	310	179	134	196	109	Mesh (10) and solid (11) unexpectedly slow.	6/03	1747	Dell Precision M50 laptop	64	1024	Redhat 8.0	4363	feyerabend.cgl.ucsf.edu

Scores for Chimera Versions < 1711

The scores in the following table are for Chimera versions prior to 1711. Optimizations for surface and mesh rendering (12) were added in Chimera version 1711 and contouring (13) in Chimera 1716 that greatly increased scores.

Graphics	Processor	Clock (MHz)	OS Type	Surface	Mesh	Contour	Solid	Recolor	Notes	Test Date	Chimera Version	Machine	Graphics Memory (Mb)	Main Memory (Mb)	OS Version	Driver Version	Host name
GeForce 4 Ti 4600	Dual G4	1000	Mac	290	290	70	256	67	Using prerelease Mac OS 10.2.5 build 6L14.	3/03	1700	Power Mac	64	1280	10.2.5		mach.cgl.ucsf.edu
Radeon 9000	Dual G4	1000	Mac	249	254	70	228	75	Using prerelease Mac OS 10.2.5 build 6L14.	3/03	1700	Power Mac	64	1280	10.2.5		mach.cgl.ucsf.edu
NVidia Quadro4 900 XGL	Dual Xeon	2400	Windows XP	223	223	104	256	118	Solid of size 256 gives 51.4 frames per second, size 257 gives 0.9 fps. (1)	10/02	1592	Compaq Workstation W8000	128	1024	XP Professional	3.0.8.7	whimsey.cgl.ucsf.edu
GeForce 4 Ti 4400	P4	1800	Linux	212	201	98	256	100	Solid of size 256 gives 20 frames per second, size 257 gives 3 fps. (1) Also see (2), (3).	8/02	1522		128	1024	Debian 3.0	NVIDIA 28.80	tamarin.ucsf.edu
FireGL 2	P4	1400	Windows 2000	209	208	47	61	92	Slow contouring in Chimera 1602 (9).	12/02	1602	Compaq Deskpro	64	768	2000 Professional	5.12.2195.2082	arnauld.cgl.ucsf.edu
FireGL 2	P4	1400	Windows 2000	208	207	76	63	79		7/02	1516	Compaq Deskpro	64	768	2000 Professional	5.12.2195.2082	arnauld.cgl.ucsf.edu
FireGL 2	P4	1400	Linux	194	194	82	62	78		7/02	1516	Compaq Deskpro	64	768	Redhat 7.2	firegl23 1.9.7	tolkien.cgl.ucsf.edu
FireGL 2	P4	2000	Linux	192	194	100	60	92	Scores reported by Dave Konerding.	9/02	1516	White box PC	64	1024	Redhat 7.3	FireGL2 1.9.20
Quadro4 500 GoGL	P4 Mobile	2200	Windows XP	179	179	71	256	115	Crashed on solid and recolor unless run separately. Also see (1), (9).	12/02	1602	Dell Precision M50 laptop	64	1024	XP Professional	6.13.10.2835
Quadro4 500 GoGL	P4 Mobile	2200	Linux	176	176	94	256	106	Solid of size 256 gives 18 frames per second. (1)	12/02	1602	Dell Precision M50 laptop	64	1024	Redhat 8.0	4191
GeForce 2 GTS/Pro	Athlon	1300	Windows XP	165	169	75	256	96	Solid of size 256 gives 18 frames per second. (1) Also see (2).	7/02	1516		64	512	XP	2.8.3.2	chagall.nmr.ucsf.edu
VPro 12	R12000	300	IRIX	157	157	57	146	69		7/02	1516	SGI Octane	128	640	6.5.14f		larson.cgl.ucsf.edu
Quadro2 Go (Dell Mobile)	P3 Mobile	1200	Windows 2000	140	140	81	128	66	Crashes on solid test. (5)	10/02	1587	Dell Precision M40 laptop		512	2000	1.4.2.7
ELSA Gloria II Quadro	Dual P3?	866	Windows NT	140	139	66	57	?	Erratic solid display frame rates. Also see (2).	7/02	1516	Compaq AP550	64	256	NT	4.00.1381.2832	edman.cgl.ucsf.edu
ATI Radeon 9000	Dual G4	1000	Mac	119	119	66	240	75	Using prerelease Mac OS X Aqua Chimera. (8).	12/02	1619	Power Mac	64	1280	10.2.2		mach.cgl.ucsf.edu
ATI Radeon 9000	Dual G4	1000	Mac	118	118	71	239	57	Using XFree86 4.2.99.1 (7).	10/02	1574	Power Mac	64	1280	10.2.1	ATI Radeon 8500 OpenGL Engine	mach.cgl.ucsf.edu
Radeon 9000 Mobility	G4	1000	Mac	98	98	53	227	64	Using Apple's X11 beta. Results from Rey Banatao.	2/03	1602	PowerBook	64	1024	10.2.3
Infinite Reality 2	Dual R12000	400	IRIX	94	87	65	48	68		7/02	1516	SGI Onyx	64	1024	6.5.14f		spinoza.cgl.ucsf.edu
GeForce4 420 Go (14)	G4	867	Mac	78	84	63	143	64	Apple X11 beta 3	7/03	1700	12" PowerBook	32	640	10.2.6		Patsy Babbit's machine
GeForce 2 MX	G4	733	Mac	70	76	60	48	40	Using XFree86 4.2.99.1 (7). Display 1600x1024.	10/02	1574	Power Mac	32	768	10.2.1		essex.cgl.ucsf.edu
Cobalt	Dual P2	600	Windows NT	70	70	57	85	48		7/02	1516	SGI 320		256	NT	5.2.3.000, 5.2.3	ryle.cgl.ucsf.edu
PowerStorm 350	Dual Alpha EV6.7	667	Tru64	66	73	73	50	71	Solid rendering shows white cube when using glFog(). Also see (2).	7/02	1516	Compaq DS-20	64	2096	5.1		retief.cgl.ucsf.edu
ATI Radeon 7500 Mobility	G4	800	Mac	63	63	65	128	30	Using Xtools from Tenon (6). Results from Mike Lowenstein.	8/19	1538	PowerBook	32	1024	10.2
Voodoo 5 5500	P3	800	Linux	54	44	67	106	54	Solid rendering shows texture junk on small 137 x 138 x 15 data when rotated	7/02	1516		64	~380	Debian 3.0		drill.ucsf.edu
Mesa 3.4.2 / Radeon 9000	Dual G4	1000	Mac	45	45	70	1	2	Using XDarwin 4.2.0.1. No hardware acceleration (4).	9/02	1574	Power Mac	64	1280	10.2.1		mach.cgl.ucsf.edu
Solid Impact	R10000	195	IRIX	43	41	49	2	14		7/02	1516	SGI Octane		256	6.5.15m		monet.nmr.ucsf.edu
CRM	R5000	300	IRIX	32	34	45	28	45		7/02	1516	SGI O2		128	6.5.6m		utopia3.ucsf.edu
Mesa 3.4.2 / Matrox G400	P3	800	Linux	30	35	69	2	15	No hardware acceleration (4).	7/02	1516			~380	Debian 3.0		langur.ucsf.edu
4D51T	Alpha	180	Tru64	19	18	29	45	35	Solid and recolor benchmarks sometimes cause system hang.	7/02	1516	Compaq AlphaStation 200 4/233		128	5.1		feyerabend.cgl.ucsf.edu
GR2	R5000	180	IRIX	13	14	37	1	9		7/02	1516	SGI INDY		64	6.5.15m		lautrec.nmr.ucsf.edu

Notes

(1) Textures must have power of 2 sizes. Going from size 256 to 257 makes the 2 dimensional textures jump from size 256 by 256 to 512 by 512. The bigger textures may not fit in texture memory reducing performance significantly. A better implementation could render a data set with edge size 257 using several smaller blocks to avoid this big jump in performance.

(2) These systems have a bug in the OpenGL implementation where glPixelMap() does not work when loading a texture. A fix to work around the problem was put into Chimera 1522 but the speed of changing thresholds in colormap solid rendering is compromised. The colormap solid rendering is not tested by the benchmarks but is the default solid rendering mode in the Chimera volume viewer extension. No data is shown in colormap solid mode on these (pre 1522) systems, but an outline box can be seen and rotated normally. The colormap mode can be turned off in the volume viewer options panel. The direct RGBA texture filling mode is then used which is slower when changing thresholds.

(3) Turning on the smooth mesh lines option makes the mesh lines black. This is probably a graphics driver bug.

(4) Mesa is a software implementation of OpenGL. It does not use hardware acceleration provided by the graphics card. The scores for systems using Mesa are probably determined by computer CPU speed rather than graphics card performance.

(5) Crashed on solid test after calculating frame rate for several data sizes. Crash may occur only for volume size greater than 128. Test solid model displays correctly.

(6) Xtools X server from Tenon was reported to behave badly running Chimera. Menus sometimes appear behind windows. Volume viewer dialog window sometimes resizes to zero size and must be resized by hand. Many OpenGL error messages written to shell by Chimera.

(7) Compiled XFree86 4.2.99.1 from Oct 21, 2002 source code. This version of XFree86 has hardware OpenGL acceleration for Mac OS X. X server crashed when started with OroborOSX 0.8b2 built for XFree86 4.2.0.1 when I moved the Chimera window. Worked when X server started with startx command.

(8) Prerelease Aqua Chimera causes Mac OS 10.2.2 system crash when displaying side view window and showing 1gcn when switching to sphere representation. This crash also occurs on the X windows version of Chimera for the Mac. It is probably a Radeon 9000 graphics driver bug. This does not effect benchmarks.

(9) Chimera 1602 for Windows appears to have had the contouring code compiled without optimization making it 5 times slower to calculate isosurfaces than version 1516. This effects the speed of the volume viewer extension.

(10) Mesh rendering did not improve on Quadro4 500 GoGL when limiting display list size. Factor of 3 speed ups were observed on other Quadro4 and GeForce4 cards. The NVidia driver apparently behaves differently for this card.

(11) Solid rendering with the NVidia 4363 driver appears to be slower than with the version 4191 driver on Quadro4 500 GoGL card.

(12) Surface and mesh rendering is about a factor of 3 faster on GeForce4 cards with early 2003 drivers (4191 linux) if display lists are limited to less than about 64K triangles or lines. I couldn't find anything on the web describing this. I changed the Chimera isosurface and mesh rendering to use multiple display lists with limited size.

(13) The contouring benchmark was bottle-necked by slow STL::map<> behaviour. This was especially severe on Windows. The map was used to avoid making multiple copies of triangle vertices. A more efficient array based method increased the calculation speed by a factor of 3 for a 100^3 grid on most platforms, and a factor of 10 on Windows. Larger grids benefit less from this optimization.

(14) Apple web pages say the 12" PowerBook has GeForce4 420 Go graphics. The Apple system profiler program identifies the graphics as GeForce4 MX.

(15) N/A ribbon scores The ribbon score on the molecule benchmark was not valid for Chimera versions 1.2382 through 1.2501 because a bug caused the ribbon not to be displayed during the timing. Only atoms as dots were shown.